ros_filter.cpp

Go to the documentation of this file.

/*
 * Copyright (c) 2014, 2015, 2016, Charles River Analytics, Inc.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 * 1. Redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above
 * copyright notice, this list of conditions and the following
 * disclaimer in the documentation and/or other materials provided
 * with the distribution.
 * 3. Neither the name of the copyright holder nor the names of its
 * contributors may be used to endorse or promote products derived
 * from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

#include "robot_localization/ros_filter.h"
#include "robot_localization/filter_utilities.h"
#include "robot_localization/ekf.h"
#include "robot_localization/ukf.h"

#include <tf2_geometry_msgs/tf2_geometry_msgs.h>

#include <algorithm>
#include <map>
#include <string>
#include <utility>
#include <vector>
#include <limits>

namespace RobotLocalization
{
  template<typename T>
  RosFilter<T>::RosFilter(std::vector<double> args) :
      staticDiagErrorLevel_(diagnostic_msgs::DiagnosticStatus::OK),
      tfListener_(tfBuffer_),
      dynamicDiagErrorLevel_(diagnostic_msgs::DiagnosticStatus::OK),
      filter_(args),
      frequency_(30.0),
      historyLength_(0),
      lastSetPoseTime_(0),
      latestControl_(),
      latestControlTime_(0),
      tfTimeout_(ros::Duration(0)),
      nhLocal_("~"),
      predictToCurrentTime_(false),
      printDiagnostics_(true),
      gravitationalAcc_(9.80665),
      publishTransform_(true),
      publishAcceleration_(false),
      twoDMode_(false),
      useControl_(false),
      smoothLaggedData_(false),
      disabledAtStartup_(false),
      enabled_(false),
      toggledOn_(true)
  {
    stateVariableNames_.push_back("X");
    stateVariableNames_.push_back("Y");
    stateVariableNames_.push_back("Z");
    stateVariableNames_.push_back("ROLL");
    stateVariableNames_.push_back("PITCH");
    stateVariableNames_.push_back("YAW");
    stateVariableNames_.push_back("X_VELOCITY");
    stateVariableNames_.push_back("Y_VELOCITY");
    stateVariableNames_.push_back("Z_VELOCITY");
    stateVariableNames_.push_back("ROLL_VELOCITY");
    stateVariableNames_.push_back("PITCH_VELOCITY");
    stateVariableNames_.push_back("YAW_VELOCITY");
    stateVariableNames_.push_back("X_ACCELERATION");
    stateVariableNames_.push_back("Y_ACCELERATION");
    stateVariableNames_.push_back("Z_ACCELERATION");

    diagnosticUpdater_.setHardwareID("none");
  }

  template<typename T>
  RosFilter<T>::~RosFilter()
  {
    topicSubs_.clear();
  }

  template<typename T>
  void RosFilter<T>::reset()
  {
    // Get rid of any initial poses (pretend we've never had a measurement)
    initialMeasurements_.clear();
    previousMeasurements_.clear();
    previousMeasurementCovariances_.clear();

    clearMeasurementQueue();

    filterStateHistory_.clear();
    measurementHistory_.clear();

    // Also set the last set pose time, so we ignore all messages
    // that occur before it
    lastSetPoseTime_ = ros::Time(0);

    // clear tf buffer to avoid TF_OLD_DATA errors
    tfBuffer_.clear();

    // clear last message timestamp, so older messages will be accepted
    lastMessageTimes_.clear();

    // reset filter to uninitialized state
    filter_.reset();

    // clear all waiting callbacks
    ros::getGlobalCallbackQueue()->clear();
  }

  template<typename T>
  bool RosFilter<T>::toggleFilterProcessingCallback(robot_localization::ToggleFilterProcessing::Request& req,
                                                    robot_localization::ToggleFilterProcessing::Response& resp)
  {
    if (req.on == toggledOn_)
    {
      ROS_WARN_STREAM("Service was called to toggle filter processing but state was already as requested.");
      resp.status = false;
    }
    else
    {
      ROS_INFO("Toggling filter measurement filtering to %s.", req.on ? "On" : "Off");
      toggledOn_ = req.on;
      resp.status = true;
    }
    return true;
  }

  // @todo: Replace with AccelWithCovarianceStamped
  template<typename T>
  void RosFilter<T>::accelerationCallback(const sensor_msgs::Imu::ConstPtr &msg, const CallbackData &callbackData,
    const std::string &targetFrame)
  {
    // If we've just reset the filter, then we want to ignore any messages
    // that arrive with an older timestamp
    if (msg->header.stamp <= lastSetPoseTime_)
    {
      return;
    }

    const std::string &topicName = callbackData.topicName_;

    RF_DEBUG("------ RosFilter::accelerationCallback (" << topicName << ") ------\n"
             "Twist message:\n" << *msg);

    if (lastMessageTimes_.count(topicName) == 0)
    {
      lastMessageTimes_.insert(std::pair<std::string, ros::Time>(topicName, msg->header.stamp));
    }

    // Make sure this message is newer than the last one
    if (msg->header.stamp >= lastMessageTimes_[topicName])
    {
      RF_DEBUG("Update vector for " << topicName << " is:\n" << topicName);

      Eigen::VectorXd measurement(STATE_SIZE);
      Eigen::MatrixXd measurementCovariance(STATE_SIZE, STATE_SIZE);

      measurement.setZero();
      measurementCovariance.setZero();

      // Make sure we're actually updating at least one of these variables
      std::vector<int> updateVectorCorrected = callbackData.updateVector_;

      // Prepare the twist data for inclusion in the filter
      if (prepareAcceleration(msg, topicName, targetFrame, updateVectorCorrected, measurement,
            measurementCovariance))
      {
        // Store the measurement. Add an "acceleration" suffix so we know what kind of measurement
        // we're dealing with when we debug the core filter logic.
        enqueueMeasurement(topicName,
                           measurement,
                           measurementCovariance,
                           updateVectorCorrected,
                           callbackData.rejectionThreshold_,
                           msg->header.stamp);

        RF_DEBUG("Enqueued new measurement for " << topicName << "_acceleration\n");
      }
      else
      {
        RF_DEBUG("Did *not* enqueue measurement for " << topicName << "_acceleration\n");
      }

      lastMessageTimes_[topicName] = msg->header.stamp;

      RF_DEBUG("Last message time for " << topicName << " is now " <<
        lastMessageTimes_[topicName] << "\n");
    }
    else if (resetOnTimeJump_ && ros::Time::isSimTime())
    {
      reset();
    }
    else
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp before that of the previous message received," <<
                " this message will be ignored. This may indicate a bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);

      RF_DEBUG("Message is too old. Last message time for " << topicName <<
               " is " << lastMessageTimes_[topicName] << ", current message time is " <<
               msg->header.stamp << ".\n");
    }

    RF_DEBUG("\n----- /RosFilter::accelerationCallback (" << topicName << ") ------\n");
  }

  template<typename T>
  void RosFilter<T>::controlCallback(const geometry_msgs::Twist::ConstPtr &msg)
  {
    geometry_msgs::TwistStampedPtr twistStampedPtr = geometry_msgs::TwistStampedPtr(new geometry_msgs::TwistStamped());
    twistStampedPtr->twist = *msg;
    twistStampedPtr->header.frame_id = baseLinkFrameId_;
    twistStampedPtr->header.stamp = ros::Time::now();
    controlCallback(twistStampedPtr);
  }

  template<typename T>
  void RosFilter<T>::controlCallback(const geometry_msgs::TwistStamped::ConstPtr &msg)
  {
    if (msg->header.frame_id == baseLinkFrameId_ || msg->header.frame_id == "")
    {
      latestControl_(ControlMemberVx) = msg->twist.linear.x;
      latestControl_(ControlMemberVy) = msg->twist.linear.y;
      latestControl_(ControlMemberVz) = msg->twist.linear.z;
      latestControl_(ControlMemberVroll) = msg->twist.angular.x;
      latestControl_(ControlMemberVpitch) = msg->twist.angular.y;
      latestControl_(ControlMemberVyaw) = msg->twist.angular.z;
      latestControlTime_ = msg->header.stamp;

      // Update the filter with this control term
      filter_.setControl(latestControl_, msg->header.stamp.toSec());
    }
    else
    {
      ROS_WARN_STREAM_THROTTLE(5.0, "Commanded velocities must be given in the robot's body frame (" <<
        baseLinkFrameId_ << "). Message frame was " << msg->header.frame_id);
    }
  }

  template<typename T>
  void RosFilter<T>::enqueueMeasurement(const std::string &topicName,
                                        const Eigen::VectorXd &measurement,
                                        const Eigen::MatrixXd &measurementCovariance,
                                        const std::vector<int> &updateVector,
                                        const double mahalanobisThresh,
                                        const ros::Time &time)
  {
    MeasurementPtr meas = MeasurementPtr(new Measurement());

    meas->topicName_ = topicName;
    meas->measurement_ = measurement;
    meas->covariance_ = measurementCovariance;
    meas->updateVector_ = updateVector;
    meas->time_ = time.toSec();
    meas->mahalanobisThresh_ = mahalanobisThresh;
    meas->latestControl_ = latestControl_;
    meas->latestControlTime_ = latestControlTime_.toSec();
    measurementQueue_.push(meas);
  }

  template<typename T>
  void RosFilter<T>::forceTwoD(Eigen::VectorXd &measurement,
                               Eigen::MatrixXd &measurementCovariance,
                               std::vector<int> &updateVector)
  {
    measurement(StateMemberZ) = 0.0;
    measurement(StateMemberRoll) = 0.0;
    measurement(StateMemberPitch) = 0.0;
    measurement(StateMemberVz) = 0.0;
    measurement(StateMemberVroll) = 0.0;
    measurement(StateMemberVpitch) = 0.0;
    measurement(StateMemberAz) = 0.0;

    measurementCovariance(StateMemberZ, StateMemberZ) = 1e-6;
    measurementCovariance(StateMemberRoll, StateMemberRoll) = 1e-6;
    measurementCovariance(StateMemberPitch, StateMemberPitch) = 1e-6;
    measurementCovariance(StateMemberVz, StateMemberVz) = 1e-6;
    measurementCovariance(StateMemberVroll, StateMemberVroll) = 1e-6;
    measurementCovariance(StateMemberVpitch, StateMemberVpitch) = 1e-6;
    measurementCovariance(StateMemberAz, StateMemberAz) = 1e-6;

    updateVector[StateMemberZ] = 1;
    updateVector[StateMemberRoll] = 1;
    updateVector[StateMemberPitch] = 1;
    updateVector[StateMemberVz] = 1;
    updateVector[StateMemberVroll] = 1;
    updateVector[StateMemberVpitch] = 1;
    updateVector[StateMemberAz] = 1;
  }

  template<typename T>
  bool RosFilter<T>::getFilteredOdometryMessage(nav_msgs::Odometry &message)
  {
    // If the filter has received a measurement at some point...
    if (filter_.getInitializedStatus())
    {
      // Grab our current state and covariance estimates
      const Eigen::VectorXd &state = filter_.getState();
      const Eigen::MatrixXd &estimateErrorCovariance = filter_.getEstimateErrorCovariance();

      // Convert from roll, pitch, and yaw back to quaternion for
      // orientation values
      tf2::Quaternion quat;
      quat.setRPY(state(StateMemberRoll), state(StateMemberPitch), state(StateMemberYaw));

      // Fill out the message
      message.pose.pose.position.x = state(StateMemberX);
      message.pose.pose.position.y = state(StateMemberY);
      message.pose.pose.position.z = state(StateMemberZ);
      message.pose.pose.orientation.x = quat.x();
      message.pose.pose.orientation.y = quat.y();
      message.pose.pose.orientation.z = quat.z();
      message.pose.pose.orientation.w = quat.w();
      message.twist.twist.linear.x = state(StateMemberVx);
      message.twist.twist.linear.y = state(StateMemberVy);
      message.twist.twist.linear.z = state(StateMemberVz);
      message.twist.twist.angular.x = state(StateMemberVroll);
      message.twist.twist.angular.y = state(StateMemberVpitch);
      message.twist.twist.angular.z = state(StateMemberVyaw);

      // Our covariance matrix layout doesn't quite match
      for (size_t i = 0; i < POSE_SIZE; i++)
      {
        for (size_t j = 0; j < POSE_SIZE; j++)
        {
          message.pose.covariance[POSE_SIZE * i + j] = estimateErrorCovariance(i, j);
        }
      }

      // POSE_SIZE and TWIST_SIZE are currently the same size, but we can spare a few
      // cycles to be meticulous and not index a twist covariance array on the size of
      // a pose covariance array
      for (size_t i = 0; i < TWIST_SIZE; i++)
      {
        for (size_t j = 0; j < TWIST_SIZE; j++)
        {
          message.twist.covariance[TWIST_SIZE * i + j] =
              estimateErrorCovariance(i + POSITION_V_OFFSET, j + POSITION_V_OFFSET);
        }
      }

      message.header.stamp = ros::Time(filter_.getLastMeasurementTime());
      message.header.frame_id = worldFrameId_;
      message.child_frame_id = baseLinkOutputFrameId_;
    }

    return filter_.getInitializedStatus();
  }

  template<typename T>
  bool RosFilter<T>::getFilteredAccelMessage(geometry_msgs::AccelWithCovarianceStamped &message)
  {
    // If the filter has received a measurement at some point...
    if (filter_.getInitializedStatus())
    {
      // Grab our current state and covariance estimates
      const Eigen::VectorXd &state = filter_.getState();
      const Eigen::MatrixXd &estimateErrorCovariance = filter_.getEstimateErrorCovariance();

      message.accel.accel.linear.x = state(StateMemberAx);
      message.accel.accel.linear.y = state(StateMemberAy);
      message.accel.accel.linear.z = state(StateMemberAz);

      // Fill the covariance (only the left-upper matrix since we are not estimating
      // the rotational accelerations arround the axes
      for (size_t i = 0; i < ACCELERATION_SIZE; i++)
      {
        for (size_t j = 0; j < ACCELERATION_SIZE; j++)
        {
          // We use the POSE_SIZE since the accel cov matrix of ROS is 6x6
          message.accel.covariance[POSE_SIZE * i + j] =
              estimateErrorCovariance(i + POSITION_A_OFFSET, j + POSITION_A_OFFSET);
        }
      }

      // Fill header information
      message.header.stamp = ros::Time(filter_.getLastMeasurementTime());
      message.header.frame_id = baseLinkOutputFrameId_;
    }

    return filter_.getInitializedStatus();
  }

  template<typename T>
  void RosFilter<T>::imuCallback(const sensor_msgs::Imu::ConstPtr &msg,
                                 const std::string &topicName,
                                 const CallbackData &poseCallbackData,
                                 const CallbackData &twistCallbackData,
                                 const CallbackData &accelCallbackData)
  {
    RF_DEBUG("------ RosFilter::imuCallback (" << topicName << ") ------\n" << "IMU message:\n" << *msg);

    // If we've just reset the filter, then we want to ignore any messages
    // that arrive with an older timestamp
    if (msg->header.stamp <= lastSetPoseTime_)
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp equal to or before the last filter reset, " <<
                "this message will be ignored. This may indicate an empty or bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);
      RF_DEBUG("Received message that preceded the most recent pose reset. Ignoring...");

      return;
    }

    // As with the odometry message, we can separate out the pose- and twist-related variables
    // in the IMU message and pass them to the pose and twist callbacks (filters)
    if (poseCallbackData.updateSum_ > 0)
    {
      // Per the IMU message specification, if the IMU does not provide orientation,
      // then its first covariance value should be set to -1, and we should ignore
      // that portion of the message. robot_localization allows users to explicitly
      // ignore data using its parameters, but we should also be compliant with
      // message specs.
      if (::fabs(msg->orientation_covariance[0] + 1) < 1e-9)
      {
        RF_DEBUG("Received IMU message with -1 as its first covariance value for orientation. "
                 "Ignoring orientation...");
      }
      else
      {
        // Extract the pose (orientation) data, pass it to its filter
        geometry_msgs::PoseWithCovarianceStamped *posPtr = new geometry_msgs::PoseWithCovarianceStamped();
        posPtr->header = msg->header;
        posPtr->pose.pose.orientation = msg->orientation;

        // Copy the covariance for roll, pitch, and yaw
        for (size_t i = 0; i < ORIENTATION_SIZE; i++)
        {
          for (size_t j = 0; j < ORIENTATION_SIZE; j++)
          {
            posPtr->pose.covariance[POSE_SIZE * (i + ORIENTATION_SIZE) + (j + ORIENTATION_SIZE)] =
                msg->orientation_covariance[ORIENTATION_SIZE * i + j];
          }
        }

        // IMU data gets handled a bit differently, since the message is ambiguous and has only a single frame_id,
        // even though the data in it is reported in two different frames. As we assume users will specify a base_link
        // to imu transform, we make the target frame baseLinkFrameId_ and tell the poseCallback that it is working
        // with IMU data. This will cause it to apply different logic to the data.
        geometry_msgs::PoseWithCovarianceStampedConstPtr pptr(posPtr);
        poseCallback(pptr, poseCallbackData, baseLinkFrameId_, true);
      }
    }

    if (twistCallbackData.updateSum_ > 0)
    {
      // Ignore rotational velocity if the first covariance value is -1
      if (::fabs(msg->angular_velocity_covariance[0] + 1) < 1e-9)
      {
        RF_DEBUG("Received IMU message with -1 as its first covariance value for angular "
                 "velocity. Ignoring angular velocity...");
      }
      else
      {
        // Repeat for velocity
        geometry_msgs::TwistWithCovarianceStamped *twistPtr = new geometry_msgs::TwistWithCovarianceStamped();
        twistPtr->header = msg->header;
        twistPtr->twist.twist.angular = msg->angular_velocity;

        // Copy the covariance
        for (size_t i = 0; i < ORIENTATION_SIZE; i++)
        {
          for (size_t j = 0; j < ORIENTATION_SIZE; j++)
          {
            twistPtr->twist.covariance[TWIST_SIZE * (i + ORIENTATION_SIZE) + (j + ORIENTATION_SIZE)] =
              msg->angular_velocity_covariance[ORIENTATION_SIZE * i + j];
          }
        }

        geometry_msgs::TwistWithCovarianceStampedConstPtr tptr(twistPtr);
        twistCallback(tptr, twistCallbackData, baseLinkFrameId_);
      }
    }

    if (accelCallbackData.updateSum_ > 0)
    {
      // Ignore linear acceleration if the first covariance value is -1
      if (::fabs(msg->linear_acceleration_covariance[0] + 1) < 1e-9)
      {
        RF_DEBUG("Received IMU message with -1 as its first covariance value for linear "
                 "acceleration. Ignoring linear acceleration...");
      }
      else
      {
        // Pass the message on
        accelerationCallback(msg, accelCallbackData, baseLinkFrameId_);
      }
    }

    RF_DEBUG("\n----- /RosFilter::imuCallback (" << topicName << ") ------\n");
  }

  template<typename T>
  void RosFilter<T>::integrateMeasurements(const ros::Time &currentTime)
  {
    const double currentTimeSec = currentTime.toSec();

    RF_DEBUG("------ RosFilter::integrateMeasurements ------\n\n"
             "Integration time is " << std::setprecision(20) << currentTimeSec << "\n"
             << measurementQueue_.size() << " measurements in queue.\n");

    bool predictToCurrentTime = predictToCurrentTime_;

    // If we have any measurements in the queue, process them
    if (!measurementQueue_.empty())
    {
      // Check if the first measurement we're going to process is older than the filter's last measurement.
      // This means we have received an out-of-sequence message (one with an old timestamp), and we need to
      // revert both the filter state and measurement queue to the first state that preceded the time stamp
      // of our first measurement.
      const MeasurementPtr& firstMeasurement = measurementQueue_.top();
      int restoredMeasurementCount = 0;
      if (smoothLaggedData_ && firstMeasurement->time_ < filter_.getLastMeasurementTime())
      {
        RF_DEBUG("Received a measurement that was " << filter_.getLastMeasurementTime() - firstMeasurement->time_ <<
                 " seconds in the past. Reverting filter state and measurement queue...");

        int originalCount = static_cast<int>(measurementQueue_.size());
        const double firstMeasurementTime =  firstMeasurement->time_;
        const std::string firstMeasurementTopic =  firstMeasurement->topicName_;
        if (!revertTo(firstMeasurementTime - 1e-9))  // revertTo may invalidate firstMeasurement
        {
          RF_DEBUG("ERROR: history interval is too small to revert to time " << firstMeasurementTime << "\n");
          ROS_WARN_STREAM_DELAYED_THROTTLE(historyLength_, "Received old measurement for topic " <<
            firstMeasurementTopic << ", but history interval is insufficiently sized. Measurement time is " <<
            std::setprecision(20) << firstMeasurementTime << ", current time is " << currentTime.toSec() <<
            ", history length is " << historyLength_ << ".");
          restoredMeasurementCount = 0;
        }

        restoredMeasurementCount = static_cast<int>(measurementQueue_.size()) - originalCount;
      }

      while (!measurementQueue_.empty() && ros::ok())
      {
        MeasurementPtr measurement = measurementQueue_.top();

        // If we've reached a measurement that has a time later than now, it should wait until a future iteration.
        // Since measurements are stored in a priority queue, all remaining measurements will be in the future.
        if (measurement->time_ > currentTime.toSec())
        {
          break;
        }

        measurementQueue_.pop();

        // When we receive control messages, we call this directly in the control callback. However, we also associate
        // a control with each sensor message so that we can support lagged smoothing. As we cannot guarantee that the
        // new control callback will fire before a new measurement, we should only perform this operation if we are
        // processing messages from the history. Otherwise, we may get a new measurement, store the "old" latest
        // control, then receive a control, call setControl, and then overwrite that value with this one (i.e., with
        // the "old" control we associated with the measurement).
        if (useControl_ && restoredMeasurementCount > 0)
        {
          filter_.setControl(measurement->latestControl_, measurement->latestControlTime_);
          restoredMeasurementCount--;
        }

        // This will call predict and, if necessary, correct
        filter_.processMeasurement(*(measurement.get()));

        // Store old states and measurements if we're smoothing
        if (smoothLaggedData_)
        {
          // Invariant still holds: measurementHistoryDeque_.back().time_ < measurementQueue_.top().time_
          measurementHistory_.push_back(measurement);

          // We should only save the filter state once per unique timstamp
          if (measurementQueue_.empty() ||
              ::fabs(measurementQueue_.top()->time_ - filter_.getLastMeasurementTime()) > 1e-9)
          {
            saveFilterState(filter_);
          }
        }
      }
    }
    else if (filter_.getInitializedStatus())
    {
      // In the event that we don't get any measurements for a long time,
      // we still need to continue to estimate our state. Therefore, we
      // should project the state forward here.
      double lastUpdateDelta = currentTimeSec - filter_.getLastMeasurementTime();

      // If we get a large delta, then continuously predict until
      if (lastUpdateDelta >= filter_.getSensorTimeout())
      {
        predictToCurrentTime = true;

        RF_DEBUG("Sensor timeout! Last measurement time was " << filter_.getLastMeasurementTime() <<
                 ", current time is " << currentTimeSec <<
                 ", delta is " << lastUpdateDelta << "\n");
      }
    }
    else
    {
      RF_DEBUG("Filter not yet initialized.\n");
    }

    if (filter_.getInitializedStatus() && predictToCurrentTime)
    {
      double lastUpdateDelta = currentTimeSec - filter_.getLastMeasurementTime();

      filter_.validateDelta(lastUpdateDelta);
      filter_.predict(currentTimeSec, lastUpdateDelta);

      // Update the last measurement time and last update time
      filter_.setLastMeasurementTime(filter_.getLastMeasurementTime() + lastUpdateDelta);
    }

    RF_DEBUG("\n----- /RosFilter::integrateMeasurements ------\n");
  }

  template<typename T>
  void RosFilter<T>::loadParams()
  {
    /* For diagnostic purposes, collect information about how many different
     * sources are measuring each absolute pose variable and do not have
     * differential integration enabled.
     */
    std::map<StateMembers, int> absPoseVarCounts;
    absPoseVarCounts[StateMemberX] = 0;
    absPoseVarCounts[StateMemberY] = 0;
    absPoseVarCounts[StateMemberZ] = 0;
    absPoseVarCounts[StateMemberRoll] = 0;
    absPoseVarCounts[StateMemberPitch] = 0;
    absPoseVarCounts[StateMemberYaw] = 0;

    // Same for twist variables
    std::map<StateMembers, int> twistVarCounts;
    twistVarCounts[StateMemberVx] = 0;
    twistVarCounts[StateMemberVy] = 0;
    twistVarCounts[StateMemberVz] = 0;
    twistVarCounts[StateMemberVroll] = 0;
    twistVarCounts[StateMemberVpitch] = 0;
    twistVarCounts[StateMemberVyaw] = 0;

    // Determine if we'll be printing diagnostic information
    nhLocal_.param("print_diagnostics", printDiagnostics_, true);

    // Check for custom gravitational acceleration value
    nhLocal_.param("gravitational_acceleration", gravitationalAcc_, 9.80665);

    // Grab the debug param. If true, the node will produce a LOT of output.
    bool debug;
    nhLocal_.param("debug", debug, false);

    if (debug)
    {
      std::string debugOutFile;

      try
      {
        nhLocal_.param("debug_out_file", debugOutFile, std::string("robot_localization_debug.txt"));
        debugStream_.open(debugOutFile.c_str());

        // Make sure we succeeded
        if (debugStream_.is_open())
        {
          filter_.setDebug(debug, &debugStream_);
        }
        else
        {
          ROS_WARN_STREAM("RosFilter::loadParams() - unable to create debug output file " << debugOutFile);
        }
      }
      catch(const std::exception &e)
      {
        ROS_WARN_STREAM("RosFilter::loadParams() - unable to create debug output file" << debugOutFile
                        << ". Error was " << e.what() << "\n");
      }
    }

    // These params specify the name of the robot's body frame (typically
    // base_link) and odometry frame (typically odom)
    nhLocal_.param("map_frame", mapFrameId_, std::string("map"));
    nhLocal_.param("odom_frame", odomFrameId_, std::string("odom"));
    nhLocal_.param("base_link_frame", baseLinkFrameId_, std::string("base_link"));
    nhLocal_.param("base_link_frame_output", baseLinkOutputFrameId_, baseLinkFrameId_);

    /*
     * These parameters are designed to enforce compliance with REP-105:
     * http://www.ros.org/reps/rep-0105.html
     * When fusing absolute position data from sensors such as GPS, the state
     * estimate can undergo discrete jumps. According to REP-105, we want three
     * coordinate frames: map, odom, and base_link. The map frame can have
     * discontinuities, but is the frame with the most accurate position estimate
     * for the robot and should not suffer from drift. The odom frame drifts over
     * time, but is guaranteed to be continuous and is accurate enough for local
     * planning and navigation. The base_link frame is affixed to the robot. The
     * intention is that some odometry source broadcasts the odom->base_link
     * transform. The localization software should broadcast map->base_link.
     * However, tf does not allow multiple parents for a coordinate frame, so
     * we must *compute* map->base_link, but then use the existing odom->base_link
     * transform to compute *and broadcast* map->odom.
     *
     * The state estimation nodes in robot_localization therefore have two "modes."
     * If your world_frame parameter value matches the odom_frame parameter value,
     * then robot_localization will assume someone else is broadcasting a transform
     * from odom_frame->base_link_frame, and it will compute the
     * map_frame->odom_frame transform. Otherwise, it will simply compute the
     * odom_frame->base_link_frame transform.
     *
     * The default is the latter behavior (broadcast of odom->base_link).
     */
    nhLocal_.param("world_frame", worldFrameId_, odomFrameId_);

    ROS_FATAL_COND(mapFrameId_ == odomFrameId_ ||
                   odomFrameId_ == baseLinkFrameId_ ||
                   mapFrameId_ == baseLinkFrameId_ ||
                   odomFrameId_ == baseLinkOutputFrameId_ ||
                   mapFrameId_ == baseLinkOutputFrameId_,
                   "Invalid frame configuration! The values for map_frame, odom_frame, "
                   "and base_link_frame must be unique. If using a base_link_frame_output values, it "
                   "must not match the map_frame or odom_frame.");

    // Try to resolve tf_prefix
    std::string tfPrefix = "";
    std::string tfPrefixPath = "";
    if (nhLocal_.searchParam("tf_prefix", tfPrefixPath))
    {
      nhLocal_.getParam(tfPrefixPath, tfPrefix);
    }

    // Append the tf prefix in a tf2-friendly manner
    FilterUtilities::appendPrefix(tfPrefix, mapFrameId_);
    FilterUtilities::appendPrefix(tfPrefix, odomFrameId_);
    FilterUtilities::appendPrefix(tfPrefix, baseLinkFrameId_);
    FilterUtilities::appendPrefix(tfPrefix, baseLinkOutputFrameId_);
    FilterUtilities::appendPrefix(tfPrefix, worldFrameId_);

    // Whether we're publshing the world_frame->base_link_frame transform
    nhLocal_.param("publish_tf", publishTransform_, true);

    // Whether we're publishing the acceleration state transform
    nhLocal_.param("publish_acceleration", publishAcceleration_, false);

    // Transform future dating
    double offsetTmp;
    nhLocal_.param("transform_time_offset", offsetTmp, 0.0);
    tfTimeOffset_.fromSec(offsetTmp);

    // Transform timeout
    double timeoutTmp;
    nhLocal_.param("transform_timeout", timeoutTmp, 0.0);
    tfTimeout_.fromSec(timeoutTmp);

    // Update frequency and sensor timeout
    double sensorTimeout;
    nhLocal_.param("frequency", frequency_, 30.0);
    nhLocal_.param("sensor_timeout", sensorTimeout, 1.0 / frequency_);
    filter_.setSensorTimeout(sensorTimeout);

    // Determine if we're in 2D mode
    nhLocal_.param("two_d_mode", twoDMode_, false);

    // Smoothing window size
    nhLocal_.param("smooth_lagged_data", smoothLaggedData_, false);
    nhLocal_.param("history_length", historyLength_, 0.0);

    // Wether we reset filter on jump back in time
    nhLocal_.param("reset_on_time_jump", resetOnTimeJump_, false);

    if (!smoothLaggedData_ && ::fabs(historyLength_) > 1e-9)
    {
      ROS_WARN_STREAM("Filter history interval of " << historyLength_ <<
                      " specified, but smooth_lagged_data is set to false. Lagged data will not be smoothed.");
    }

    if (smoothLaggedData_ && historyLength_ < -1e9)
    {
      ROS_WARN_STREAM("Negative history interval of " << historyLength_ <<
                      " specified. Absolute value will be assumed.");
    }

    historyLength_ = ::fabs(historyLength_);

    nhLocal_.param("predict_to_current_time", predictToCurrentTime_, false);

    // Determine if we're using a control term
    bool stampedControl = false;
    double controlTimeout = sensorTimeout;
    std::vector<int> controlUpdateVector(TWIST_SIZE, 0);
    std::vector<double> accelerationLimits(TWIST_SIZE, 1.0);
    std::vector<double> accelerationGains(TWIST_SIZE, 1.0);
    std::vector<double> decelerationLimits(TWIST_SIZE, 1.0);
    std::vector<double> decelerationGains(TWIST_SIZE, 1.0);

    nhLocal_.param("use_control", useControl_, false);
    nhLocal_.param("stamped_control", stampedControl, false);
    nhLocal_.param("control_timeout", controlTimeout, sensorTimeout);

    if (useControl_)
    {
      if (nhLocal_.getParam("control_config", controlUpdateVector))
      {
        if (controlUpdateVector.size() != TWIST_SIZE)
        {
          ROS_ERROR_STREAM("Control configuration must be of size " << TWIST_SIZE << ". Provided config was of "
            "size " << controlUpdateVector.size() << ". No control term will be used.");
          useControl_ = false;
        }
      }
      else
      {
        ROS_ERROR_STREAM("use_control is set to true, but control_config is missing. No control term will be used.");
        useControl_ = false;
      }

      if (nhLocal_.getParam("acceleration_limits", accelerationLimits))
      {
        if (accelerationLimits.size() != TWIST_SIZE)
        {
          ROS_ERROR_STREAM("Acceleration configuration must be of size " << TWIST_SIZE << ". Provided config was of "
            "size " << accelerationLimits.size() << ". No control term will be used.");
          useControl_ = false;
        }
      }
      else
      {
        ROS_WARN_STREAM("use_control is set to true, but acceleration_limits is missing. Will use default values.");
      }

      if (nhLocal_.getParam("acceleration_gains", accelerationGains))
      {
        const int size = accelerationGains.size();
        if (size != TWIST_SIZE)
        {
          ROS_ERROR_STREAM("Acceleration gain configuration must be of size " << TWIST_SIZE <<
            ". Provided config was of size " << size << ". All gains will be assumed to be 1.");
          std::fill_n(accelerationGains.begin(), std::min(size, TWIST_SIZE), 1.0);
          accelerationGains.resize(TWIST_SIZE, 1.0);
        }
      }

      if (nhLocal_.getParam("deceleration_limits", decelerationLimits))
      {
        if (decelerationLimits.size() != TWIST_SIZE)
        {
          ROS_ERROR_STREAM("Deceleration configuration must be of size " << TWIST_SIZE <<
            ". Provided config was of size " << decelerationLimits.size() << ". No control term will be used.");
          useControl_ = false;
        }
      }
      else
      {
        ROS_INFO_STREAM("use_control is set to true, but no deceleration_limits specified. Will use acceleration "
          "limits.");
        decelerationLimits = accelerationLimits;
      }

      if (nhLocal_.getParam("deceleration_gains", decelerationGains))
      {
        const int size = decelerationGains.size();
        if (size != TWIST_SIZE)
        {
          ROS_ERROR_STREAM("Deceleration gain configuration must be of size " << TWIST_SIZE <<
            ". Provided config was of size " << size << ". All gains will be assumed to be 1.");
          std::fill_n(decelerationGains.begin(), std::min(size, TWIST_SIZE), 1.0);
          decelerationGains.resize(TWIST_SIZE, 1.0);
        }
      }
      else
      {
        ROS_INFO_STREAM("use_control is set to true, but no deceleration_gains specified. Will use acceleration "
          "gains.");
        decelerationGains = accelerationGains;
      }
    }

    bool dynamicProcessNoiseCovariance = false;
    nhLocal_.param("dynamic_process_noise_covariance", dynamicProcessNoiseCovariance, false);
    filter_.setUseDynamicProcessNoiseCovariance(dynamicProcessNoiseCovariance);

    std::vector<double> initialState(STATE_SIZE, 0.0);
    if (nhLocal_.getParam("initial_state", initialState))
    {
      if (initialState.size() != STATE_SIZE)
      {
        ROS_ERROR_STREAM("Initial state must be of size " << STATE_SIZE << ". Provided config was of size " <<
          initialState.size() << ". The initial state will be ignored.");
      }
      else
      {
        Eigen::Map<Eigen::VectorXd> eigenState(initialState.data(), initialState.size());
        filter_.setState(eigenState);
      }
    }

    // Check if the filter should start or not
    nhLocal_.param("disabled_at_startup", disabledAtStartup_, false);
    enabled_ = !disabledAtStartup_;


    // Debugging writes to file
    RF_DEBUG("tf_prefix is " << tfPrefix <<
             "\nmap_frame is " << mapFrameId_ <<
             "\nodom_frame is " << odomFrameId_ <<
             "\nbase_link_frame is " << baseLinkFrameId_ <<
             "\base_link_frame_output is " << baseLinkOutputFrameId_ <<
             "\nworld_frame is " << worldFrameId_ <<
             "\ntransform_time_offset is " << tfTimeOffset_.toSec() <<
             "\ntransform_timeout is " << tfTimeout_.toSec() <<
             "\nfrequency is " << frequency_ <<
             "\nsensor_timeout is " << filter_.getSensorTimeout() <<
             "\ntwo_d_mode is " << (twoDMode_ ? "true" : "false") <<
             "\nsmooth_lagged_data is " << (smoothLaggedData_ ? "true" : "false") <<
             "\nhistory_length is " << historyLength_ <<
             "\nuse_control is " << (useControl_ ? "true" : "false") <<
             "\nstamped_control is " << (stampedControl ? "true" : "false") <<
             "\ncontrol_config is " << controlUpdateVector <<
             "\ncontrol_timeout is " << controlTimeout <<
             "\nacceleration_limits are " << accelerationLimits <<
             "\nacceleration_gains are " << accelerationGains <<
             "\ndeceleration_limits are " << decelerationLimits <<
             "\ndeceleration_gains are " << decelerationGains <<
             "\ninitial state is " << filter_.getState() <<
             "\ndynamic_process_noise_covariance is " << (dynamicProcessNoiseCovariance ? "true" : "false") <<
             "\nprint_diagnostics is " << (printDiagnostics_ ? "true" : "false") << "\n");

    // Create a subscriber for manually setting/resetting pose
    setPoseSub_ = nh_.subscribe("set_pose",
                                1,
                                &RosFilter<T>::setPoseCallback,
                                this, ros::TransportHints().tcpNoDelay(false));

    // Create a service for manually setting/resetting pose
    setPoseSrv_ = nh_.advertiseService("set_pose", &RosFilter<T>::setPoseSrvCallback, this);

    // Create a service for manually enabling the filter
    enableFilterSrv_ = nhLocal_.advertiseService("enable", &RosFilter<T>::enableFilterSrvCallback, this);

    // Create a service for toggling processing new measurements while still publishing
    toggleFilterProcessingSrv_ =
      nhLocal_.advertiseService("toggle", &RosFilter<T>::toggleFilterProcessingCallback, this);

    // Init the last measurement time so we don't get a huge initial delta
    filter_.setLastMeasurementTime(ros::Time::now().toSec());

    // Now pull in each topic to which we want to subscribe.
    // Start with odom.
    size_t topicInd = 0;
    bool moreParams = false;
    do
    {
      // Build the string in the form of "odomX", where X is the odom topic number,
      // then check if we have any parameters with that value. Users need to make
      // sure they don't have gaps in their configs (e.g., odom0 and then odom2)
      std::stringstream ss;
      ss << "odom" << topicInd++;
      std::string odomTopicName = ss.str();
      moreParams = nhLocal_.hasParam(odomTopicName);

      if (moreParams)
      {
        // Determine if we want to integrate this sensor differentially
        bool differential;
        nhLocal_.param(odomTopicName + std::string("_differential"), differential, false);

        // Determine if we want to integrate this sensor relatively
        bool relative;
        nhLocal_.param(odomTopicName + std::string("_relative"), relative, false);

        if (relative && differential)
        {
          ROS_WARN_STREAM("Both " << odomTopicName << "_differential" << " and " << odomTopicName <<
                          "_relative were set to true. Using differential mode.");

          relative = false;
        }

        std::string odomTopic;
        nhLocal_.getParam(odomTopicName, odomTopic);

        // Check for pose rejection threshold
        double poseMahalanobisThresh;
        nhLocal_.param(odomTopicName + std::string("_pose_rejection_threshold"),
                       poseMahalanobisThresh,
                       std::numeric_limits<double>::max());

        // Check for twist rejection threshold
        double twistMahalanobisThresh;
        nhLocal_.param(odomTopicName + std::string("_twist_rejection_threshold"),
                       twistMahalanobisThresh,
                       std::numeric_limits<double>::max());

        // Now pull in its boolean update vector configuration. Create separate vectors for pose
        // and twist data, and then zero out the opposite values in each vector (no pose data in
        // the twist update vector and vice-versa).
        std::vector<int> updateVec = loadUpdateConfig(odomTopicName);
        std::vector<int> poseUpdateVec = updateVec;
        std::fill(poseUpdateVec.begin() + POSITION_V_OFFSET, poseUpdateVec.begin() + POSITION_V_OFFSET + TWIST_SIZE, 0);
        std::vector<int> twistUpdateVec = updateVec;
        std::fill(twistUpdateVec.begin() + POSITION_OFFSET, twistUpdateVec.begin() + POSITION_OFFSET + POSE_SIZE, 0);

        int poseUpdateSum = std::accumulate(poseUpdateVec.begin(), poseUpdateVec.end(), 0);
        int twistUpdateSum = std::accumulate(twistUpdateVec.begin(), twistUpdateVec.end(), 0);
        int odomQueueSize = 1;
        nhLocal_.param(odomTopicName + "_queue_size", odomQueueSize, 1);

        const CallbackData poseCallbackData(odomTopicName + "_pose", poseUpdateVec, poseUpdateSum, differential,
          relative, poseMahalanobisThresh);
        const CallbackData twistCallbackData(odomTopicName + "_twist", twistUpdateVec, twistUpdateSum, false, false,
          twistMahalanobisThresh);

        bool nodelayOdom = false;
        nhLocal_.param(odomTopicName + "_nodelay", nodelayOdom, false);

        // Store the odometry topic subscribers so they don't go out of scope.
        if (poseUpdateSum + twistUpdateSum > 0)
        {
          topicSubs_.push_back(
            nh_.subscribe<nav_msgs::Odometry>(odomTopic, odomQueueSize,
              boost::bind(&RosFilter::odometryCallback, this, _1, odomTopicName, poseCallbackData, twistCallbackData),
              ros::VoidPtr(), ros::TransportHints().tcpNoDelay(nodelayOdom)));
        }
        else
        {
          std::stringstream stream;
          stream << odomTopic << " is listed as an input topic, but all update variables are false";

          addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                        odomTopic + "_configuration",
                        stream.str(),
                        true);
        }

        if (poseUpdateSum > 0)
        {
          if (differential)
          {
            twistVarCounts[StateMemberVx] += poseUpdateVec[StateMemberX];
            twistVarCounts[StateMemberVy] += poseUpdateVec[StateMemberY];
            twistVarCounts[StateMemberVz] += poseUpdateVec[StateMemberZ];
            twistVarCounts[StateMemberVroll] += poseUpdateVec[StateMemberRoll];
            twistVarCounts[StateMemberVpitch] += poseUpdateVec[StateMemberPitch];
            twistVarCounts[StateMemberVyaw] += poseUpdateVec[StateMemberYaw];
          }
          else
          {
            absPoseVarCounts[StateMemberX] += poseUpdateVec[StateMemberX];
            absPoseVarCounts[StateMemberY] += poseUpdateVec[StateMemberY];
            absPoseVarCounts[StateMemberZ] += poseUpdateVec[StateMemberZ];
            absPoseVarCounts[StateMemberRoll] += poseUpdateVec[StateMemberRoll];
            absPoseVarCounts[StateMemberPitch] += poseUpdateVec[StateMemberPitch];
            absPoseVarCounts[StateMemberYaw] += poseUpdateVec[StateMemberYaw];
          }
        }

        if (twistUpdateSum > 0)
        {
          twistVarCounts[StateMemberVx] += twistUpdateVec[StateMemberVx];
          twistVarCounts[StateMemberVy] += twistUpdateVec[StateMemberVx];
          twistVarCounts[StateMemberVz] += twistUpdateVec[StateMemberVz];
          twistVarCounts[StateMemberVroll] += twistUpdateVec[StateMemberVroll];
          twistVarCounts[StateMemberVpitch] += twistUpdateVec[StateMemberVpitch];
          twistVarCounts[StateMemberVyaw] += twistUpdateVec[StateMemberVyaw];
        }

        RF_DEBUG("Subscribed to " << odomTopic << " (" << odomTopicName << ")\n\t" <<
                 odomTopicName << "_differential is " << (differential ? "true" : "false") << "\n\t" <<
                 odomTopicName << "_pose_rejection_threshold is " << poseMahalanobisThresh << "\n\t" <<
                 odomTopicName << "_twist_rejection_threshold is " << twistMahalanobisThresh << "\n\t" <<
                 odomTopicName << "_queue_size is " << odomQueueSize << "\n\t" <<
                 odomTopicName << " pose update vector is " << poseUpdateVec << "\t"<<
                 odomTopicName << " twist update vector is " << twistUpdateVec);
      }
    }
    while (moreParams);

    // Repeat for pose
    topicInd = 0;
    moreParams = false;
    do
    {
      std::stringstream ss;
      ss << "pose" << topicInd++;
      std::string poseTopicName = ss.str();
      moreParams = nhLocal_.hasParam(poseTopicName);

      if (moreParams)
      {
        bool differential;
        nhLocal_.param(poseTopicName + std::string("_differential"), differential, false);

        // Determine if we want to integrate this sensor relatively
        bool relative;
        nhLocal_.param(poseTopicName + std::string("_relative"), relative, false);

        if (relative && differential)
        {
          ROS_WARN_STREAM("Both " << poseTopicName << "_differential" << " and " << poseTopicName <<
                          "_relative were set to true. Using differential mode.");

          relative = false;
        }

        std::string poseTopic;
        nhLocal_.getParam(poseTopicName, poseTopic);

        // Check for pose rejection threshold
        double poseMahalanobisThresh;
        nhLocal_.param(poseTopicName + std::string("_rejection_threshold"),
                       poseMahalanobisThresh,
                       std::numeric_limits<double>::max());

        int poseQueueSize = 1;
        nhLocal_.param(poseTopicName + "_queue_size", poseQueueSize, 1);

        bool nodelayPose = false;
        nhLocal_.param(poseTopicName + "_nodelay", nodelayPose, false);

        // Pull in the sensor's config, zero out values that are invalid for the pose type
        std::vector<int> poseUpdateVec = loadUpdateConfig(poseTopicName);
        std::fill(poseUpdateVec.begin() + POSITION_V_OFFSET,
                  poseUpdateVec.begin() + POSITION_V_OFFSET + TWIST_SIZE,
                  0);
        std::fill(poseUpdateVec.begin() + POSITION_A_OFFSET,
                  poseUpdateVec.begin() + POSITION_A_OFFSET + ACCELERATION_SIZE,
                  0);

        int poseUpdateSum = std::accumulate(poseUpdateVec.begin(), poseUpdateVec.end(), 0);

        if (poseUpdateSum > 0)
        {
          const CallbackData callbackData(poseTopicName, poseUpdateVec, poseUpdateSum, differential, relative,
            poseMahalanobisThresh);

          topicSubs_.push_back(
            nh_.subscribe<geometry_msgs::PoseWithCovarianceStamped>(poseTopic, poseQueueSize,
              boost::bind(&RosFilter::poseCallback, this, _1, callbackData, worldFrameId_, false),
              ros::VoidPtr(), ros::TransportHints().tcpNoDelay(nodelayPose)));

          if (differential)
          {
            twistVarCounts[StateMemberVx] += poseUpdateVec[StateMemberX];
            twistVarCounts[StateMemberVy] += poseUpdateVec[StateMemberY];
            twistVarCounts[StateMemberVz] += poseUpdateVec[StateMemberZ];
            twistVarCounts[StateMemberVroll] += poseUpdateVec[StateMemberRoll];
            twistVarCounts[StateMemberVpitch] += poseUpdateVec[StateMemberPitch];
            twistVarCounts[StateMemberVyaw] += poseUpdateVec[StateMemberYaw];
          }
          else
          {
            absPoseVarCounts[StateMemberX] += poseUpdateVec[StateMemberX];
            absPoseVarCounts[StateMemberY] += poseUpdateVec[StateMemberY];
            absPoseVarCounts[StateMemberZ] += poseUpdateVec[StateMemberZ];
            absPoseVarCounts[StateMemberRoll] += poseUpdateVec[StateMemberRoll];
            absPoseVarCounts[StateMemberPitch] += poseUpdateVec[StateMemberPitch];
            absPoseVarCounts[StateMemberYaw] += poseUpdateVec[StateMemberYaw];
          }
        }
        else
        {
          ROS_WARN_STREAM("Warning: " << poseTopic << " is listed as an input topic, "
                          "but all pose update variables are false");
        }

        RF_DEBUG("Subscribed to " << poseTopic << " (" << poseTopicName << ")\n\t" <<
                 poseTopicName << "_differential is " << (differential ? "true" : "false") << "\n\t" <<
                 poseTopicName << "_rejection_threshold is " << poseMahalanobisThresh << "\n\t" <<
                 poseTopicName << "_queue_size is " << poseQueueSize << "\n\t" <<
                 poseTopicName << " update vector is " << poseUpdateVec);
      }
    }
    while (moreParams);

    // Repeat for twist
    topicInd = 0;
    moreParams = false;
    do
    {
      std::stringstream ss;
      ss << "twist" << topicInd++;
      std::string twistTopicName = ss.str();
      moreParams = nhLocal_.hasParam(twistTopicName);

      if (moreParams)
      {
        std::string twistTopic;
        nhLocal_.getParam(twistTopicName, twistTopic);

        // Check for twist rejection threshold
        double twistMahalanobisThresh;
        nhLocal_.param(twistTopicName + std::string("_rejection_threshold"),
                       twistMahalanobisThresh,
                       std::numeric_limits<double>::max());

        int twistQueueSize = 1;
        nhLocal_.param(twistTopicName + "_queue_size", twistQueueSize, 1);

        bool nodelayTwist = false;
        nhLocal_.param(twistTopicName + "_nodelay", nodelayTwist, false);

        // Pull in the sensor's config, zero out values that are invalid for the twist type
        std::vector<int> twistUpdateVec = loadUpdateConfig(twistTopicName);
        std::fill(twistUpdateVec.begin() + POSITION_OFFSET, twistUpdateVec.begin() + POSITION_OFFSET + POSE_SIZE, 0);

        int twistUpdateSum = std::accumulate(twistUpdateVec.begin(), twistUpdateVec.end(), 0);

        if (twistUpdateSum > 0)
        {
          const CallbackData callbackData(twistTopicName, twistUpdateVec, twistUpdateSum, false, false,
            twistMahalanobisThresh);

          topicSubs_.push_back(
            nh_.subscribe<geometry_msgs::TwistWithCovarianceStamped>(twistTopic, twistQueueSize,
              boost::bind(&RosFilter<T>::twistCallback, this, _1, callbackData, baseLinkFrameId_),
              ros::VoidPtr(), ros::TransportHints().tcpNoDelay(nodelayTwist)));

          twistVarCounts[StateMemberVx] += twistUpdateVec[StateMemberVx];
          twistVarCounts[StateMemberVy] += twistUpdateVec[StateMemberVy];
          twistVarCounts[StateMemberVz] += twistUpdateVec[StateMemberVz];
          twistVarCounts[StateMemberVroll] += twistUpdateVec[StateMemberVroll];
          twistVarCounts[StateMemberVpitch] += twistUpdateVec[StateMemberVpitch];
          twistVarCounts[StateMemberVyaw] += twistUpdateVec[StateMemberVyaw];
        }
        else
        {
          ROS_WARN_STREAM("Warning: " << twistTopic << " is listed as an input topic, "
                          "but all twist update variables are false");
        }

        RF_DEBUG("Subscribed to " << twistTopic << " (" << twistTopicName << ")\n\t" <<
                 twistTopicName << "_rejection_threshold is " << twistMahalanobisThresh << "\n\t" <<
                 twistTopicName << "_queue_size is " << twistQueueSize << "\n\t" <<
                 twistTopicName << " update vector is " << twistUpdateVec);
      }
    }
    while (moreParams);

    // Repeat for IMU
    topicInd = 0;
    moreParams = false;
    do
    {
      std::stringstream ss;
      ss << "imu" << topicInd++;
      std::string imuTopicName = ss.str();
      moreParams = nhLocal_.hasParam(imuTopicName);

      if (moreParams)
      {
        bool differential;
        nhLocal_.param(imuTopicName + std::string("_differential"), differential, false);

        // Determine if we want to integrate this sensor relatively
        bool relative;
        nhLocal_.param(imuTopicName + std::string("_relative"), relative, false);

        if (relative && differential)
        {
          ROS_WARN_STREAM("Both " << imuTopicName << "_differential" << " and " << imuTopicName <<
                          "_relative were set to true. Using differential mode.");

          relative = false;
        }

        std::string imuTopic;
        nhLocal_.getParam(imuTopicName, imuTopic);

        // Check for pose rejection threshold
        double poseMahalanobisThresh;
        nhLocal_.param(imuTopicName + std::string("_pose_rejection_threshold"),
                       poseMahalanobisThresh,
                       std::numeric_limits<double>::max());

        // Check for angular velocity rejection threshold
        double twistMahalanobisThresh;
        std::string imuTwistRejectionName =
          imuTopicName + std::string("_twist_rejection_threshold");
        nhLocal_.param(imuTwistRejectionName, twistMahalanobisThresh, std::numeric_limits<double>::max());

        // Check for acceleration rejection threshold
        double accelMahalanobisThresh;
        nhLocal_.param(imuTopicName + std::string("_linear_acceleration_rejection_threshold"),
                       accelMahalanobisThresh,
                       std::numeric_limits<double>::max());

        bool removeGravAcc = false;
        nhLocal_.param(imuTopicName + "_remove_gravitational_acceleration", removeGravAcc, false);
        removeGravitationalAcc_[imuTopicName + "_acceleration"] = removeGravAcc;

        // Now pull in its boolean update vector configuration and differential
        // update configuration (as this contains pose information)
        std::vector<int> updateVec = loadUpdateConfig(imuTopicName);

        std::vector<int> poseUpdateVec = updateVec;
        std::fill(poseUpdateVec.begin() + POSITION_V_OFFSET,
                  poseUpdateVec.begin() + POSITION_V_OFFSET + TWIST_SIZE,
                  0);
        std::fill(poseUpdateVec.begin() + POSITION_A_OFFSET,
                  poseUpdateVec.begin() + POSITION_A_OFFSET + ACCELERATION_SIZE,
                  0);

        std::vector<int> twistUpdateVec = updateVec;
        std::fill(twistUpdateVec.begin() + POSITION_OFFSET,
                  twistUpdateVec.begin() + POSITION_OFFSET + POSE_SIZE,
                  0);
        std::fill(twistUpdateVec.begin() + POSITION_A_OFFSET,
                  twistUpdateVec.begin() + POSITION_A_OFFSET + ACCELERATION_SIZE,
                  0);

        std::vector<int> accelUpdateVec = updateVec;
        std::fill(accelUpdateVec.begin() + POSITION_OFFSET,
                  accelUpdateVec.begin() + POSITION_OFFSET + POSE_SIZE,
                  0);
        std::fill(accelUpdateVec.begin() + POSITION_V_OFFSET,
                  accelUpdateVec.begin() + POSITION_V_OFFSET + TWIST_SIZE,
                  0);

        int poseUpdateSum = std::accumulate(poseUpdateVec.begin(), poseUpdateVec.end(), 0);
        int twistUpdateSum = std::accumulate(twistUpdateVec.begin(), twistUpdateVec.end(), 0);
        int accelUpdateSum = std::accumulate(accelUpdateVec.begin(), accelUpdateVec.end(), 0);

        // Check if we're using control input for any of the acceleration variables; turn off if so
        if (static_cast<bool>(controlUpdateVector[ControlMemberVx]) && static_cast<bool>(accelUpdateVec[StateMemberAx]))
        {
          ROS_WARN_STREAM("X acceleration is being measured from IMU; X velocity control input is disabled");
          controlUpdateVector[ControlMemberVx] = 0;
        }
        if (static_cast<bool>(controlUpdateVector[ControlMemberVy]) && static_cast<bool>(accelUpdateVec[StateMemberAy]))
        {
          ROS_WARN_STREAM("Y acceleration is being measured from IMU; Y velocity control input is disabled");
          controlUpdateVector[ControlMemberVy] = 0;
        }
        if (static_cast<bool>(controlUpdateVector[ControlMemberVz]) && static_cast<bool>(accelUpdateVec[StateMemberAz]))
        {
          ROS_WARN_STREAM("Z acceleration is being measured from IMU; Z velocity control input is disabled");
          controlUpdateVector[ControlMemberVz] = 0;
        }

        int imuQueueSize = 1;
        nhLocal_.param(imuTopicName + "_queue_size", imuQueueSize, 1);

        bool nodelayImu = false;
        nhLocal_.param(imuTopicName + "_nodelay", nodelayImu, false);

        if (poseUpdateSum + twistUpdateSum + accelUpdateSum > 0)
        {
          const CallbackData poseCallbackData(imuTopicName + "_pose", poseUpdateVec, poseUpdateSum, differential,
            relative, poseMahalanobisThresh);
          const CallbackData twistCallbackData(imuTopicName + "_twist", twistUpdateVec, twistUpdateSum, differential,
            relative, twistMahalanobisThresh);
          const CallbackData accelCallbackData(imuTopicName + "_acceleration", accelUpdateVec, accelUpdateSum,
            differential, relative, accelMahalanobisThresh);

          topicSubs_.push_back(
            nh_.subscribe<sensor_msgs::Imu>(imuTopic, imuQueueSize,
              boost::bind(&RosFilter<T>::imuCallback, this, _1, imuTopicName, poseCallbackData, twistCallbackData,
                accelCallbackData), ros::VoidPtr(), ros::TransportHints().tcpNoDelay(nodelayImu)));
        }
        else
        {
          ROS_WARN_STREAM("Warning: " << imuTopic << " is listed as an input topic, "
                          "but all its update variables are false");
        }

        if (poseUpdateSum > 0)
        {
          if (differential)
          {
            twistVarCounts[StateMemberVroll] += poseUpdateVec[StateMemberRoll];
            twistVarCounts[StateMemberVpitch] += poseUpdateVec[StateMemberPitch];
            twistVarCounts[StateMemberVyaw] += poseUpdateVec[StateMemberYaw];
          }
          else
          {
            absPoseVarCounts[StateMemberRoll] += poseUpdateVec[StateMemberRoll];
            absPoseVarCounts[StateMemberPitch] += poseUpdateVec[StateMemberPitch];
            absPoseVarCounts[StateMemberYaw] += poseUpdateVec[StateMemberYaw];
          }
        }

        if (twistUpdateSum > 0)
        {
          twistVarCounts[StateMemberVroll] += twistUpdateVec[StateMemberVroll];
          twistVarCounts[StateMemberVpitch] += twistUpdateVec[StateMemberVpitch];
          twistVarCounts[StateMemberVyaw] += twistUpdateVec[StateMemberVyaw];
        }

        RF_DEBUG("Subscribed to " << imuTopic << " (" << imuTopicName << ")\n\t" <<
                 imuTopicName << "_differential is " << (differential ? "true" : "false") << "\n\t" <<
                 imuTopicName << "_pose_rejection_threshold is " << poseMahalanobisThresh << "\n\t" <<
                 imuTopicName << "_twist_rejection_threshold is " << twistMahalanobisThresh << "\n\t" <<
                 imuTopicName << "_linear_acceleration_rejection_threshold is " << accelMahalanobisThresh << "\n\t" <<
                 imuTopicName << "_remove_gravitational_acceleration is " <<
                                 (removeGravAcc ? "true" : "false") << "\n\t" <<
                 imuTopicName << "_queue_size is " << imuQueueSize << "\n\t" <<
                 imuTopicName << " pose update vector is " << poseUpdateVec << "\t"<<
                 imuTopicName << " twist update vector is " << twistUpdateVec << "\t" <<
                 imuTopicName << " acceleration update vector is " << accelUpdateVec);
      }
    }
    while (moreParams);

    // Now that we've checked if IMU linear acceleration is being used, we can determine our final control parameters
    if (useControl_ && std::accumulate(controlUpdateVector.begin(), controlUpdateVector.end(), 0) == 0)
    {
      ROS_ERROR_STREAM("use_control is set to true, but control_config has only false values. No control term "
        "will be used.");
      useControl_ = false;
    }

    // If we're using control, set the parameters and create the necessary subscribers
    if (useControl_)
    {
      latestControl_.resize(TWIST_SIZE);
      latestControl_.setZero();

      filter_.setControlParams(controlUpdateVector, controlTimeout, accelerationLimits, accelerationGains,
        decelerationLimits, decelerationGains);

      if (stampedControl)
      {
        controlSub_ = nh_.subscribe<geometry_msgs::TwistStamped>("cmd_vel", 1, &RosFilter<T>::controlCallback, this);
      }
      else
      {
        controlSub_ = nh_.subscribe<geometry_msgs::Twist>("cmd_vel", 1, &RosFilter<T>::controlCallback, this);
      }
    }

    /* Warn users about:
    *    1. Multiple non-differential input sources
    *    2. No absolute *or* velocity measurements for pose variables
    */
    if (printDiagnostics_)
    {
      for (int stateVar = StateMemberX; stateVar <= StateMemberYaw; ++stateVar)
      {
        if (absPoseVarCounts[static_cast<StateMembers>(stateVar)] > 1)
        {
          std::stringstream stream;
          stream <<  absPoseVarCounts[static_cast<StateMembers>(stateVar - POSITION_OFFSET)] <<
              " absolute pose inputs detected for " << stateVariableNames_[stateVar] <<
              ". This may result in oscillations. Please ensure that your variances for each "
              "measured variable are set appropriately.";

          addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                        stateVariableNames_[stateVar] + "_configuration",
                        stream.str(),
                        true);
        }
        else if (absPoseVarCounts[static_cast<StateMembers>(stateVar)] == 0)
        {
          if ((static_cast<StateMembers>(stateVar) == StateMemberX &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVx)] == 0) ||
              (static_cast<StateMembers>(stateVar) == StateMemberY &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVy)] == 0) ||
              (static_cast<StateMembers>(stateVar) == StateMemberZ &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVz)] == 0 &&
               twoDMode_ == false) ||
              (static_cast<StateMembers>(stateVar) == StateMemberRoll &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVroll)] == 0 &&
               twoDMode_ == false) ||
              (static_cast<StateMembers>(stateVar) == StateMemberPitch &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVpitch)] == 0 &&
               twoDMode_ == false) ||
              (static_cast<StateMembers>(stateVar) == StateMemberYaw &&
               twistVarCounts[static_cast<StateMembers>(StateMemberVyaw)] == 0))
          {
            std::stringstream stream;
            stream << "Neither " << stateVariableNames_[stateVar] << " nor its "
                      "velocity is being measured. This will result in unbounded "
                      "error growth and erratic filter behavior.";

            addDiagnostic(diagnostic_msgs::DiagnosticStatus::ERROR,
                          stateVariableNames_[stateVar] + "_configuration",
                          stream.str(),
                          true);
          }
        }
      }
    }

    // Load up the process noise covariance (from the launch file/parameter server)
    Eigen::MatrixXd processNoiseCovariance(STATE_SIZE, STATE_SIZE);
    processNoiseCovariance.setZero();
    XmlRpc::XmlRpcValue processNoiseCovarConfig;

    if (nhLocal_.hasParam("process_noise_covariance"))
    {
      try
      {
        nhLocal_.getParam("process_noise_covariance", processNoiseCovarConfig);

        ROS_ASSERT(processNoiseCovarConfig.getType() == XmlRpc::XmlRpcValue::TypeArray);

        int matSize = processNoiseCovariance.rows();

        for (int i = 0; i < matSize; i++)
        {
          for (int j = 0; j < matSize; j++)
          {
            try
            {
              // These matrices can cause problems if all the types
              // aren't specified with decimal points. Handle that
              // using string streams.
              std::ostringstream ostr;
              ostr << processNoiseCovarConfig[matSize * i + j];
              std::istringstream istr(ostr.str());
              istr >> processNoiseCovariance(i, j);
            }
            catch(XmlRpc::XmlRpcException &e)
            {
              throw e;
            }
            catch(...)
            {
              throw;
            }
          }
        }

        RF_DEBUG("Process noise covariance is:\n" << processNoiseCovariance << "\n");
      }
      catch (XmlRpc::XmlRpcException &e)
      {
        ROS_ERROR_STREAM("ERROR reading sensor config: " <<
                         e.getMessage() <<
                         " for process_noise_covariance (type: " <<
                         processNoiseCovarConfig.getType() << ")");
      }

      filter_.setProcessNoiseCovariance(processNoiseCovariance);
    }

    // Load up the process noise covariance (from the launch file/parameter server)
    Eigen::MatrixXd initialEstimateErrorCovariance(STATE_SIZE, STATE_SIZE);
    initialEstimateErrorCovariance.setZero();
    XmlRpc::XmlRpcValue estimateErrorCovarConfig;

    if (nhLocal_.hasParam("initial_estimate_covariance"))
    {
      try
      {
        nhLocal_.getParam("initial_estimate_covariance", estimateErrorCovarConfig);

        ROS_ASSERT(estimateErrorCovarConfig.getType() == XmlRpc::XmlRpcValue::TypeArray);

        int matSize = initialEstimateErrorCovariance.rows();

        for (int i = 0; i < matSize; i++)
        {
          for (int j = 0; j < matSize; j++)
          {
            try
            {
              // These matrices can cause problems if all the types
              // aren't specified with decimal points. Handle that
              // using string streams.
              std::ostringstream ostr;
              ostr << estimateErrorCovarConfig[matSize * i + j];
              std::istringstream istr(ostr.str());
              istr >> initialEstimateErrorCovariance(i, j);
            }
            catch(XmlRpc::XmlRpcException &e)
            {
              throw e;
            }
            catch(...)
            {
              throw;
            }
          }
        }

        RF_DEBUG("Initial estimate error covariance is:\n" << initialEstimateErrorCovariance << "\n");
      }
      catch (XmlRpc::XmlRpcException &e)
      {
        ROS_ERROR_STREAM("ERROR reading initial_estimate_covariance (type: " <<
                         estimateErrorCovarConfig.getType() <<
                         "): " <<
                         e.getMessage());
      }
      catch(...)
      {
        ROS_ERROR_STREAM(
          "ERROR reading initial_estimate_covariance (type: " << estimateErrorCovarConfig.getType() << ")");
      }

      filter_.setEstimateErrorCovariance(initialEstimateErrorCovariance);
    }
  }

  template<typename T>
  void RosFilter<T>::odometryCallback(const nav_msgs::Odometry::ConstPtr &msg, const std::string &topicName,
    const CallbackData &poseCallbackData, const CallbackData &twistCallbackData)
  {
    // If we've just reset the filter, then we want to ignore any messages
    // that arrive with an older timestamp
    if (msg->header.stamp <= lastSetPoseTime_)
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp equal to or before the last filter reset, " <<
                "this message will be ignored. This may indicate an empty or bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);
      RF_DEBUG("Received message that preceded the most recent pose reset. Ignoring...");
      return;
    }

    RF_DEBUG("------ RosFilter::odometryCallback (" << topicName << ") ------\n" << "Odometry message:\n" << *msg);

    if (poseCallbackData.updateSum_ > 0)
    {
      // Grab the pose portion of the message and pass it to the poseCallback
      geometry_msgs::PoseWithCovarianceStamped *posPtr = new geometry_msgs::PoseWithCovarianceStamped();
      posPtr->header = msg->header;
      posPtr->pose = msg->pose;  // Entire pose object, also copies covariance

      geometry_msgs::PoseWithCovarianceStampedConstPtr pptr(posPtr);
      poseCallback(pptr, poseCallbackData, worldFrameId_, false);
    }

    if (twistCallbackData.updateSum_ > 0)
    {
      // Grab the twist portion of the message and pass it to the twistCallback
      geometry_msgs::TwistWithCovarianceStamped *twistPtr = new geometry_msgs::TwistWithCovarianceStamped();
      twistPtr->header = msg->header;
      twistPtr->header.frame_id = msg->child_frame_id;
      twistPtr->twist = msg->twist;  // Entire twist object, also copies covariance

      geometry_msgs::TwistWithCovarianceStampedConstPtr tptr(twistPtr);
      twistCallback(tptr, twistCallbackData, baseLinkFrameId_);
    }

    RF_DEBUG("\n----- /RosFilter::odometryCallback (" << topicName << ") ------\n");
  }

  template<typename T>
  void RosFilter<T>::poseCallback(const geometry_msgs::PoseWithCovarianceStamped::ConstPtr &msg,
                                  const CallbackData &callbackData,
                                  const std::string &targetFrame,
                                  const bool imuData)
  {
    const std::string &topicName = callbackData.topicName_;

    // If we've just reset the filter, then we want to ignore any messages
    // that arrive with an older timestamp
    if (msg->header.stamp <= lastSetPoseTime_)
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp equal to or before the last filter reset, " <<
                "this message will be ignored. This may indicate an empty or bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);
      return;
    }

    RF_DEBUG("------ RosFilter::poseCallback (" << topicName << ") ------\n" <<
             "Pose message:\n" << *msg);

    // Put the initial value in the lastMessagTimes_ for this variable if it's empty
    if (lastMessageTimes_.count(topicName) == 0)
    {
      lastMessageTimes_.insert(std::pair<std::string, ros::Time>(topicName, msg->header.stamp));
    }

    // Make sure this message is newer than the last one
    if (msg->header.stamp >= lastMessageTimes_[topicName])
    {
      RF_DEBUG("Update vector for " << topicName << " is:\n" << callbackData.updateVector_);

      Eigen::VectorXd measurement(STATE_SIZE);
      Eigen::MatrixXd measurementCovariance(STATE_SIZE, STATE_SIZE);

      measurement.setZero();
      measurementCovariance.setZero();

      // Make sure we're actually updating at least one of these variables
      std::vector<int> updateVectorCorrected = callbackData.updateVector_;

      // Prepare the pose data for inclusion in the filter
      if (preparePose(msg,
                      topicName,
                      targetFrame,
                      callbackData.differential_,
                      callbackData.relative_,
                      imuData,
                      updateVectorCorrected,
                      measurement,
                      measurementCovariance))
      {
        // Store the measurement. Add a "pose" suffix so we know what kind of measurement
        // we're dealing with when we debug the core filter logic.
        enqueueMeasurement(topicName,
                           measurement,
                           measurementCovariance,
                           updateVectorCorrected,
                           callbackData.rejectionThreshold_,
                           msg->header.stamp);

        RF_DEBUG("Enqueued new measurement for " << topicName << "\n");
      }
      else
      {
        RF_DEBUG("Did *not* enqueue measurement for " << topicName << "\n");
      }

      lastMessageTimes_[topicName] = msg->header.stamp;

      RF_DEBUG("Last message time for " << topicName << " is now " <<
        lastMessageTimes_[topicName] << "\n");
    }
    else if (resetOnTimeJump_ && ros::Time::isSimTime())
    {
      reset();
    }
    else
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp before that of the previous message received," <<
                " this message will be ignored. This may indicate a bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);

      RF_DEBUG("Message is too old. Last message time for " << topicName << " is "
               << lastMessageTimes_[topicName] << ", current message time is "
               << msg->header.stamp << ".\n");
    }

    RF_DEBUG("\n----- /RosFilter::poseCallback (" << topicName << ") ------\n");
  }

  template<typename T>
  void RosFilter<T>::run()
  {
    ROS_INFO("Waiting for valid clock time...");
    ros::Time::waitForValid();
    ROS_INFO("Valid clock time received. Starting node.");

    loadParams();

    if (printDiagnostics_)
    {
      diagnosticUpdater_.add("Filter diagnostic updater", this, &RosFilter<T>::aggregateDiagnostics);
    }

    // Set up the frequency diagnostic
    double minFrequency = frequency_ - 2;
    double maxFrequency = frequency_ + 2;
    diagnostic_updater::HeaderlessTopicDiagnostic freqDiag("odometry/filtered",
                                                           diagnosticUpdater_,
                                                           diagnostic_updater::FrequencyStatusParam(&minFrequency,
                                                                                                    &maxFrequency,
                                                                                                    0.1, 10));

    // We may need to broadcast a different transform than
    // the one we've already calculated.
    tf2::Transform mapOdomTrans;
    tf2::Transform baseLinkOdomTrans;
    geometry_msgs::TransformStamped mapOdomTransMsg;
    ros::Time curTime;
    ros::Time lastDiagTime = ros::Time::now();

    // Clear out the transforms
    worldBaseLinkTransMsg_.transform = tf2::toMsg(tf2::Transform::getIdentity());
    mapOdomTransMsg.transform = tf2::toMsg(tf2::Transform::getIdentity());

    // Publisher
    ros::Publisher positionPub = nh_.advertise<nav_msgs::Odometry>("odometry/filtered", 20);
    tf2_ros::TransformBroadcaster worldTransformBroadcaster;

    // Optional acceleration publisher
    ros::Publisher accelPub;
    if (publishAcceleration_)
    {
      accelPub = nh_.advertise<geometry_msgs::AccelWithCovarianceStamped>("accel/filtered", 20);
    }

    const ros::Duration loop_cycle_time(1.0 / frequency_);
    ros::Time loop_end_time = ros::Time::now();

    // Wait for the filter to be enabled
    while (!enabled_ && ros::ok())
    {
      ROS_WARN_STREAM_ONCE("[" << ros::this_node::getName() <<
        ":] This filter is disabled. To enable it call the service " << ros::this_node::getName() << "/enable");
      ros::spinOnce();
      if (enabled_)
      {
        break;
      }
    }

    while (ros::ok())
    {
      // The spin will call all the available callbacks and enqueue
      // their received measurements
      ros::spinOnce();
      curTime = ros::Time::now();

      if (toggledOn_)
      {
        // Now we'll integrate any measurements we've received if requested
        integrateMeasurements(curTime);
      }
      else
      {
        // clear out measurements since we're not currently processing new entries
        clearMeasurementQueue();

        // Reset last measurement time so we don't get a large time delta on toggle on
        if (filter_.getInitializedStatus())
        {
          filter_.setLastMeasurementTime(ros::Time::now().toSec());
        }
      }

      // Get latest state and publish it
      nav_msgs::Odometry filteredPosition;

      if (getFilteredOdometryMessage(filteredPosition))
      {
        worldBaseLinkTransMsg_.header.stamp = filteredPosition.header.stamp + tfTimeOffset_;
        worldBaseLinkTransMsg_.header.frame_id = filteredPosition.header.frame_id;
        worldBaseLinkTransMsg_.child_frame_id = filteredPosition.child_frame_id;

        worldBaseLinkTransMsg_.transform.translation.x = filteredPosition.pose.pose.position.x;
        worldBaseLinkTransMsg_.transform.translation.y = filteredPosition.pose.pose.position.y;
        worldBaseLinkTransMsg_.transform.translation.z = filteredPosition.pose.pose.position.z;
        worldBaseLinkTransMsg_.transform.rotation = filteredPosition.pose.pose.orientation;

        // the filteredPosition is the message containing the state and covariances: nav_msgs Odometry

        if (!validateFilterOutput(filteredPosition))
        {
          ROS_ERROR_STREAM("Critical Error, NaNs were detected in the output state of the filter." <<
                " This was likely due to poorly coniditioned process, noise, or sensor covariances.");
        }

        // If the worldFrameId_ is the odomFrameId_ frame, then we can just send the transform. If the
        // worldFrameId_ is the mapFrameId_ frame, we'll have some work to do.
        if (publishTransform_)
        {
          if (filteredPosition.header.frame_id == odomFrameId_)
          {
            worldTransformBroadcaster.sendTransform(worldBaseLinkTransMsg_);
          }
          else if (filteredPosition.header.frame_id == mapFrameId_)
          {
            try
            {
              tf2::Transform worldBaseLinkTrans;
              tf2::fromMsg(worldBaseLinkTransMsg_.transform, worldBaseLinkTrans);

              if (!RosFilterUtilities::lookupTransformSafe(
                     tfBuffer_,
                     baseLinkFrameId_,
                     odomFrameId_,
                     ros::Time(filter_.getLastMeasurementTime()),
                     baseLinkOdomTrans,
                     true))
              {
                ROS_ERROR_STREAM_DELAYED_THROTTLE(1.0, "Unable to retrieve " << odomFrameId_ << "->" <<
                  baseLinkFrameId_ << " transform. Skipping iteration...");
                continue;
              }

              /*
               * First, see these two references:
               * http://wiki.ros.org/tf/Overview/Using%20Published%20Transforms#lookupTransform
               * http://wiki.ros.org/geometry/CoordinateFrameConventions#Transform_Direction
               * We have a transform from mapFrameId_->baseLinkFrameId_, but it would actually transform
               * a given pose from baseLinkFrameId_->mapFrameId_. We then used lookupTransform, whose
               * first two arguments are target frame and source frame, to get a transform from
               * baseLinkFrameId_->odomFrameId_. However, this transform would actually transform data
               * from odomFrameId_->baseLinkFrameId_. Now imagine that we have a position in the
               * mapFrameId_ frame. First, we multiply it by the inverse of the
               * mapFrameId_->baseLinkFrameId, which will transform that data from mapFrameId_ to
               * baseLinkFrameId_. Now we want to go from baseLinkFrameId_->odomFrameId_, but the
               * transform we have takes data from odomFrameId_->baseLinkFrameId_, so we need its
               * inverse as well. We have now transformed our data from mapFrameId_ to odomFrameId_.
               * However, if we want other users to be able to do the same, we need to broadcast
               * the inverse of that entire transform.
              */

              mapOdomTrans.mult(worldBaseLinkTrans, baseLinkOdomTrans);

              mapOdomTransMsg.transform = tf2::toMsg(mapOdomTrans);
              mapOdomTransMsg.header.stamp = filteredPosition.header.stamp + tfTimeOffset_;
              mapOdomTransMsg.header.frame_id = mapFrameId_;
              mapOdomTransMsg.child_frame_id = odomFrameId_;

              worldTransformBroadcaster.sendTransform(mapOdomTransMsg);
            }
            catch(...)
            {
              ROS_ERROR_STREAM_DELAYED_THROTTLE(5.0, "Could not obtain transform from "
                                                << odomFrameId_ << "->" << baseLinkFrameId_);
            }
          }
          else
          {
            ROS_ERROR_STREAM("Odometry message frame_id was " << filteredPosition.header.frame_id <<
                             ", expected " << mapFrameId_ << " or " << odomFrameId_);
          }
        }

        // Fire off the position and the transform
        positionPub.publish(filteredPosition);

        if (printDiagnostics_)
        {
          freqDiag.tick();
        }
      }

      // Publish the acceleration if desired and filter is initialized
      geometry_msgs::AccelWithCovarianceStamped filteredAcceleration;
      if (publishAcceleration_ && getFilteredAccelMessage(filteredAcceleration))
      {
        accelPub.publish(filteredAcceleration);
      }

      /* Diagnostics can behave strangely when playing back from bag
       * files and using simulated time, so we have to check for
       * time suddenly moving backwards as well as the standard
       * timeout criterion before publishing. */
      double diagDuration = (curTime - lastDiagTime).toSec();
      if (printDiagnostics_ && (diagDuration >= diagnosticUpdater_.getPeriod() || diagDuration < 0.0))
      {
        diagnosticUpdater_.force_update();
        lastDiagTime = curTime;
      }

      // Clear out expired history data
      if (smoothLaggedData_)
      {
        clearExpiredHistory(filter_.getLastMeasurementTime() - historyLength_);
      }

      ros::Duration loop_elapsed_time = ros::Time::now() - loop_end_time;

      if (loop_elapsed_time > loop_cycle_time)
      {
        ROS_WARN_STREAM_DELAYED_THROTTLE(1.0, "Failed to meet update rate! Took " << std::setprecision(20) <<
          loop_elapsed_time.toSec() << " seconds. Try decreasing the rate, limiting sensor output frequency, or "
          "limiting the number of sensors.");
      }
      else
      {
        ros::Duration sleep_time = loop_cycle_time - loop_elapsed_time;
        sleep_time.sleep();
      }

      loop_end_time = ros::Time::now();
    }
  }

  template<typename T>
  void RosFilter<T>::setPoseCallback(const geometry_msgs::PoseWithCovarianceStamped::ConstPtr &msg)
  {
    RF_DEBUG("------ RosFilter::setPoseCallback ------\nPose message:\n" << *msg);

    ROS_INFO_STREAM("Received set_pose request with value\n" << *msg);

    std::string topicName("setPose");

    // Get rid of any initial poses (pretend we've never had a measurement)
    initialMeasurements_.clear();
    previousMeasurements_.clear();
    previousMeasurementCovariances_.clear();

    clearMeasurementQueue();

    filterStateHistory_.clear();
    measurementHistory_.clear();

    // Also set the last set pose time, so we ignore all messages
    // that occur before it
    lastSetPoseTime_ = msg->header.stamp;

    // Set the state vector to the reported pose
    Eigen::VectorXd measurement(STATE_SIZE);
    Eigen::MatrixXd measurementCovariance(STATE_SIZE, STATE_SIZE);
    std::vector<int> updateVector(STATE_SIZE, true);

    // We only measure pose variables, so initialize the vector to 0
    measurement.setZero();

    // Set this to the identity and let the message reset it
    measurementCovariance.setIdentity();
    measurementCovariance *= 1e-6;

    // Prepare the pose data (really just using this to transform it into the target frame).
    // Twist data is going to get zeroed out.
    preparePose(msg, topicName, worldFrameId_, false, false, false, updateVector, measurement, measurementCovariance);

    // For the state
    filter_.setState(measurement);
    filter_.setEstimateErrorCovariance(measurementCovariance);

    filter_.setLastMeasurementTime(ros::Time::now().toSec());

    RF_DEBUG("\n------ /RosFilter::setPoseCallback ------\n");
  }

  template<typename T>
  bool RosFilter<T>::setPoseSrvCallback(robot_localization::SetPose::Request& request,
                          robot_localization::SetPose::Response&)
  {
    geometry_msgs::PoseWithCovarianceStamped::Ptr msg;
    msg = boost::make_shared<geometry_msgs::PoseWithCovarianceStamped>(request.pose);
    setPoseCallback(msg);

    return true;
  }

  template<typename T>
  bool RosFilter<T>::enableFilterSrvCallback(std_srvs::Empty::Request&,
                                             std_srvs::Empty::Response&)
  {
    RF_DEBUG("\n[" << ros::this_node::getName() << ":]" << " ------ /RosFilter::enableFilterSrvCallback ------\n");
    if (enabled_)
    {
      ROS_WARN_STREAM("[" << ros::this_node::getName() <<
        ":] Asking for enabling filter service, but the filter was already enabled! Use param disabled_at_startup.");
    }
    else
    {
      ROS_INFO_STREAM("[" << ros::this_node::getName() << ":] Enabling filter...");
      enabled_ = true;
    }
    return true;
  }

  template<typename T>
  void RosFilter<T>::twistCallback(const geometry_msgs::TwistWithCovarianceStamped::ConstPtr &msg,
                                   const CallbackData &callbackData,
                                   const std::string &targetFrame)
  {
    const std::string &topicName = callbackData.topicName_;

    // If we've just reset the filter, then we want to ignore any messages
    // that arrive with an older timestamp
    if (msg->header.stamp <= lastSetPoseTime_)
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp equal to or before the last filter reset, " <<
                "this message will be ignored. This may indicate an empty or bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);
      return;
    }

    RF_DEBUG("------ RosFilter::twistCallback (" << topicName << ") ------\n"
             "Twist message:\n" << *msg);

    if (lastMessageTimes_.count(topicName) == 0)
    {
      lastMessageTimes_.insert(std::pair<std::string, ros::Time>(topicName, msg->header.stamp));
    }

    // Make sure this message is newer than the last one
    if (msg->header.stamp >= lastMessageTimes_[topicName])
    {
      RF_DEBUG("Update vector for " << topicName << " is:\n" << callbackData.updateVector_);

      Eigen::VectorXd measurement(STATE_SIZE);
      Eigen::MatrixXd measurementCovariance(STATE_SIZE, STATE_SIZE);

      measurement.setZero();
      measurementCovariance.setZero();

      // Make sure we're actually updating at least one of these variables
      std::vector<int> updateVectorCorrected = callbackData.updateVector_;

      // Prepare the twist data for inclusion in the filter
      if (prepareTwist(msg, topicName, targetFrame, updateVectorCorrected, measurement, measurementCovariance))
      {
        // Store the measurement. Add a "twist" suffix so we know what kind of measurement
        // we're dealing with when we debug the core filter logic.
        enqueueMeasurement(topicName,
                           measurement,
                           measurementCovariance,
                           updateVectorCorrected,
                           callbackData.rejectionThreshold_,
                           msg->header.stamp);

        RF_DEBUG("Enqueued new measurement for " << topicName << "_twist\n");
      }
      else
      {
        RF_DEBUG("Did *not* enqueue measurement for " << topicName << "_twist\n");
      }

      lastMessageTimes_[topicName] = msg->header.stamp;

      RF_DEBUG("Last message time for " << topicName << " is now " <<
        lastMessageTimes_[topicName] << "\n");
    }
    else if (resetOnTimeJump_ && ros::Time::isSimTime())
    {
      reset();
    }
    else
    {
      std::stringstream stream;
      stream << "The " << topicName << " message has a timestamp before that of the previous message received," <<
                " this message will be ignored. This may indicate a bad timestamp. (message time: " <<
                msg->header.stamp.toSec() << ")";
      addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                    topicName + "_timestamp",
                    stream.str(),
                    false);

      RF_DEBUG("Message is too old. Last message time for " << topicName << " is " << lastMessageTimes_[topicName] <<
        ", current message time is " << msg->header.stamp << ".\n");
    }

    RF_DEBUG("\n----- /RosFilter::twistCallback (" << topicName << ") ------\n");
  }

  template<typename T>
  void RosFilter<T>::addDiagnostic(const int errLevel,
                                   const std::string &topicAndClass,
                                   const std::string &message,
                                   const bool staticDiag)
  {
    if (staticDiag)
    {
      staticDiagnostics_[topicAndClass] = message;
      staticDiagErrorLevel_ = std::max(staticDiagErrorLevel_, errLevel);
    }
    else
    {
      dynamicDiagnostics_[topicAndClass] = message;
      dynamicDiagErrorLevel_ = std::max(dynamicDiagErrorLevel_, errLevel);
    }
  }

  template<typename T>
  void RosFilter<T>::aggregateDiagnostics(diagnostic_updater::DiagnosticStatusWrapper &wrapper)
  {
    wrapper.clear();
    wrapper.clearSummary();

    int maxErrLevel = std::max(staticDiagErrorLevel_, dynamicDiagErrorLevel_);

    // Report the overall status
    switch (maxErrLevel)
    {
      case diagnostic_msgs::DiagnosticStatus::ERROR:
        wrapper.summary(maxErrLevel,
                        "Erroneous data or settings detected for a robot_localization state estimation node.");
        break;
      case diagnostic_msgs::DiagnosticStatus::WARN:
        wrapper.summary(maxErrLevel,
                        "Potentially erroneous data or settings detected for "
                        "a robot_localization state estimation node.");
        break;
      case diagnostic_msgs::DiagnosticStatus::STALE:
        wrapper.summary(maxErrLevel,
                        "The state of the robot_localization state estimation node is stale.");
        break;
      case diagnostic_msgs::DiagnosticStatus::OK:
        wrapper.summary(maxErrLevel,
                        "The robot_localization state estimation node appears to be functioning properly.");
        break;
      default:
        break;
    }

    // Aggregate all the static messages
    for (std::map<std::string, std::string>::iterator diagIt = staticDiagnostics_.begin();
        diagIt != staticDiagnostics_.end();
        ++diagIt)
    {
      wrapper.add(diagIt->first, diagIt->second);
    }

    // Aggregate all the dynamic messages, then clear them
    for (std::map<std::string, std::string>::iterator diagIt = dynamicDiagnostics_.begin();
        diagIt != dynamicDiagnostics_.end();
        ++diagIt)
    {
      wrapper.add(diagIt->first, diagIt->second);
    }
    dynamicDiagnostics_.clear();

    // Reset the warning level for the dynamic diagnostic messages
    dynamicDiagErrorLevel_ = diagnostic_msgs::DiagnosticStatus::OK;
  }

  template<typename T>
  void RosFilter<T>::copyCovariance(const double *arr,
                                    Eigen::MatrixXd &covariance,
                                    const std::string &topicName,
                                    const std::vector<int> &updateVector,
                                    const size_t offset,
                                    const size_t dimension)
  {
    for (size_t i = 0; i < dimension; i++)
    {
      for (size_t j = 0; j < dimension; j++)
      {
        covariance(i, j) = arr[dimension * i + j];

        if (printDiagnostics_)
        {
          std::string iVar = stateVariableNames_[offset + i];

          if (covariance(i, j) > 1e3 && (updateVector[offset  + i] || updateVector[offset  + j]))
          {
            std::string jVar = stateVariableNames_[offset + j];

            std::stringstream stream;
            stream << "The covariance at position (" << dimension * i + j << "), which corresponds to " <<
                (i == j ? iVar + " variance" : iVar + " and " + jVar + " covariance") <<
                ", the value is extremely large (" << covariance(i, j) << "), but the update vector for " <<
                (i == j ? iVar : iVar + " and/or " + jVar) << " is set to true. This may produce undesirable results.";

            addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                          topicName + "_covariance",
                          stream.str(),
                          false);
          }
          else if (updateVector[i] && i == j && covariance(i, j) == 0)
          {
            std::stringstream stream;
            stream << "The covariance at position (" << dimension * i + j << "), which corresponds to " <<
                       iVar << " variance, was zero. This will be replaced with a small value to maintain filter "
                       "stability, but should be corrected at the message origin node.";

            addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                          topicName + "_covariance",
                          stream.str(),
                          false);
          }
          else if (updateVector[i] && i == j && covariance(i, j) < 0)
          {
            std::stringstream stream;
            stream << "The covariance at position (" << dimension * i + j <<
                      "), which corresponds to " << iVar << " variance, was negative. This will be replaced with a "
                      "small positive value to maintain filter stability, but should be corrected at the message "
                      "origin node.";

            addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                          topicName + "_covariance",
                          stream.str(),
                          false);
          }
        }
      }
    }
  }

  template<typename T>
  void RosFilter<T>::copyCovariance(const Eigen::MatrixXd &covariance,
                                 double *arr,
                                 const size_t dimension)
  {
    for (size_t i = 0; i < dimension; i++)
    {
      for (size_t j = 0; j < dimension; j++)
      {
        arr[dimension * i + j] = covariance(i, j);
      }
    }
  }

  template<typename T>
  std::vector<int> RosFilter<T>::loadUpdateConfig(const std::string &topicName)
  {
    XmlRpc::XmlRpcValue topicConfig;
    std::vector<int> updateVector(STATE_SIZE, 0);
    std::string topicConfigName = topicName + "_config";

    try
    {
      nhLocal_.getParam(topicConfigName, topicConfig);

      ROS_ASSERT(topicConfig.getType() == XmlRpc::XmlRpcValue::TypeArray);

      if (topicConfig.size() != STATE_SIZE)
      {
        ROS_WARN_STREAM("Configuration vector for " << topicConfigName << " should have 15 entries.");
      }

      for (int i = 0; i < topicConfig.size(); i++)
      {
        // The double cast looks strange, but we'll get exceptions if we
        // remove the cast to bool. vector<bool> is discouraged, so updateVector
        // uses integers.
        updateVector[i] = static_cast<int>(static_cast<bool>(topicConfig[i]));
      }
    }
    catch (XmlRpc::XmlRpcException &e)
    {
      ROS_FATAL_STREAM("Could not read sensor update configuration for topic " << topicName <<
                       " (type: " << topicConfig.getType() << ", expected: " << XmlRpc::XmlRpcValue::TypeArray
                       << "). Error was " << e.getMessage() << "\n");
    }

    return updateVector;
  }

  template<typename T>
  bool RosFilter<T>::prepareAcceleration(const sensor_msgs::Imu::ConstPtr &msg,
                           const std::string &topicName,
                           const std::string &targetFrame,
                           std::vector<int> &updateVector,
                           Eigen::VectorXd &measurement,
                           Eigen::MatrixXd &measurementCovariance)
  {
    RF_DEBUG("------ RosFilter::prepareAcceleration (" << topicName << ") ------\n");

    // 1. Get the measurement into a vector
    tf2::Vector3 accTmp(msg->linear_acceleration.x,
                        msg->linear_acceleration.y,
                        msg->linear_acceleration.z);

    // Set relevant header info
    std::string msgFrame = (msg->header.frame_id == "" ? baseLinkFrameId_ : msg->header.frame_id);

    // 2. robot_localization lets users configure which variables from the sensor should be
    //    fused with the filter. This is specified at the sensor level. However, the data
    //    may go through transforms before being fused with the state estimate. In that case,
    //    we need to know which of the transformed variables came from the pre-transformed
    //    "approved" variables (i.e., the ones that had "true" in their xxx_config parameter).
    //    To do this, we create a pose from the original upate vector, which contains only
    //    zeros and ones. This pose goes through the same transforms as the measurement. The
    //    non-zero values that result will be used to modify the updateVector.
    tf2::Matrix3x3 maskAcc(updateVector[StateMemberAx], 0, 0,
                           0, updateVector[StateMemberAy], 0,
                           0, 0, updateVector[StateMemberAz]);

    // 3. We'll need to rotate the covariance as well
    Eigen::MatrixXd covarianceRotated(ACCELERATION_SIZE, ACCELERATION_SIZE);
    covarianceRotated.setZero();

    this->copyCovariance(&(msg->linear_acceleration_covariance[0]),
                         covarianceRotated,
                         topicName,
                         updateVector,
                         POSITION_A_OFFSET,
                         ACCELERATION_SIZE);

    RF_DEBUG("Original measurement as tf object: " << accTmp <<
             "\nOriginal update vector:\n" << updateVector <<
             "\nOriginal covariance matrix:\n" << covarianceRotated << "\n");

    // 4. We need to transform this into the target frame (probably base_link)
    // It's unlikely that we'll get a velocity measurement in another frame, but
    // we have to handle the situation.
    tf2::Transform targetFrameTrans;
    bool canTransform = RosFilterUtilities::lookupTransformSafe(tfBuffer_,
                                                                targetFrame,
                                                                msgFrame,
                                                                msg->header.stamp,
                                                                tfTimeout_,
                                                                targetFrameTrans);

    if (canTransform)
    {
      // We don't know if the user has already handled the removal
      // of normal forces, so we use a parameter
      if (removeGravitationalAcc_[topicName])
      {
        tf2::Vector3 normAcc(0, 0, gravitationalAcc_);
        tf2::Quaternion curAttitude;
        tf2::Transform trans;

        if (::fabs(msg->orientation_covariance[0] + 1) < 1e-9)
        {
          // Imu message contains no orientation, so we should use orientation
          // from filter state to transform and remove acceleration
          const Eigen::VectorXd &state = filter_.getState();
          tf2::Vector3 stateTmp(state(StateMemberRoll),
                                state(StateMemberPitch),
                                state(StateMemberYaw));
          // transform state orientation to IMU frame
          tf2::Transform imuFrameTrans;
          RosFilterUtilities::lookupTransformSafe(tfBuffer_,
                                                  msgFrame,
                                                  targetFrame,
                                                  msg->header.stamp,
                                                  tfTimeout_,
                                                  imuFrameTrans);
          stateTmp = imuFrameTrans.getBasis() * stateTmp;
          curAttitude.setRPY(stateTmp.getX(), stateTmp.getY(), stateTmp.getZ());
        }
        else
        {
          tf2::fromMsg(msg->orientation, curAttitude);
          if (fabs(curAttitude.length() - 1.0) > 0.01)
          {
            ROS_WARN_ONCE("An input was not normalized, this should NOT happen, but will normalize.");
            curAttitude.normalize();
          }
        }
        trans.setRotation(curAttitude);
        tf2::Vector3 rotNorm = trans.getBasis().inverse() * normAcc;
        accTmp.setX(accTmp.getX() - rotNorm.getX());
        accTmp.setY(accTmp.getY() - rotNorm.getY());
        accTmp.setZ(accTmp.getZ() - rotNorm.getZ());

        RF_DEBUG("Orientation is " << curAttitude <<
                 "Acceleration due to gravity is " << rotNorm <<
                 "After removing acceleration due to gravity, acceleration is " << accTmp << "\n");
      }

      // Transform to correct frame
      // @todo: This needs to take into account offsets from the origin. Right now,
      // it assumes that if the sensor is placed at some non-zero offset from the
      // vehicle's center, that the vehicle turns with constant velocity. This needs
      // to be something like
      // accTmp = targetFrameTrans.getBasis() * accTmp - targetFrameTrans.getOrigin().cross(rotation_acceleration);
      // We can get rotational acceleration by differentiating the rotational velocity
      // (if it's available)
      accTmp = targetFrameTrans.getBasis() * accTmp;
      maskAcc = targetFrameTrans.getBasis() * maskAcc;

      // Now use the mask values to determine which update vector values should be true
      updateVector[StateMemberAx] = static_cast<int>(
        maskAcc.getRow(StateMemberAx - POSITION_A_OFFSET).length() >= 1e-6);
      updateVector[StateMemberAy] = static_cast<int>(
        maskAcc.getRow(StateMemberAy - POSITION_A_OFFSET).length() >= 1e-6);
      updateVector[StateMemberAz] = static_cast<int>(
        maskAcc.getRow(StateMemberAz - POSITION_A_OFFSET).length() >= 1e-6);

      RF_DEBUG(msg->header.frame_id << "->" << targetFrame << " transform:\n" << targetFrameTrans <<
               "\nAfter applying transform to " << targetFrame << ", update vector is:\n" << updateVector <<
               "\nAfter applying transform to " << targetFrame << ", measurement is:\n" << accTmp << "\n");

      // 5. Now rotate the covariance: create an augmented
      // matrix that contains a 3D rotation matrix in the
      // upper-left and lower-right quadrants, and zeros
      // elsewhere
      tf2::Matrix3x3 rot(targetFrameTrans.getRotation());
      Eigen::MatrixXd rot3d(ACCELERATION_SIZE, ACCELERATION_SIZE);
      rot3d.setIdentity();

      for (size_t rInd = 0; rInd < ACCELERATION_SIZE; ++rInd)
      {
        rot3d(rInd, 0) = rot.getRow(rInd).getX();
        rot3d(rInd, 1) = rot.getRow(rInd).getY();
        rot3d(rInd, 2) = rot.getRow(rInd).getZ();
      }

      // Carry out the rotation
      covarianceRotated = rot3d * covarianceRotated.eval() * rot3d.transpose();

      RF_DEBUG("Transformed covariance is \n" << covarianceRotated << "\n");

      // 6. Store our corrected measurement and covariance
      measurement(StateMemberAx) = accTmp.getX();
      measurement(StateMemberAy) = accTmp.getY();
      measurement(StateMemberAz) = accTmp.getZ();

      // Copy the covariances
      measurementCovariance.block(POSITION_A_OFFSET, POSITION_A_OFFSET, ACCELERATION_SIZE, ACCELERATION_SIZE) =
        covarianceRotated.block(0, 0, ACCELERATION_SIZE, ACCELERATION_SIZE);

      // 7. Handle 2D mode
      if (twoDMode_)
      {
        forceTwoD(measurement, measurementCovariance, updateVector);
      }
    }
    else
    {
      RF_DEBUG("Could not transform measurement into " << targetFrame << ". Ignoring...\n");
    }

    RF_DEBUG("\n----- /RosFilter::prepareAcceleration(" << topicName << ") ------\n");

    return canTransform;
  }

  template<typename T>
  bool RosFilter<T>::preparePose(const geometry_msgs::PoseWithCovarianceStamped::ConstPtr &msg,
                                 const std::string &topicName,
                                 const std::string &targetFrame,
                                 const bool differential,
                                 const bool relative,
                                 const bool imuData,
                                 std::vector<int> &updateVector,
                                 Eigen::VectorXd &measurement,
                                 Eigen::MatrixXd &measurementCovariance)
  {
    bool retVal = false;

    RF_DEBUG("------ RosFilter::preparePose (" << topicName << ") ------\n");

    // 1. Get the measurement into a tf-friendly transform (pose) object
    tf2::Stamped<tf2::Transform> poseTmp;

    // We'll need this later for storing this measurement for differential integration
    tf2::Transform curMeasurement;

    // Handle issues where frame_id data is not filled out properly
    // @todo: verify that this is necessary still. New IMU handling may
    // have rendered this obsolete.
    std::string finalTargetFrame;
    if (targetFrame == "" && msg->header.frame_id == "")
    {
      // Blank target and message frames mean we can just
      // use our world_frame
      finalTargetFrame = worldFrameId_;
      poseTmp.frame_id_ = finalTargetFrame;
    }
    else if (targetFrame == "")
    {
      // A blank target frame means we shouldn't bother
      // transforming the data
      finalTargetFrame = msg->header.frame_id;
      poseTmp.frame_id_ = finalTargetFrame;
    }
    else
    {
      // Otherwise, we should use our target frame
      finalTargetFrame = targetFrame;
      poseTmp.frame_id_ = (differential ? finalTargetFrame : msg->header.frame_id);
    }

    RF_DEBUG("Final target frame for " << topicName << " is " << finalTargetFrame << "\n");

    poseTmp.stamp_ = msg->header.stamp;

    // Fill out the position data
    poseTmp.setOrigin(tf2::Vector3(msg->pose.pose.position.x,
                                   msg->pose.pose.position.y,
                                   msg->pose.pose.position.z));

    tf2::Quaternion orientation;

    // Handle bad (empty) quaternions
    if (msg->pose.pose.orientation.x == 0 && msg->pose.pose.orientation.y == 0 &&
       msg->pose.pose.orientation.z == 0 && msg->pose.pose.orientation.w == 0)
    {
      orientation.setValue(0.0, 0.0, 0.0, 1.0);

      if (updateVector[StateMemberRoll] || updateVector[StateMemberPitch] || updateVector[StateMemberYaw])
      {
        std::stringstream stream;
        stream << "The " << topicName << " message contains an invalid orientation quaternion, " <<
                  "but its configuration is such that orientation data is being used. Correcting...";

        addDiagnostic(diagnostic_msgs::DiagnosticStatus::WARN,
                      topicName + "_orientation",
                      stream.str(),
                      false);
      }
    }
    else
    {
      tf2::fromMsg(msg->pose.pose.orientation, orientation);
      if (fabs(orientation.length() - 1.0) > 0.01)
      {
        ROS_WARN_ONCE("An input was not normalized, this should NOT happen, but will normalize.");
        orientation.normalize();
      }
    }

    // Fill out the orientation data
    poseTmp.setRotation(orientation);

    // 2. Get the target frame transformation
    tf2::Transform targetFrameTrans;
    bool canTransform = RosFilterUtilities::lookupTransformSafe(tfBuffer_,
                                                                finalTargetFrame,
                                                                poseTmp.frame_id_,
                                                                poseTmp.stamp_,
                                                                tfTimeout_,
                                                                targetFrameTrans);

    // 3. Make sure we can work with this data before carrying on
    if (canTransform)
    {
      /* 4. robot_localization lets users configure which variables from the sensor should be
       *    fused with the filter. This is specified at the sensor level. However, the data
       *    may go through transforms before being fused with the state estimate. In that case,
       *    we need to know which of the transformed variables came from the pre-transformed
       *    "approved" variables (i.e., the ones that had "true" in their xxx_config parameter).
       *    To do this, we construct matrices using the update vector values on the diagonals,
       *    pass this matrix through the rotation, and use the length of each row to determine
       *    the transformed update vector. The process is slightly different for IMUs, as the
       *    coordinate frame transform is really the base_link->imu_frame transform, and not
       *    a transform from some other world-fixed frame (even though the IMU data itself *is*
       *    reported in a world fixed frame). */
      tf2::Matrix3x3 maskPosition(updateVector[StateMemberX], 0, 0,
                                  0, updateVector[StateMemberY], 0,
                                  0, 0, updateVector[StateMemberZ]);

      tf2::Matrix3x3 maskOrientation(updateVector[StateMemberRoll], 0, 0,
                                     0, updateVector[StateMemberPitch], 0,
                                     0, 0, updateVector[StateMemberYaw]);

      if (imuData)
      {
        /* We have to treat IMU orientation data differently. Even though we are dealing with pose
         * data when we work with orientations, for IMUs, the frame_id is the frame in which the
         * sensor is mounted, and not the coordinate frame of the IMU. Imagine an IMU that is mounted
         * facing sideways. The pitch in the IMU frame becomes roll for the vehicle. This means that
         * we need to rotate roll and pitch angles by the IMU's mounting yaw offset, and we must apply
         * similar treatment to its update mask and covariance. */

        double dummy, yaw;
        targetFrameTrans.getBasis().getRPY(dummy, dummy, yaw);
        tf2::Matrix3x3 transTmp;
        transTmp.setRPY(0.0, 0.0, yaw);

        maskPosition = transTmp * maskPosition;
        maskOrientation = transTmp * maskOrientation;
      }
      else
      {
        maskPosition = targetFrameTrans.getBasis() * maskPosition;
        maskOrientation = targetFrameTrans.getBasis() * maskOrientation;
      }

      // Now copy the mask values back into the update vector: any row with a significant vector length
      // indicates that we want to set that variable to true in the update vector.
      updateVector[StateMemberX] = static_cast<int>(
        maskPosition.getRow(StateMemberX - POSITION_OFFSET).length() >= 1e-6);
      updateVector[StateMemberY] = static_cast<int>(
        maskPosition.getRow(StateMemberY - POSITION_OFFSET).length() >= 1e-6);
      updateVector[StateMemberZ] = static_cast<int>(
        maskPosition.getRow(StateMemberZ - POSITION_OFFSET).length() >= 1e-6);
      updateVector[StateMemberRoll] = static_cast<int>(
        maskOrientation.getRow(StateMemberRoll - ORIENTATION_OFFSET).length() >= 1e-6);
      updateVector[StateMemberPitch] = static_cast<int>(
        maskOrientation.getRow(StateMemberPitch - ORIENTATION_OFFSET).length() >= 1e-6);
      updateVector[StateMemberYaw] = static_cast<int>(
        maskOrientation.getRow(StateMemberYaw - ORIENTATION_OFFSET).length() >= 1e-6);

      // 5a. We'll need to rotate the covariance as well. Create a container and copy over the
      // covariance data
      Eigen::MatrixXd covariance(POSE_SIZE, POSE_SIZE);
      covariance.setZero();
      copyCovariance(&(msg->pose.covariance[0]), covariance, topicName, updateVector, POSITION_OFFSET, POSE_SIZE);

      // 5b. Now rotate the covariance: create an augmented matrix that
      // contains a 3D rotation matrix in the upper-left and lower-right
      // quadrants, with zeros elsewhere.
      tf2::Matrix3x3 rot;
      Eigen::MatrixXd rot6d(POSE_SIZE, POSE_SIZE);
      rot6d.setIdentity();
      Eigen::MatrixXd covarianceRotated;

      if (imuData)
      {
        // Apply the same special logic to the IMU covariance rotation
        double dummy, yaw;
        targetFrameTrans.getBasis().getRPY(dummy, dummy, yaw);
        rot.setRPY(0.0, 0.0, yaw);
      }
      else
      {
        rot.setRotation(targetFrameTrans.getRotation());
      }

      for (size_t rInd = 0; rInd < POSITION_SIZE; ++rInd)
      {
        rot6d(rInd, 0) = rot.getRow(rInd).getX();
        rot6d(rInd, 1) = rot.getRow(rInd).getY();
        rot6d(rInd, 2) = rot.getRow(rInd).getZ();
        rot6d(rInd+POSITION_SIZE, 3) = rot.getRow(rInd).getX();
        rot6d(rInd+POSITION_SIZE, 4) = rot.getRow(rInd).getY();
        rot6d(rInd+POSITION_SIZE, 5) = rot.getRow(rInd).getZ();
      }

      // Now carry out the rotation
      covarianceRotated = rot6d * covariance * rot6d.transpose();

      RF_DEBUG("After rotating into the " << finalTargetFrame <<
               " frame, covariance is \n" << covarianceRotated <<  "\n");

      /* 6a. For IMU data, the transform that we get is the transform from the body
       * frame of the robot (e.g., base_link) to the mounting frame of the robot. It
       * is *not* the coordinate frame in which the IMU orientation data is reported.
       * If the IMU is mounted in a non-neutral orientation, we need to remove those
       * offsets, and then we need to potentially "swap" roll and pitch.
       * Note that this transform does NOT handle NED->ENU conversions. Data is assumed
       * to be in the ENU frame when it is received.
       * */
      if (imuData)
      {
        // First, convert the transform and measurement rotation to RPY
        // @todo: There must be a way to handle this with quaternions. Need to look into it.
        double rollOffset = 0;
        double pitchOffset = 0;
        double yawOffset = 0;
        double roll = 0;
        double pitch = 0;
        double yaw = 0;
        RosFilterUtilities::quatToRPY(targetFrameTrans.getRotation(), rollOffset, pitchOffset, yawOffset);
        RosFilterUtilities::quatToRPY(poseTmp.getRotation(), roll, pitch, yaw);

        // 6b. Apply the offset (making sure to bound them), and throw them in a vector
        tf2::Vector3 rpyAngles(FilterUtilities::clampRotation(roll - rollOffset),
                               FilterUtilities::clampRotation(pitch - pitchOffset),
                               FilterUtilities::clampRotation(yaw - yawOffset));

        // 6c. Now we need to rotate the roll and pitch by the yaw offset value.
        // Imagine a case where an IMU is mounted facing sideways. In that case
        // pitch for the IMU's world frame is roll for the robot.
        tf2::Matrix3x3 mat;
        mat.setRPY(0.0, 0.0, yawOffset);
        rpyAngles = mat * rpyAngles;
        poseTmp.getBasis().setRPY(rpyAngles.getX(), rpyAngles.getY(), rpyAngles.getZ());

        // We will use this target transformation later on, but
        // we've already transformed this data as if the IMU
        // were mounted neutrall on the robot, so we can just
        // make the transform the identity.
        targetFrameTrans.setIdentity();
      }

      // 7. Two cases: if we're in differential mode, we need to generate a twist
      // message. Otherwise, we just transform it to the target frame.
      if (differential)
      {
        bool success = false;

        // We're going to be playing with poseTmp, so store it,
        // as we'll need to save its current value for the next
        // measurement.
        curMeasurement = poseTmp;

        // Make sure we have previous measurements to work with
        if (previousMeasurements_.count(topicName) > 0 && previousMeasurementCovariances_.count(topicName) > 0)
        {
          // 7a. If we are carrying out differential integration and
          // we have a previous measurement for this sensor,then we
          // need to apply the inverse of that measurement to this new
          // measurement to produce a "delta" measurement between the two.
          // Even if we're not using all of the variables from this sensor,
          // we need to use the whole measurement to determine the delta
          // to the new measurement
          tf2::Transform prevMeasurement = previousMeasurements_[topicName];
          poseTmp.setData(prevMeasurement.inverseTimes(poseTmp));

          RF_DEBUG("Previous measurement:\n" << previousMeasurements_[topicName] <<
                   "\nAfter removing previous measurement, measurement delta is:\n" << poseTmp << "\n");

          // 7b. Now we we have a measurement delta in the frame_id of the
          // message, but we want that delta to be in the target frame, so
          // we need to apply the rotation of the target frame transform.
          targetFrameTrans.setOrigin(tf2::Vector3(0.0, 0.0, 0.0));
          poseTmp.mult(targetFrameTrans, poseTmp);

          RF_DEBUG("After rotating to the target frame, measurement delta is:\n" << poseTmp << "\n");

          // 7c. Now use the time difference from the last message to compute
          // translational and rotational velocities
          double dt = msg->header.stamp.toSec() - lastMessageTimes_[topicName].toSec();
          double xVel = poseTmp.getOrigin().getX() / dt;
          double yVel = poseTmp.getOrigin().getY() / dt;
          double zVel = poseTmp.getOrigin().getZ() / dt;

          double rollVel = 0;
          double pitchVel = 0;
          double yawVel = 0;

          RosFilterUtilities::quatToRPY(poseTmp.getRotation(), rollVel, pitchVel, yawVel);
          rollVel /= dt;
          pitchVel /= dt;
          yawVel /= dt;

          RF_DEBUG("Previous message time was " << lastMessageTimes_[topicName].toSec() <<
                   ", current message time is " << msg->header.stamp.toSec() << ", delta is " <<
                   dt << ", velocity is (vX, vY, vZ): (" << xVel << ", " << yVel << ", " << zVel <<
                   ")\n" << "(vRoll, vPitch, vYaw): (" << rollVel << ", " << pitchVel << ", " <<
                   yawVel << ")\n");

          // 7d. Fill out the velocity data in the message
          geometry_msgs::TwistWithCovarianceStamped *twistPtr = new geometry_msgs::TwistWithCovarianceStamped();
          twistPtr->header = msg->header;
          twistPtr->header.frame_id = baseLinkFrameId_;
          twistPtr->twist.twist.linear.x = xVel;
          twistPtr->twist.twist.linear.y = yVel;
          twistPtr->twist.twist.linear.z = zVel;
          twistPtr->twist.twist.angular.x = rollVel;
          twistPtr->twist.twist.angular.y = pitchVel;
          twistPtr->twist.twist.angular.z = yawVel;
          std::vector<int> twistUpdateVec(STATE_SIZE, false);
          std::copy(updateVector.begin() + POSITION_OFFSET,
                    updateVector.begin() + POSE_SIZE,
                    twistUpdateVec.begin() + POSITION_V_OFFSET);
          std::copy(twistUpdateVec.begin(), twistUpdateVec.end(), updateVector.begin());
          geometry_msgs::TwistWithCovarianceStampedConstPtr ptr(twistPtr);

          // 7e. Now rotate the previous covariance for this measurement to get it
          // into the target frame, and add the current measurement's rotated covariance
          // to the previous measurement's rotated covariance, and multiply by the time delta.
          Eigen::MatrixXd prevCovarRotated = rot6d * previousMeasurementCovariances_[topicName] * rot6d.transpose();
          covarianceRotated = (covarianceRotated.eval() + prevCovarRotated) * dt;
          copyCovariance(covarianceRotated, &(twistPtr->twist.covariance[0]), POSE_SIZE);

          RF_DEBUG("Previous measurement covariance:\n" << previousMeasurementCovariances_[topicName] <<
                   "\nPrevious measurement covariance rotated:\n" << prevCovarRotated <<
                   "\nFinal twist covariance:\n" << covarianceRotated << "\n");

          // Now pass this on to prepareTwist, which will convert it to the required frame
          success = prepareTwist(ptr,
                                 topicName + "_twist",
                                 twistPtr->header.frame_id,
                                 updateVector,
                                 measurement,
                                 measurementCovariance);
        }

        // 7f. Update the previous measurement and measurement covariance
        previousMeasurements_[topicName] = curMeasurement;
        previousMeasurementCovariances_[topicName] = covariance;

        retVal = success;
      }
      else
      {
        // 7g. If we're in relative mode, remove the initial measurement
        if (relative)
        {
          if (initialMeasurements_.count(topicName) == 0)
          {
            initialMeasurements_.insert(std::pair<std::string, tf2::Transform>(topicName, poseTmp));
          }

          tf2::Transform initialMeasurement = initialMeasurements_[topicName];
          poseTmp.setData(initialMeasurement.inverseTimes(poseTmp));
        }

        // 7h. Apply the target frame transformation to the pose object.
        poseTmp.mult(targetFrameTrans, poseTmp);
        poseTmp.frame_id_ = finalTargetFrame;

        // 7i. Finally, copy everything into our measurement and covariance objects
        measurement(StateMemberX) = poseTmp.getOrigin().x();
        measurement(StateMemberY) = poseTmp.getOrigin().y();
        measurement(StateMemberZ) = poseTmp.getOrigin().z();

        // The filter needs roll, pitch, and yaw values instead of quaternions
        double roll, pitch, yaw;
        RosFilterUtilities::quatToRPY(poseTmp.getRotation(), roll, pitch, yaw);
        measurement(StateMemberRoll) = roll;
        measurement(StateMemberPitch) = pitch;
        measurement(StateMemberYaw) = yaw;

        measurementCovariance.block(0, 0, POSE_SIZE, POSE_SIZE) = covarianceRotated.block(0, 0, POSE_SIZE, POSE_SIZE);

        // 8. Handle 2D mode
        if (twoDMode_)
        {
          forceTwoD(measurement, measurementCovariance, updateVector);
        }

        retVal = true;
      }
    }
    else
    {
      retVal = false;

      RF_DEBUG("Could not transform measurement into " << finalTargetFrame << ". Ignoring...");
    }

    RF_DEBUG("\n----- /RosFilter::preparePose (" << topicName << ") ------\n");

    return retVal;
  }

  template<typename T>
  bool RosFilter<T>::prepareTwist(const geometry_msgs::TwistWithCovarianceStamped::ConstPtr &msg,
                               const std::string &topicName,
                               const std::string &targetFrame,
                               std::vector<int> &updateVector,
                               Eigen::VectorXd &measurement,
                               Eigen::MatrixXd &measurementCovariance)
  {
    RF_DEBUG("------ RosFilter::prepareTwist (" << topicName << ") ------\n");

    // 1. Get the measurement into two separate vector objects.
    tf2::Vector3 twistLin(msg->twist.twist.linear.x,
                          msg->twist.twist.linear.y,
                          msg->twist.twist.linear.z);
    tf2::Vector3 measTwistRot(msg->twist.twist.angular.x,
                              msg->twist.twist.angular.y,
                              msg->twist.twist.angular.z);

    // 1a. This sensor may or may not measure rotational velocity. Regardless,
    // if it measures linear velocity, then later on, we'll need to remove "false"
    // linear velocity resulting from angular velocity and the translational offset
    // of the sensor from the vehicle origin.
    const Eigen::VectorXd &state = filter_.getState();
    tf2::Vector3 stateTwistRot(state(StateMemberVroll),
                               state(StateMemberVpitch),
                               state(StateMemberVyaw));

    // Determine the frame_id of the data
    std::string msgFrame = (msg->header.frame_id == "" ? targetFrame : msg->header.frame_id);

    // 2. robot_localization lets users configure which variables from the sensor should be
    //    fused with the filter. This is specified at the sensor level. However, the data
    //    may go through transforms before being fused with the state estimate. In that case,
    //    we need to know which of the transformed variables came from the pre-transformed
    //    "approved" variables (i.e., the ones that had "true" in their xxx_config parameter).
    //    To do this, we construct matrices using the update vector values on the diagonals,
    //    pass this matrix through the rotation, and use the length of each row to determine
    //    the transformed update vector.
    tf2::Matrix3x3 maskLin(updateVector[StateMemberVx], 0, 0,
                           0, updateVector[StateMemberVy], 0,
                           0, 0, updateVector[StateMemberVz]);

    tf2::Matrix3x3 maskRot(updateVector[StateMemberVroll], 0, 0,
                           0, updateVector[StateMemberVpitch], 0,
                           0, 0, updateVector[StateMemberVyaw]);

    // 3. We'll need to rotate the covariance as well
    Eigen::MatrixXd covarianceRotated(TWIST_SIZE, TWIST_SIZE);
    covarianceRotated.setZero();

    copyCovariance(&(msg->twist.covariance[0]),
                   covarianceRotated,
                   topicName,
                   updateVector,
                   POSITION_V_OFFSET,
                   TWIST_SIZE);

    RF_DEBUG("Original measurement as tf object:\nLinear: " << twistLin <<
             "Rotational: " << measTwistRot <<
             "\nOriginal update vector:\n" << updateVector <<
             "\nOriginal covariance matrix:\n" << covarianceRotated << "\n");

    // 4. We need to transform this into the target frame (probably base_link)
    tf2::Transform targetFrameTrans;
    bool canTransform = RosFilterUtilities::lookupTransformSafe(tfBuffer_,
                                                                targetFrame,
                                                                msgFrame,
                                                                msg->header.stamp,
                                                                tfTimeout_,
                                                                targetFrameTrans);

    if (canTransform)
    {
      // Transform to correct frame. Note that we can get linear velocity
      // as a result of the sensor offset and rotational velocity
      measTwistRot = targetFrameTrans.getBasis() * measTwistRot;
      twistLin = targetFrameTrans.getBasis() * twistLin + targetFrameTrans.getOrigin().cross(stateTwistRot);
      maskLin = targetFrameTrans.getBasis() * maskLin;
      maskRot = targetFrameTrans.getBasis() * maskRot;

      // Now copy the mask values back into the update vector
      updateVector[StateMemberVx] = static_cast<int>(
        maskLin.getRow(StateMemberVx - POSITION_V_OFFSET).length() >= 1e-6);
      updateVector[StateMemberVy] = static_cast<int>(
        maskLin.getRow(StateMemberVy - POSITION_V_OFFSET).length() >= 1e-6);
      updateVector[StateMemberVz] = static_cast<int>(
        maskLin.getRow(StateMemberVz - POSITION_V_OFFSET).length() >= 1e-6);
      updateVector[StateMemberVroll] = static_cast<int>(
        maskRot.getRow(StateMemberVroll - ORIENTATION_V_OFFSET).length() >= 1e-6);
      updateVector[StateMemberVpitch] = static_cast<int>(
        maskRot.getRow(StateMemberVpitch - ORIENTATION_V_OFFSET).length() >= 1e-6);
      updateVector[StateMemberVyaw] = static_cast<int>(
        maskRot.getRow(StateMemberVyaw - ORIENTATION_V_OFFSET).length() >= 1e-6);

      RF_DEBUG(msg->header.frame_id << "->" << targetFrame << " transform:\n" << targetFrameTrans <<
               "\nAfter applying transform to " << targetFrame << ", update vector is:\n" << updateVector <<
               "\nAfter applying transform to " << targetFrame << ", measurement is:\n" <<
               "Linear: " << twistLin << "Rotational: " << measTwistRot << "\n");

      // 5. Now rotate the covariance: create an augmented
      // matrix that contains a 3D rotation matrix in the
      // upper-left and lower-right quadrants, and zeros
      // elsewhere
      tf2::Matrix3x3 rot(targetFrameTrans.getRotation());
      Eigen::MatrixXd rot6d(TWIST_SIZE, TWIST_SIZE);
      rot6d.setIdentity();

      for (size_t rInd = 0; rInd < POSITION_SIZE; ++rInd)
      {
        rot6d(rInd, 0) = rot.getRow(rInd).getX();
        rot6d(rInd, 1) = rot.getRow(rInd).getY();
        rot6d(rInd, 2) = rot.getRow(rInd).getZ();
        rot6d(rInd+POSITION_SIZE, 3) = rot.getRow(rInd).getX();
        rot6d(rInd+POSITION_SIZE, 4) = rot.getRow(rInd).getY();
        rot6d(rInd+POSITION_SIZE, 5) = rot.getRow(rInd).getZ();
      }

      // Carry out the rotation
      covarianceRotated = rot6d * covarianceRotated.eval() * rot6d.transpose();

      RF_DEBUG("Transformed covariance is \n" << covarianceRotated << "\n");

      // 6. Store our corrected measurement and covariance
      measurement(StateMemberVx) = twistLin.getX();
      measurement(StateMemberVy) = twistLin.getY();
      measurement(StateMemberVz) = twistLin.getZ();
      measurement(StateMemberVroll) = measTwistRot.getX();
      measurement(StateMemberVpitch) = measTwistRot.getY();
      measurement(StateMemberVyaw) = measTwistRot.getZ();

      // Copy the covariances
      measurementCovariance.block(POSITION_V_OFFSET, POSITION_V_OFFSET, TWIST_SIZE, TWIST_SIZE) =
        covarianceRotated.block(0, 0, TWIST_SIZE, TWIST_SIZE);

      // 7. Handle 2D mode
      if (twoDMode_)
      {
        forceTwoD(measurement, measurementCovariance, updateVector);
      }
    }
    else
    {
      RF_DEBUG("Could not transform measurement into " << targetFrame << ". Ignoring...");
    }

    RF_DEBUG("\n----- /RosFilter::prepareTwist (" << topicName << ") ------\n");

    return canTransform;
  }

  template<typename T>
  void RosFilter<T>::saveFilterState(FilterBase& filter)
  {
    FilterStatePtr state = FilterStatePtr(new FilterState());
    state->state_ = Eigen::VectorXd(filter.getState());
    state->estimateErrorCovariance_ = Eigen::MatrixXd(filter.getEstimateErrorCovariance());
    state->lastMeasurementTime_ = filter.getLastMeasurementTime();
    state->latestControl_ = Eigen::VectorXd(filter.getControl());
    state->latestControlTime_ = filter.getControlTime();
    filterStateHistory_.push_back(state);
    RF_DEBUG("Saved state with timestamp " << std::setprecision(20) << state->lastMeasurementTime_ <<
             " to history. " << filterStateHistory_.size() << " measurements are in the queue.\n");
  }

  template<typename T>
  bool RosFilter<T>::revertTo(const double time)
  {
    RF_DEBUG("\n----- RosFilter::revertTo -----\n");
    RF_DEBUG("\nRequested time was " << std::setprecision(20) << time << "\n")

    size_t history_size = filterStateHistory_.size();

    // Walk back through the queue until we reach a filter state whose time stamp is less than or equal to the
    // requested time. Since every saved state after that time will be overwritten/corrected, we can pop from
    // the queue. If the history is insufficiently short, we just take the oldest state we have.
    FilterStatePtr lastHistoryState;
    while (!filterStateHistory_.empty() && filterStateHistory_.back()->lastMeasurementTime_ > time)
    {
      lastHistoryState = filterStateHistory_.back();
      filterStateHistory_.pop_back();
    }

    // If the state history is not empty at this point, it means that our history was large enough, and we
    // should revert to the state at the back of the history deque.
    bool retVal = false;
    if (!filterStateHistory_.empty())
    {
      retVal = true;
      lastHistoryState = filterStateHistory_.back();
    }
    else
    {
      RF_DEBUG("Insufficient history to revert to time " << time << "\n");

      if (lastHistoryState)
      {
        RF_DEBUG("Will revert to oldest state at " << lastHistoryState->latestControlTime_ << ".\n");
        ROS_WARN_STREAM_DELAYED_THROTTLE(historyLength_, "Could not revert to state with time " <<
          std::setprecision(20) << time << ". Instead reverted to state with time " <<
          lastHistoryState->lastMeasurementTime_ << ". History size was " << history_size);
      }
    }

    // If we have a valid reversion state, revert
    if (lastHistoryState)
    {
      // Reset filter to the latest state from the queue.
      const FilterStatePtr &state = lastHistoryState;
      filter_.setState(state->state_);
      filter_.setEstimateErrorCovariance(state->estimateErrorCovariance_);
      filter_.setLastMeasurementTime(state->lastMeasurementTime_);

      RF_DEBUG("Reverted to state with time " << std::setprecision(20) << state->lastMeasurementTime_ << "\n");

      // Repeat for measurements, but push every measurement onto the measurement queue as we go
      int restored_measurements = 0;
      while (!measurementHistory_.empty() && measurementHistory_.back()->time_ > time)
      {
        // Don't need to restore measurements that predate our earliest state time
        if (state->lastMeasurementTime_ <= measurementHistory_.back()->time_)
        {
          measurementQueue_.push(measurementHistory_.back());
          restored_measurements++;
        }

        measurementHistory_.pop_back();
      }

      RF_DEBUG("Restored " << restored_measurements << " to measurement queue.\n");
    }

    RF_DEBUG("\n----- /RosFilter::revertTo\n");

    return retVal;
  }

  template<typename T>
  bool RosFilter<T>::validateFilterOutput(const nav_msgs::Odometry &message)
  {
    return !std::isnan(message.pose.pose.position.x) && !std::isinf(message.pose.pose.position.x) &&
           !std::isnan(message.pose.pose.position.y) && !std::isinf(message.pose.pose.position.y) &&
           !std::isnan(message.pose.pose.position.z) && !std::isinf(message.pose.pose.position.z) &&
           !std::isnan(message.pose.pose.orientation.x) && !std::isinf(message.pose.pose.orientation.x) &&
           !std::isnan(message.pose.pose.orientation.y) && !std::isinf(message.pose.pose.orientation.y) &&
           !std::isnan(message.pose.pose.orientation.z) && !std::isinf(message.pose.pose.orientation.z) &&
           !std::isnan(message.pose.pose.orientation.w) && !std::isinf(message.pose.pose.orientation.w) &&
           !std::isnan(message.twist.twist.linear.x) && !std::isinf(message.twist.twist.linear.x) &&
           !std::isnan(message.twist.twist.linear.y) && !std::isinf(message.twist.twist.linear.y) &&
           !std::isnan(message.twist.twist.linear.z) && !std::isinf(message.twist.twist.linear.z) &&
           !std::isnan(message.twist.twist.angular.x) && !std::isinf(message.twist.twist.angular.x) &&
           !std::isnan(message.twist.twist.angular.y) && !std::isinf(message.twist.twist.angular.y) &&
           !std::isnan(message.twist.twist.angular.z) && !std::isinf(message.twist.twist.angular.z);
  }

  template<typename T>
  void RosFilter<T>::clearExpiredHistory(const double cutOffTime)
  {
    RF_DEBUG("\n----- RosFilter::clearExpiredHistory -----" <<
             "\nCutoff time is " << cutOffTime << "\n");

    int poppedMeasurements = 0;
    int poppedStates = 0;

    while (!measurementHistory_.empty() && measurementHistory_.front()->time_ < cutOffTime)
    {
      measurementHistory_.pop_front();
      poppedMeasurements++;
    }

    while (!filterStateHistory_.empty() && filterStateHistory_.front()->lastMeasurementTime_ < cutOffTime)
    {
      filterStateHistory_.pop_front();
      poppedStates++;
    }

    RF_DEBUG("\nPopped " << poppedMeasurements << " measurements and " <<
             poppedStates << " states from their respective queues." <<
             "\n---- /RosFilter::clearExpiredHistory ----\n");
  }

  template<typename T>
  void RosFilter<T>::clearMeasurementQueue()
  {
    while (!measurementQueue_.empty() && ros::ok())
    {
      measurementQueue_.pop();
    }
    return;
  }
}  // namespace RobotLocalization

// Instantiations of classes is required when template class code
// is placed in a .cpp file.
template class RobotLocalization::RosFilter<RobotLocalization::Ekf>;
template class RobotLocalization::RosFilter<RobotLocalization::Ukf>;