linear_feedback_controller
- roscontrol controller package conputing a linear feedback. The user needs
to provide a model of the robot and a list of controlled joint and the controller computes a linear feedback on the user defined state.
README
linear-feedback-controller
A ros2_control package implementing linear feedback control strategies for robotic systems. It supports both fixed-base and free-flying robot configurations and is designed to integrate with external model predictive controllers.
The public message interface is defined in linear_feedback_controller_msgs.
Both packages are released under the BSD-2 license. Current versions are tracked in their respective package.xml files and must have an associated git tag of the form vX.Y.Z.
Full documentation is available on DeepWiki.
Controllers
The package provides three ros2_control controller plugins:
Controller |
Base class |
Role |
|---|---|---|
|
|
Main control loop: PD → linear feedback transition with external MPC |
|
|
Joint state filtering and velocity estimation |
|
|
Interface renaming / forwarding for hardware–simulation parity |
LinearFeedbackControllerRos
The main controller implements a two-phase control strategy:
PD phase — Proportional-derivative control for initial stabilization using configurable per-joint gains (
p_gains,d_gains).LF phase — Linear feedback control driven by an external controller (e.g. a whole-body MPC based on Crocoddyl).
The transition between phases is time-based and configurable via pd_to_lf_transition_duration.
The controller publishes sensor data on a ROS 2 topic:
Base pose and velocity (free-flying robots; identity for fixed-base robots)
Joint positions, velocities, and efforts
It receives control commands containing:
A feedback gain matrix
A feedforward torque term
The linearization state
JointStateEstimator
Reads hardware state interfaces, applies exponential smoothing to joint velocities, and writes the filtered values to command interfaces. This makes raw sensor data suitable for downstream chainable controllers.
PassthroughController
Forwards and renames reference interfaces to command interfaces. Useful when simulation and hardware use different interface names but share the same parameter file.
Dependencies
Package |
Role |
|---|---|
Controller framework and hardware abstraction |
|
Rigid body dynamics and URDF model loading |
|
Linear algebra |
|
ROS topic message definitions and Eigen/NumPy conversions |
Build
With colcon (recommended)
mkdir -p workspace/src
git -C workspace/src clone https://github.com/loco-3d/linear-feedback-controller.git
colcon build
With CMake directly
git clone https://github.com/loco-3d/linear-feedback-controller.git
cd linear-feedback-controller
mkdir _build && cd _build
cmake .. -DCMAKE_INSTALL_PREFIX=/path/to/your/install/folder
make install
With Nix
git clone https://github.com/loco-3d/linear-feedback-controller.git
cd linear-feedback-controller
# Check (build + tests)
nix flake check -L
# Build only
nix build -L
# Enter development shell
nix develop
nix-direnv is also supported for automatic shell activation.
Configuration
An example YAML configuration for the Tiago Pro robot is available at config/tiago_pro_lfc_params.yaml.
Real-world usage can be found in the agimus-demos repository, in particular in the agimus_demos_common package.