STOMP Yaml file

Basic STOMP planner configuration

A basic configuration file is as follows:

• stomp_config_example.yaml

  stomp/manipulator:
    group_name: manipulator
    optimization:
      num_timesteps: 20
      num_iterations: 50
      num_iterations_after_valid: 0    
      num_rollouts: 10
      max_rollouts: 100
      initialization_method: 1 #[1 : LINEAR_INTERPOLATION, 2 : CUBIC_POLYNOMIAL, 3 : MININUM_CONTROL_COST
      control_cost_weight: 0.0
    task:
      noise_generator:
        - class: stomp_moveit/NormalDistributionSampling
          stddev: [0.3, 1.0, 0.5, 0.4, 0.3, 0.3]
      cost_functions:
        - class: stomp_moveit/CollisionCheck
          collision_penalty: 1.0
          cost_weight: 1.0 
          kernel_window_percentage: 0.4
      noisy_filters:
        - class: stomp_moveit/JointLimits
          lock_start: True
          lock_goal: True
      update_filters:
        - class: stomp_moveit/ControlCostProjectionMatrix
  stomp/manipulator_rail:
    group_name: manipulator_rail
    optimization:
      num_timesteps: 40
      num_iterations: 40
      num_iterations_after_valid: 0    
      num_rollouts: 10
      max_rollouts: 10 
      initialization_method: 3 #[1 : LINEAR_INTERPOLATION, 2 : CUBIC_POLYNOMIAL, 3 : MININUM_CONTROL_COST
      control_cost_weight: 0.0
    task:
      noise_generator:
        - class: stomp_moveit/NormalDistributionSampling
          stddev: [0.05, 0.4, 1.2, 0.4, 0.4, 0.1, 0.1]
      cost_functions:
        - class: stomp_moveit/CollisionCheck
          collision_penalty: 1.0
          cost_weight: 1.0
          kernel_window_percentage: 0.2
          longest_valid_joint_move: 0.05 
      noisy_filters:
        - class: stomp_moveit/JointLimits
          lock_start: True
          lock_goal: True
        - class: stomp_moveit/MultiTrajectoryVisualization
          line_width: 0.02
          rgb: [255, 255, 0]
          marker_array_topic: stomp_trajectories
          marker_namespace: noisy
      update_filters:
        - class: stomp_moveit/ControlCostProjectionMatrix
        - class: stomp_moveit/UpdateLogger
          package: stomp_moveit
          directory: log
          filename: smoothed_update.txt
  

  Each "stomp/group_name" section is specific to that planning group. The group name must be a valid planning group in the SRDF file

Explanation of parameters

Optimization Parameters

All the parameters under the optimization field affect the behavior of the optimization process. These parameters are as follows:

• num_timesteps: The number of timesteps in the trajectory.
• num_iterations: How many times to iterate in order to find a valid trajectory. If the allowed number of iterations is reached before a solution is found then the trajectory is invalid.
• num_iterations_after_valid: How many times to iterate after success in order to further improve the trajectory.
• num_rollouts: Number of new noisy trajectories to generate at each iteration.
• max_rollouts: How many rollouts STOMP will store from previous iterations. When the maximum is reached, old rollouts with higher costs will be discarded in place of rollouts with less cost.
• initialization_method: An integer value indicated the method for creating the initial trajectory.
  • Linear Interpolation (1): An interpolation from start to end.
  • Cubic Polynomial(2): Uses a cubic spline.
  • Minimum Control Cost(3): Builds a covariance matrix and uses it to generate an initial trajectory with low accelerations.
• control_cost_weight: Weighting factor applied to the acceleration costs, using zero is recommended.

Tasks Parameters

At each iteration, STOMP invokes a StompTaks object. The taks object holds all of the active plugins and invokes them at specific stages of the optimization process. Thus each of the plugins is listed under a plugin specifier entry below the "task" field. Each plugin entry should at least have a "class" field which is used by pluginlib to identify and load the plugin. There are four types of plugins that can be loaded into the StompTask and these are:

• noise_generator: Adds random noise onto the trajectory in order to explore the workspace. Only one can be loaded
• cost_functions: Evaluate the state costs of each noisy trajectory.
• noisy_filters: Apply various filtering methods to the noisy trajectories.
• update_filters: Apply various filtering methods to the update values that will be used in improving the current trajectory.