Freespace Planner Scaling

[ahmaad-ansari]

Linked to one of my past posts, I was able to configure the config files to allow Autoware to handle my specific scaled model vehicle. I have been able to run simulations of various lanelet configurations with success. I am now having trouble scaling the freespace planner down. I keep getting the error shown in the screenshot:

[component_container-29] [ERROR] [1709447497.150241928] [planning.scenario_planning.motion_velocity_smoother]: input trajectory size must > 1. Skip computation.
[component_container-29] [ERROR] [1709447497.150311920] [planning.scenario_planning.motion_velocity_smoother]: No enough points in trajectory after overlap points removal
[component_container_mt-32] [INFO] [1709447497.301982456] [planning.scenario_planning.parking.freespace_planner]: Can't find goal...

I think that my answer lies in one of the following files goal_planner.param.yaml,freespace_planner.param.yaml, or motion_velocity_smoother.param.yaml. I have done quite some testing to see what values change the simulation but have had no luck. If anyone knows how the freespace planner really works and what values I need to alter please reply.

The following is my autoware launch scenario planning config files: https://github.com/ahmaad-ansari/autoware_launch/tree/main/autoware_launch/config/planning/scenario_planning

[4swinv]

Hello,
I am looking into the issue, the error is probably due to the planner not being able to find the goal either because the launch configs are not proper or some other issue in code, can you give me these details,

1. How can I reproduce the error, are all parameters for your scaled model vehicle included in the autoware lauch repo you shared? If there were changes to vehicle configs please also link them.
2. The second picture shows a planned trajectory to goal, in what exact cases do you get an error?

[ahmaad-ansari]

Thank you for looking into this matter. The following are my autoware launch and vehicle launch repositories:

• https://github.com/ahmaad-ansari/autoware_launch
• https://github.com/ahmaad-ansari/r2_vehicle_launch

The outputs vary every time a path is planned:

• If the path planned has a goal within the freespace planner, it will enter the freespace area and either
  • Plan a trajectory to the goal (like the second photo) and commit to it partially
  • Plan a trajectory to the goal and not commit to it at all (fail)
• If the path planned has a goal not within the freespace planner, the car will be controlled as usual as I have enabled the path planning to be scaled down

[4swinv]

Hello,
Please make sure you have configured the search parameters for astar correctly for your scaled case in freespace_planner.param.yaml

# -- Configurations common to the all planners --
    # base configs
    time_limit: 30000.0
    # robot configs # TODO replace by vehicle_info
    robot_length: 0.235
    robot_width: 0.166
    robot_base2back: 0.045
    minimum_turning_radius: 0.5
    maximum_turning_radius: 0.5
    turning_radius_size: 1
    # search configs
    theta_size: 144
    angle_goal_range: 0.6
    curve_weight: 1.2
    reverse_weight: 2.0
    lateral_goal_range: 5.0
    longitudinal_goal_range: 2.0

Currently the lateral_goal_range: 5.0, longitudinal_goal_range: 2.0 is pretty high compared to your robot size, the algorithm stops planning if the vehicle is within these limits to goal. The astar planner also uses a cost map which you have set to 7m, if you place the goal outside 7m the planner will fail since the search area is confined to this 7m x 7m grid. This is probabably the reason for this:

If the path planned has a goal within the freespace planner, it will enter the freespace area and either
Plan a trajectory to the goal (like the second photo) and commit to it partially
Plan a trajectory to the goal and not commit to it at all (fail)

autoware_launch/config/planning/scenario_planning/common/costmap_generator.param.yaml

    grid_resolution: 0.02
    grid_length_x: 7.0
    grid_length_y: 7.0

You might want to change these parameters to better suit your map.

[maxime-clem]

@ahmaad-ansari I marked the question as answered but feel free to unmark it if needed.