Inspecting runs in Foxglove

Every CARLA and CommonRoad run can be written to an MCAP file, or streamed live, and opened in Foxglove or the open-source Lichtblick. Scrub through a run frame by frame and see what the planner saw: every candidate with its cost rank, the obstacle that blocked the cheapest candidate, the predicted paths, gap / headway / time to collision to the vehicle ahead, planning times and (in CARLA) the camera.

The recordings of the demo runs are attached to the GitHub release (carla_cut_in.mcap, carla_traffic.mcap, commonroad_zip.mcap).

Open a recording

1. Download a .mcap file and the layout docs/foxglove/fiss_planner_layout.json.
2. Open the file in Foxglove (desktop app or app.foxglove.dev; a free account is required) or Lichtblick.
3. Import the layout (Layouts → Import from file).

The layout shows a 3D view following the ego, the camera, plots of speed vs. target, headway and TTC, planning time and trajectory count, and the raw planner statistics.

Topic	Content
/map/lanes	lane boundaries
/ego	ego box and heading
/obstacles	obstacle boxes and predicted paths; vehicle ahead yellow, blocking obstacle orange
/plan/trajectory	executed plan
/plan/candidates	explored candidates, cheapest brightest; rejected ones faint
/plan/stats	JSON: planner, mode (plan / fallback / degraded / emergency), timings, counts, gap, headway, TTC, rejections
/tf	map → ego
/camera/image	JPEG chase camera (CARLA)

Record your own

pip install "fiss-planner[foxglove,commonroad]"
# CommonRoad scenario
python -m fiss_planner.commonroad.demo ZAM_Zip-1_50_T-1.xml --mcap zip.mcap
# CARLA with the chase camera (start the simulator first)
python -m fiss_planner.carla.record --scenario cut_in --mcap cut_in.mcap
# CARLA batch runs, one file per run
python -m fiss_planner.carla.closed_loop --scenario traffic --mcap recordings/

Add --foxglove-live to the CARLA commands and connect Foxglove to ws://localhost:8765 to watch a run while it drives. For the ROS nodes, record the topics with ros2 bag record -s mcap (see ROS).