AutoCastSim builds on top of Carla Scenario Runner the networking component among vehicles as well as the occlusion scenarios that single vehicle based solution often fails. To config the scenarios, we added a lot of Autocast specific flags in addition to original flags.
To view the explanation of all flags, simple run
python3 scenario_runner.py --help
In the following, we list a few important flags for our considerations. These include all the major flags to be configured in run_test.sh and run_eval.sh.
Simulation engine and setup flags:
--host HOST IP of the host server (default: localhost)
--port PORT TCP port to listen to (default: 2000)
--mqttport Set the MQTT broker port (which should be different than port)
--trafficManagerPort Port to use for the TrafficManager (default: 8000)
--trafficManagerSeed Random seed used by the TrafficManager to config background traffic (default: 0)
--emulate Flag to switch between emulation (slower, higher fidelity, e.g. for communication, object detection) vs simulation (faster, use carla labels)
Scenario flags:
--route Run a route as a scenario (input: (route_file,scenario_file,[number of route]))
An example route file: "srunner/data/routes_training_town03_autocast10.xml"
A corresponding scenario file: "srunner/data/towns03_traffic_scenarios_autocast10.json"
--reloadWorld Reload the CARLA world before starting a scenario (default=True)
--bgtraffic Set the amount of background traffic in scenario
--seed SEED random seed used for background traffic
--eval Evaluation parameters (input: (ego_speed, ego_distance, collider_speed, collider_dist,[collider_acceleration_dist]))
--nocollider Disable colliders in scenarios
Recording flags:
--full Record full records for data logger (Lidar, camera, from all vehicles nearby)
--lean Record lean records (only ego BEV and Lidar) for data logger
--hud Record the HUD display
--commlog Record communication logs among vehicles
--outputdir Set the recordings output directory
Agent flags
--agent Agent used to execute the scenario. e.g. "AVR/autocast_agents/simple_agent.py"
--sharing Enable sensor sharing
--passive_collider Disable colliders sharing in scenarios
--fullpc Enable Full Point cloud sharing with 1000X bandwidth (Caution: this runs very slow)