High-level command for RL agent

[weizhaoji]

trafficstars

hi @bitsauce, in your model,the input state of RL agent is the vae output and measurement(steer,acc,break),but not include the high-level command,such as turn left,turn right,go straight,fllow the road.I want to add these high-level command to train RL agent,what should I do?

[bitsauce]

Hi! Sorry for the late response.

I actually have an example of this in my repo. My repo consists of two different Carla environments:

1. The lap environment (where the agent must follow a pre-defined lap)
2. The route environment (where the agent must follow a high-level route, starting at a random point A and ending at a random point B)

The route environment is implemented in CarlaEnv/carla_route_env.py, and you can switch to this environment by setting USE_ROUTE_ENVIRONMENT = True in train.py (or in the .py file you want to run.) The current high-level command expected by the environment is stored in CarlaLapEnv.current_road_maneuver.

Hope this answered your question!

[xiaoqingkong]

hi @bitsauce,how about the file CarlalapEnv.current_road_maneuver? Looking forward to your reply!