IntelLabs / coach

Reinforcement Learning Coach by Intel AI Lab enables easy experimentation with state of the art Reinforcement Learning algorithms
https://intellabs.github.io/coach/
Apache License 2.0
2.33k stars 461 forks source link

Problem adding a driving benchmark to an experimental suit in CARLA_DDPG #443

Open Jidenna opened 4 years ago

Jidenna commented 4 years ago

Great work here! So I am trying to use my own Carla driving_benchmark with a specific town, pose and vehicle/pedestrian parameters or at least use the basic_experiment_suit found in Carla's experiment_suit directory. I understand to do this, I will need to add the following code to the evn_params in CARLA_DDPG.py:

env_params = CarlaEnvironmentParameters()

env_params.cameras = ['CameraRGB'] env_params.camera_height = 600 env_params.camera_width = 800 env_params.experiment_suite = BasicExperimentSuite('Town01')

Also I had to make some changes to agent_params. Specifically changing the Front_camera to CameraRGB. Here:

agent_params = DDPGAgentParameters()

agent_params.algorithm.num_consecutive_playing_steps = EnvironmentSteps(4) agent_params.network_wrappers['actor'].input_embedders_parameters['CameraRGB'] = \ agent_params.network_wrappers['actor'].input_embedders_parameters.pop('observation') agent_params.network_wrappers['critic'].input_embedders_parameters['CameraRGB'] = \ agent_params.network_wrappers['critic'].input_embedders_parameters.pop('observation')

The problem, the agent runs for 1 or 2 episodes and then returns a broken pipe error each time it starts to improve the simple_rl_graph. error_brokenpipe

What am I doing wrong? Please help