carla_cil_pytorch_eval

A pytorch implementation to evaluate the conditional imitation learning policy in "End-to-end Driving via Conditional Imitation Learning" and "CARLA: An Open Urban Driving Simulator".

Requirements

pytorch > 0.4.0
tensorboardX

Running

Start carla simulater and leave your trained policy weight in model/policy.pth run:

$ python run_CIL.py --log-name local_test --weathers 6 --model-path "model/policy.pth"

Policy Training

Please reference carla_cil_pytorh.
For the benchmark results, please check our RA-L paper VR-Goggles for Robots: Real-to-sim Domain Adaptation for Visual Control.