Utilities and scripts used to perform experiments described in "Imitating Driver Behavior with Generative Adversarial Networks". Built on rllab and source code for generative adversarial imitation learning.
Train a model from the command line by running:
python scripts/train_gail_model.py
An ego vehicle trained through Generative Adversarial Imitation Learning (blue) navigating a congested highway scene.
Julia 0.5
ForwardNets.jl (nextgen branch)
AutomotiveDrivingModels.jl (gail branch)
Note: This repository is not up to date with recent changes to the following Julia packages. We recommend using the following commits of these packages:
AutoViz.jl (commit 274dd08)
NGSIM.jl (commit f16d684)
Jonathan Ho, Stefano Ermon. "Generative Adversarial Imitation Learning". Advances in Neural Information Processing Systems (NIPS), 2016
Yan Duan, Xi Chen, Rein Houthooft, John Schulman, Pieter Abbeel. "Benchmarking Deep Reinforcement Learning for Continuous Control". Proceedings of the 33rd International Conference on Machine Learning (ICML), 2016.