This repo implements a TRPO agent ( http://arxiv.org/abs/1502.05477 ) by modifying https://github.com/wojzaremba/trpo and replacing the softmax distributions with Gaussian distributions, and adding a tiny bit of bells and whistles.
To run the code, simply type python main.py --task $TASK_NAME. Once training is complete, main.py will upload the run using your OpenAI gym account which should be stored in OPENAI_GYM_API_KEY.