Research code for the paper "Trial without Error: Towards Safe Reinforcement Learning via Human Intervention" (arxiv) (2017)
Contributors (alphabetical): Owain Evans, Vlad Firoiu, Girish Sastry, William Saunders
This repository contains the code for human intervention reinforcement learning in Atari environments (based on OpenAI's Gym). The humanrl package contains various Gym environment wrappers and utilities that allow modifying Atari environments to include catastrophes.
scripts/human_feedback.py is a script that allows a human to intervene during offline or online training of an RL agent.
To label and run the code locally, first create an Anaconda environment with our packages:
conda env create
source activate humanrlSee the human feedback README for directions on providing human feedback with the OpenAI universe starter agent.
See the catastrophe wrapper for a general purpose way to add catastrophes to Gym environments.