Fast Peer Adaptation with Context-aware Exploration

Installation

Please see install.sh for instructions. Note that you may need to copy each line into the terminal and execute them separately instead of directly running the script itself.

Running Experiments

See scripts/ for the training scripts for each environment.

Code Structure Overview

• learning/: Implementation of the context buffer and training algorithms.
• environment/: Environment implementations.
• scripts/: Training scripts.
• baselines/: The GSCU baseline. Other baselines are implemented on the same codebase as our method.
• train_.py: Main training entry point.
• online_test_.py: Evaluation code of trained checkpoints.

Acknowledgements

This repository is built upon the following projects. We sincerely thank their authors for the contributions; for references, please refer to the paper.

• PPO: https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail
• GSCU: https://github.com/YeTianJHU/GSCU
• Overcooked: https://github.com/51616/marl-lipo
• MPE: https://github.com/openai/multiagent-particle-envs