/multirl

Reinforcement Learning in Multi-Agent Settings

Primary LanguagePython

multicraft

This is our experiments on two different RL approaches in solving a multi-agent setting, in particular, a Tag scenario. The environment and scenario we used is from the Multi-Agent Particle Environments (MPE).

Dependencies

  • Python (3.5.4)
  • OpenAI gym (0.10.5)
  • Tensorflow (1.8.0)
  • numpy (1.14.5)
  • torch (1.3.1)

Running Experiments

  • under /experiment/ directory, execute train.py: this executes two learning algorithms, SARSA and DDPG respectively, and trains 2000 episodes each. By default, the scenario is rendered every 100 episodes.

  • The motivation for our experiments is to compare the performance of learned agents against that of Multi-Agent Deep Deterministic Policy Gradient (MADDPG) and to validate MADDPG as well.

  • To run our MADDPG experiment (based on https://github.com/openai/maddpg): under /maddpg/experiments/ directory, execute train.py

Results

  • SARSAR and DDPG experiments each produces a result benchmark.csv, stored at /experiment/sarsa_out and /experiment/ddpg_out espectively

  • MADDPG's experiment result (benchmark.csv) is stored at /maddpg/experiments/maddpg_out