Multiagent Reinforcement Learning: Rollout and Policy Iteration

Implementation of the Multiagent Rollout based on the paper by Dimitri Bertsekas (2021).

Environment

Simulation environment follows the rules of the Spiders-and-Flies game as specified in [1]. The environment is adapted from Anurag Koul's ma-gym [2] modifying the PredatorPrey env.

Usage

• Install the requirements with pip:

$ pip install -r requirements.txt

• Run the agent simulation from the scripts folder:

$ python run_agent.py

• Run agents' comparison from the scripts folder:

$ python run_comparison.py

Results

Baseline Policy	Standard Rollout	Agent-by-agent Rollout

Note: Baseline Policy means moving along the shortest path to the closest surviving fly.

References

1. Dimitri Bertsekas - Multiagent Reinforcement Learning: Rollout and Policy Iteration (2021). Web: https://ieeexplore.ieee.org/document/9317713

2. Anurag Koul - ma-gym: Collection of multi-agent environments based on OpenAI gym (2019). Web: https://github.com/koulanurag/ma-gym