Multi-Agent Reinforcement Learning and Genetic Policy Sharing

TL;DR

This paper investigates how sharing policies among agents in a multi-agent reinforcement learning system affects learning speed and performance, revealing that sharing accelerates convergence and enhances outcomes.

Contribution

It provides the first systematic analysis of policy sharing effects in multi-agent reinforcement learning, highlighting benefits for convergence and performance.

Findings

Policy sharing reduces time to reach asymptotic behavior

Sharing policies improves asymptotic performance

Higher population densities influence sharing effectiveness

Abstract

The effects of policy sharing between agents in a multi-agent dynamical system has not been studied extensively. I simulate a system of agents optimizing the same task using reinforcement learning, to study the effects of different population densities and policy sharing. I demonstrate that sharing policies decreases the time to reach asymptotic behavior, and results in improved asymptotic behavior.