MAMPS: Safe Multi-Agent Reinforcement Learning via Model Predictive Shielding

TL;DR

MAMPS is a novel multi-agent reinforcement learning method that guarantees safety through model predictive shielding, allowing learned policies to operate while ensuring obstacle avoidance in robotics tasks.

Contribution

The paper introduces MAMPS, a safety-guaranteeing algorithm that combines learned policies with backup strategies in multi-agent reinforcement learning.

Findings

MAMPS achieves safety guarantees in multi-agent simulations.

The method maintains high performance while ensuring safety.

It effectively prevents unsafe behaviors like obstacle collisions.

Abstract

Reinforcement learning is a promising approach to learning control policies for performing complex multi-agent robotics tasks. However, a policy learned in simulation often fails to guarantee even simple safety properties such as obstacle avoidance. To ensure safety, we propose multi-agent model predictive shielding (MAMPS), an algorithm that provably guarantees safety for an arbitrary learned policy. In particular, it operates by using the learned policy as often as possible, but instead uses a backup policy in cases where it cannot guarantee the safety of the learned policy. Using a multi-agent simulation environment, we show how MAMPS can achieve good performance while ensuring safety.