Multi-Agent Behavior Retrieval: Retrieval-Augmented Policy Training for Cooperative Push Manipulation by Mobile Robots

2023-12-04

IEEE/RJS International Conference on Intelligent RObots and Systems

AI Summary

Low data efficiency in policy learning for multi-robot collaborative object pushing hinders rapid adaptation to novel tasks. Method: We propose a few-shot adaptive framework featuring (i) the first skill database explicitly designed for multi-agent collaboration; (ii) a Transformer-based spatiotemporal interaction-aware skill encoder for high-fidelity skill representation and precise behavior retrieval; and (iii) synergistic integration of retrieval-augmented training (RAT), multi-agent reinforcement learning, and few-shot imitation learning. Contribution/Results: Our approach enables rapid generalization to unseen collaborative scenarios using only a small number of demonstrations. Extensive evaluations on both simulation and real-world wheeled robot platforms demonstrate significant improvements in pushing success rates, consistently outperforming state-of-the-art few-shot baselines.

Abstract

Due to the complex interactions between agents, learning multi-agent control policy often requires a prohibitive amount of data. This paper aims to enable multi-agent systems to effectively utilize past memories to adapt to novel collaborative tasks in a data-efficient fashion. We propose the Multi-Agent Coordination Skill Database, a repository for storing a collection of coordinated behaviors associated with key vectors distinctive to them. Our Transformer-based skill encoder effectively captures spatio-temporal interactions that contribute to coordination and provides a unique skill representation for each coordinated behavior. By leveraging only a small number of demonstrations of the target task, the database enables us to train the policy using a dataset augmented with the retrieved demonstrations. Experimental evaluations demonstrate that our method achieves a significantly higher success rate in push manipulation tasks compared with baseline methods like few-shot imitation learning. Furthermore, we validate the effectiveness of our retrieve-and-learn framework in a real environment using a team of wheeled robots.

Problem

• Multi-Robot Collaboration
• Data-Efficient Learning
• Object Manipulation

Innovation

• Skill Library
• Transformer Model
• Multi-Robot Collaboration