DeReCo: Decoupling Representation and Coordination Learning for Object-Adaptive Decentralized Multi-Robot Cooperative Transport

TL;DR

DeReCo introduces a three-stage decentralized multi-robot transport framework that decouples representation and coordination learning, enhancing sample efficiency and generalization across diverse objects and scenarios.

Contribution

The paper proposes a novel decoupled MARL framework, DeReCo, which separates representation and coordination learning to improve stability and generalization in multi-robot transport tasks.

Findings

Outperforms baselines in simulation on training objects

Generalizes effectively to unseen objects with different physical properties

Achieves superior real-robot performance on new objects

Abstract

Generalizing decentralized multi-robot cooperative transport across objects with diverse shapes and physical properties remains a fundamental challenge. Under decentralized execution, two key challenges arise: object-dependent representation learning under partial observability and coordination learning in multi-agent reinforcement learning (MARL) under non-stationarity. A typical approach jointly optimizes object-dependent representations and coordinated policies in an end-to-end manner while randomizing object shapes and physical properties during training. However, this joint optimization tightly couples representation and coordination learning, introducing bidirectional interference: inaccurate representations under partial observability destabilize coordination learning, while non-stationarity in MARL further degrades representation learning, resulting in sample-inefficient training. To address this structural coupling, we propose DeReCo, a novel MARL framework that decouples representation and coordination learning for object-adaptive multi-robot cooperative transport, improving sample efficiency and generalization across objects and transport scenarios. DeReCo adopts a three-stage training strategy: (1) centralized coordination learning with privileged object information, (2) reconstruction of object-dependent representations from local observations, and (3) progressive removal of privileged information for decentralized execution. This decoupling mitigates interference between representation and coordination learning and enables stable and sample-efficient training. Experimental results show that DeReCo outperforms baselines in simulation on three training objects, generalizes to six unseen objects with varying masses and friction coefficients, and achieves superior performance on two unseen objects in real-robot experiments.