Structure-Aware Reinforcement Learning for Cooperative Multi-Robot Systems

Multi-robot systems are increasingly deployed in environments where centralized models and fixed coordination rules are difficult to maintain. Reinforcement learning and multi-agent reinforcement learning offer a flexible route to decentralized decision-making, but their use on physical robot teams remains constrained by sample inefficiency, partial observability, changing team composition, limited communication, safety requirements, and the gap between simulation and deployment. Recent work on symmetry, graph structure, geometric priors, model-based learning, and representation sharing suggests that explicit knowledge of agent relations and task structure can improve learning efficiency and generalization. This Research Topic is needed to connect these algorithmic advances with the design, analysis, and experimental validation of cooperative multi-robot behavior.This Research Topic will examine how structural knowledge can be incorporated into RL and MARL to produce cooperative robot teams that learn efficiently, coordinate under uncertainty, and transfer across tasks, team sizes, embodiments, and environments. We seek methods that use symmetries, graphs, dynamics, constraints, communication topology, task decomposition, or domain knowledge as inductive biases while retaining the adaptability of data-driven learning. This Research Topic will bring together theoretical analysis, algorithm design, simulation studies, benchmarks, and physical-robot experiments. Particular attention will be given to safety, robustness, scalability, interpretability, and reproducibility, as well as to evidence that a proposed method improves real multi-robot decision-making rather than only performance in abstract multi-agent games. The goal is to clarify which forms of structure are most useful, how they should be represented, and how their benefits persist when assumptions or operating conditions change.We welcome Original Research, Methods, Review, Systematic Review, Perspective, and Technology and Code articles on:- structure- and symmetry-aware RL and MARL, including graph-based policies, value functions, and structured representations;- cooperative learning, communication, credit assignment, and centralized training with decentralized execution;- safe, robust, risk-sensitive, constrained, interpretable, and reproducible robot learning;- partial observability, transfer, continual, offline, foundation-model-assisted, and sim-to-real learning;- heterogeneous and variable-size teams, benchmarks, and physical multi-robot experiments;- applications in coverage, exploration, mapping, formation, task allocation, navigation, manipulation, and aerial, ground, marine, or mixed robot swarms.Submissions focused on non-robotic multi-agent systems should explain a clear contribution or pathway to multi-robot applications. Topic editors: Rongye Shi, Jun Jin, Xingcheng Fu, Huiming Zhang Source: https://www.frontiersin.org/research-topics/86911/structure-aware-reinforcement-learning-for-cooperative-multi-robot-systems