The field of multi-agent reinforcement learning (MARL) is rapidly advancing, with a focus on developing more efficient and effective algorithms for cooperative and competitive environments. Recent research has explored the use of centralized training with decentralized execution, permutation equivariant architectures, and autoregressive action world models to improve performance in complex multi-agent settings. Additionally, there is a growing interest in developing more realistic and challenging benchmarks, such as those that incorporate continuous action spaces and uncertain environments. Noteworthy papers in this area include the proposal of Centralized Permutation Equivariant (CPE) learning, which has been shown to substantially improve performance in cooperative benchmarks, and the introduction of MAPF-World, an autoregressive action world model that enables more informed and coordinated decision-making in multi-agent path finding scenarios. Other notable contributions include the development of novel communication mechanisms, such as self-attention-based modules, and the investigation of payoff manipulation in repeated multi-objective Stackelberg games.
Advances in Multi-Agent Reinforcement Learning
Sources
Nominal Evaluation Of Automatic Multi-Sections Control Potential In Comparison To A Simpler One- Or Two-Sections Alternative With Predictive Spray Switching
Centralized Permutation Equivariant Policy for Cooperative Multi-Agent Reinforcement Learning
MAPF-World: Action World Model for Multi-Agent Path Finding
The Yokai Learning Environment: Tracking Beliefs Over Space and Time
DCT-MARL: A Dynamic Communication Topology-Based MARL Algorithm for Connected Vehicle Platoon Control
CAMAR: Continuous Actions Multi-Agent Routing
MACTAS: Self-Attention-Based Module for Inter-Agent Communication in Multi-Agent Reinforcement Learning
Convergent Reinforcement Learning Algorithms for Stochastic Shortest Path Problem
Efficient Environment Design for Multi-Robot Navigation via Continuous Control
An Improved Multi-Agent Algorithm for Cooperative and Competitive Environments by Identifying and Encouraging Cooperation among Agents
Learning in Repeated Multi-Objective Stackelberg Games with Payoff Manipulation
Universal Reinforcement Learning in Coalgebras: Asynchronous Stochastic Computation via Conduction
Understanding Action Effects through Instrumental Empowerment in Multi-Agent Reinforcement Learning
An Efficient Open World Environment for Multi-Agent Social Learning
Distributed Detection of Adversarial Attacks in Multi-Agent Reinforcement Learning with Continuous Action Space