Advances in Multi-Agent Reinforcement Learning

The field of multi-agent reinforcement learning (MARL) is rapidly advancing, with a focus on developing innovative methods to improve cooperation, communication, and decision-making among agents. Recent research has explored the challenges of credit assignment, decentralized learning, and heterogeneous agent collaboration. Notably, new approaches have been proposed to address the credit assignment problem in open systems, such as conceptual and empirical analyses of openness and its impact on traditional credit assignment methods. Additionally, novel frameworks have been introduced to facilitate scalable, perception-aware imitation learning in multi-agent collaborative systems. Other notable developments include the use of Gaussian-image synergy, predictive auxiliary learning, and differentiable discrete communication learning to enhance MARL performance.

Some noteworthy papers in this area include: NegoCollab, which proposes a heterogeneous collaboration method based on negotiated common representation, effectively reducing domain gaps and improving collaborative performance. GauDP, which presents a novel Gaussian-image synergistic representation for scalable, perception-aware imitation learning in multi-agent collaborative systems, achieving superior performance over existing image-based methods. From Pixels to Cooperation, which introduces a framework based on a shared, generative Multimodal World Model to learn cooperative MARL policies from high-dimensional, multimodal sensory inputs, demonstrating orders-of-magnitude greater sample efficiency compared to state-of-the-art model-free MARL baselines.

Sources

Challenges in Credit Assignment for Multi-Agent Reinforcement Learning in Open Agent Systems

NegoCollab: A Common Representation Negotiation Approach for Heterogeneous Collaborative Perception

On the Fundamental Limitations of Decentralized Learnable Reward Shaping in Cooperative Multi-Agent Reinforcement Learning

Stochastic Shortest Path with Sparse Adversarial Costs

GauDP: Reinventing Multi-Agent Collaboration through Gaussian-Image Synergy in Diffusion Policies

Predictive Auxiliary Learning for Belief-based Multi-Agent Systems

From Pixels to Cooperation Multi Agent Reinforcement Learning based on Multimodal World Models

Learning what to say and how precisely: Efficient Communication via Differentiable Discrete Communication Learning

Learning Intractable Multimodal Policies with Reparameterization and Diversity Regularization

Game-theoretic distributed learning of generative models for heterogeneous data collections

A Quantitative Comparison of Centralised and Distributed Reinforcement Learning-Based Control for Soft Robotic Arms

Automata-Conditioned Cooperative Multi-Agent Reinforcement Learning

A Spatially Informed Gaussian Process UCB Method for Decentralized Coverage Control

From Solo to Symphony: Orchestrating Multi-Agent Collaboration with Single-Agent Demos

Regret Lower Bounds for Decentralized Multi-Agent Stochastic Shortest Path Problems

Built with on top of