Multi-Agent Reinforcement Learning
Also known as MARL
Overview
Multi-agent reinforcement learning studies how multiple decision-makers learn when each agent’s actions affect both the environment and other agents, across cooperative, competitive, and mixed-motive settings. Common approaches include centralized training with decentralized execution, independent or shared actor-critic policies, value factorization such as QMIX, self-play and game-solving methods, and graph-based communication or hierarchical policies. Researchers test these methods in benchmarks such as SMAC, MPE, Overcooked, and multi-agent MuJoCo, and apply them to robotics, traffic, wireless networks, energy systems, and language-agent workflows. A central challenge is scaling coordination and credit assignment while keeping learning stable under non-stationarity and policies robust, safe, and fair with unfamiliar partners and conditions.
Latest papers
402 papers37 papers in the last four weeks, down 27% on the four weeks before. 0.3% of all new papers.
Multi-Agent Transformer for Queue-Level XR Traffic Scheduling in TSN Networks
PRIMAL3: Pathfinding via Reinforcement and Imitation Multi-Agent Learning - Leveraging LaCAM3
History Matters: Meta-policy Delegation with Heterogeneous Multi-agent Reinforcement Learning
Is Inter-Seed Cross-Play Enough? Evaluating the Robustness of Zero-Shot Coordination Algorithms to Implementation Details
Lagrangian or Lyapunov? Multi-Agent Reinforcement Learning under Coupled Finite-Horizon Constraints
Policy Optimality Measurement for Multi-Vehicle Decision-Making: From Extrinsic Indicators to Intrinsic Quality
MA-HEAD-Net: Adaptive Rule-Guided Multi-Agent DRL for AoI Minimization in UAV-Assisted Emergency Networks
Shared Voxel-Map-Based Cooperative Indoor UAV Guidance with a Multi-Agent Soft Actor-Critic Controller
PLATO: Pointer Learner for Agent and Task Openness
Predictive Lightweight MARL for Resilient Coverage in Sparse-Signaling Aerial Networks
Coordinating from Memory: Graph-Structured Experience Reuse for Multi-Agent Adaptation in Dynamic Manufacturing
CHMAS: A Coupled Hierarchical Framework for Multi-Agent Reinforcement Learning
Comparative Study of Multi-Agent Actor-Critic Algorithms in Parameterized Action Reinforcement Learning
Strategy-Following Multi-Agent Deep Reinforcement Learning Considering Control Strategies Provided to Other Agents
A Self-Evolving Default Action for Cooperative Tasks with Continuous Action Space
Scalable Policy Optimization for Networked Multi-Agent Reinforcement Learning with Continuous State-Action Spaces
Aggregate in the Advantage, Not the Ratio: A Canonical-Form Analysis of Cooperative Multi-Agent Policy Optimization
Reinforcement Learning: From Algorithms To Foundation Models
Feedback Attribution and Representation Geometry: Metrics for Comparing Individual and Shared Rewards in MARL
protoss\_5\_vs\_5, where unit type is encoded in the observation. In an observation reward-attribution comparison (unit type observed vs.\ masked; individual damage-contribution reward vs.\ shared team reward), geometry follows observation rather than reward. With unit type observed, shared and individual rewards have similar EffRank/ ( vs.\ ) and probe accuracy ( vs.\ , both chance), while leans higher under individual rewards ( vs.\ ). Masking unit type cuts the above-chance probe signal by more than half, to in both reward arms. In short: individually rewarded agents are competent and separable by role, but on SMACv2 the observation explains the geometry and reward attribution shows up mainly in behavior. Thus geometric diagnostics must control for observed role information and test persistent roles that are not directly observed. EffRank/ and add 5% overhead.