The findings demonstrate that combining relational state representations, copula-based dependence modeling, and shared rewards improves coordination, robustness, and stability in multi-agent navigation tasks.
Abstract
Cooperative multi-agent reinforcement learning (MARL) enables autonomous agents to coordinate in complex spatial environments. This study proposes a MARL framework for goal-directed navigation that integrates entangled state embeddings, copula-based joint action transformations, and a shared reward mechanism. Entangled representations enhance cooperative awareness by incorporating peer-agent information, while copula transformations model dependence among actions to stabilize joint decision-making. A shared reward structure further aligns agent objectives toward collective performance. The framework is evaluated in synthetic continuous navigation environments and a Chicago crime-based real-data setting. Simulation studies compare three configurations: a full Copula Model, a No Copula model, and a Baseline model without cooperative enhancements. Results show that the Copula Model achieves the highest cumulative reward and the most stable coordination, whereas the Baseline model consistently underperforms. Sensitivity analysis indicates that intermediate actor learning rates provide the most stable convergence. Trajectory analyses reveal emergent cooperative behaviors such as spatial dispersion, obstacle avoidance, and coordinated movement toward target regions. In the Chicago crime experiment, agents exhibit risk-aware navigation and non-redundant exploration despite environmental complexity and sparse rewards. Overall, the findings demonstrate that combining relational state representations, copula-based dependence modeling, and shared rewards improves coordination, robustness, and stability in multi-agent navigation tasks.
Safe decision-making in partially observable dynamic environments remains a fundamental challenge for multi-agent embodied reinforcement learning, where agents must cooperate with limited local observations while responding to evolving environmental changes and safety-critical interactions. Existing methods often empha...
Bo-Zhi Zhang, Jiang-Bo Wang, Tian Jing· International journal of pat...· 0 citations
This work formulate large-scale MAPP as a partially observable networked Markov decision process as a decentralized model-based Actor-Critic using the Kronecker-factored trust region (DM-ACKTR) algorithm, which consistently obtains the highest TCR and lowest CR.
Ye-Min Liu, Jin-Hao Yang, Xiang-Yu Ma et al.· Electronics· 0 citations
Multi-agent pathfinding (MAPF) requires multiple agents to reach assigned goals while avoiding vertex, edge, and intermittent-obstacle conflicts in a shared workspace. Learning-based MAPF is attractive for online decision-making, but fully decentralized reinforcement learning remains vulnerable to sparse rewards, non-s...
This work proposes scalable coordination via optimal unified transport (SCOUT), the first offline MARL framework to combine a generative foundation model with a learned value function through test-time action refinement, and proves a single-term KL bound on the joint soft-value gap that vanishes as transport converges.
Multi-agent reinforcement learning (MARL) provides a powerful framework for learning coordinated behaviors through interactions with the environment. Developing MARL policies requires balancing expressive modeling of complex and multimodal action distributions with efficient training and execution. Generative policies,...
Zhuo-Ran Li, Yun-Zhan Li, Xun Wang et al.· 0 citations
This work proposes a multiagent belief state computation method termed mutual awareness belief (MAB), which extends belief modeling to decentralized partially observable settings by incorporating agents' behavioral intentions and collaborative relationships and proves effective in the nonmonotonic Pursuit environment.
Si-Han Zhou, Tian-Tian He, Yi-Nan Guo et al.· IEEE Transactions on Neural...· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 9, 2026