Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think about thinking. ReMA decouples the reasoning process into two hierarchical agents: a high-level meta-thinking agent responsible for generating strategic oversight and plans, and a low-level reasoning agent for detailed executions. Through iterative reinforcement learning with aligned objectives, these agents explore and learn collaboration, leading to improved generalization and robustness. Experimental results demonstrate that ReMA outperforms single-agent RL baselines on complex reasoning tasks, including competitive-level mathematical benchmarks and LLM-as-a-Judge benchmarks. Comprehensive ablation studies further illustrate the evolving dynamics of each distinct agent, providing valuable insights into how the meta-thinking reasoning process enhances the reasoning capabilities of LLMs.
This paper addresses the multi-robot pursuit problem for an unknown target, encompassing both target state estimation and pursuit control. First, in state estimation, we focus on using only bearing information, as it is readily available from vision sensors and effective for small, distant targets. Challenges such as instability due to the nonlinearity of bearing measurements and singularities in the two-angle representation are addressed through a proposed uniform bearing-only information filter. This filter integrates multiple 3D bearing measurements, provides a concise formulation, and enhances stability and resilience to target loss caused by limited field of view (FoV). Second, in target pursuit control within complex environments, where challenges such as heterogeneity and limited FoV arise, conventional methods like differential games or Voronoi partitioning often prove inadequate. To address these limitations, we propose a novel multiagent reinforcement learning (MARL) framework, enabling multiple heterogeneous vehicles to search, localize, and follow a target while effectively handling those challenges. Third, to bridge the sim-to-real gap, we propose two key techniques: incorporating adjustable low-level control gains in training to replicate the dynamics of real-world autonomous ground vehicles (AGVs), and proposing spectral-normalized RL algorithms to enhance policy smoothness and robustness. Finally, we demonstrate the successful zero-shot transfer of the MARL controllers to AGVs, validating the effectiveness and practical feasibility of our approach. The accompanying video is available at https://youtu.be/HO7FJyZiJ3E.
Multi-agent reinforcement learning (MARL) has shown significant potential in traffic signal control (TSC). However, current MARL-based methods often suffer from insufficient generalization due to the fixed traffic patterns and road network conditions used during training. This limitation results in poor adaptability to new traffic scenarios, leading to high retraining costs and complex deployment. To address this challenge, we propose two algorithms: PLight and PRLight. PLight employs a model-based reinforcement learning approach, pretraining control policies and environment models using predefined source-domain traffic scenarios. The environment model predicts the state transitions, which facilitates the comparison of environmental features. PRLight further enhances adaptability by adaptively selecting pre-trained PLight agents based on the similarity between the source and target domains to accelerate the learning process in the target domain. We evaluated the algorithms through two transfer settings: (1) adaptability to different traffic scenarios within the same road network, and (2) generalization across different road networks. The results show that PRLight significantly reduces the adaptation time compared to learning from scratch in new TSC scenarios, achieving optimal performance using similarities between available and target scenarios.
Recently, with the development of Multi-agent reinforcement learning (MARL), adaptive traffic signal control (ATSC) has achieved satisfactory results. In traffic scenarios with multiple intersections, MARL treats each intersection as an agent and optimizes traffic signal control strategies through learning and real-time decision-making. Considering that observation distributions of intersections might be different in real-world scenarios, shared parameter methods might lack diversity and thus lead to high generalization requirements in the shared-policy network. A typical solution is to increase the size of network parameters. However, simply increasing the scale of the network does not necessarily improve policy generalization, which is validated in our experiments. Accordingly, an approach that considers both the personalization of intersections and the efficiency of parameter sharing is required. To this end, we propose Hyper-Action Multi-Head Proximal Policy Optimization (HAMH-PPO), a Centralized Training with Decentralized Execution (CTDE) MARL method that utilizes a shared PPO policy network to deliver personalized policies for intersections with non-iid observation distributions. The centralized critic in HAMH-PPO uses graph attention units to calculate the graph representations of all intersections and outputs a set of value estimates with multiple output heads for each intersection. The decentralized execution actor takes the local observation history as input and output distributions of action as well as a so-called hyper-action to balance the multiple values estimated from the centralized critic to further guide the updating of TSC policies. The combination of hyper-action and multi-head values enables multiple agents to share a single actor-critic while achieving personalized policies.
We introduce VLM-Attention, a multimodal StarCraft II environment that aligns artificial agent perception with the human gameplay experience. Traditional frameworks such as SMAC rely on abstract state representations that diverge significantly from human perception, limiting the ecological validity of agent behavior. Our environment addresses this limitation by incorporating RGB visual inputs and natural language observations that more closely simulate human cognitive processes during gameplay. The VLM-Attention framework consists of three integrated components: (1) a vision-language model enhanced with specialized self-attention mechanisms for strategic unit targeting and battlefield assessment, (2) a retrieval-augmented generation system that leverages domain-specific StarCraft II knowledge to inform tactical decisions, and (3) a dynamic role-based task distribution system that enables coordinated multi-agent behavior. Our experimental evaluation across 21 custom scenarios demonstrates that VLM-based agents powered by foundation models (specifically Qwen-VL and GPT-4o) can execute complex tactical maneuvers without explicit training, achieving comparable performance to traditional MARL methods that require substantial training iterations. This work establishes a foundation for developing human-aligned StarCraft II agents and advances the broader research agenda of multimodal game AI. Our implementation is available at https://github.com/camel-ai/VLM-Play-StarCraft2.
Teams of people coordinate to perform complex tasks by forming abstract mental models of world and agent dynamics. The use of abstract models contrasts with much recent work in robot learning that uses a high-fidelity simulator and reinforcement learning (RL) to obtain policies for physical robots. Motivated by this difference, we investigate the extent to which so-called abstract simulators can be used for multi-agent reinforcement learning (MARL) and the resulting policies successfully deployed on teams of physical robots. An abstract simulator models the robot's target task at a high-level of abstraction and discards many details of the world that could impact optimal decision-making. Policies are trained in an abstract simulator then transferred to the physical robot by making use of separately-obtained low-level perception and motion control modules. We identify three key categories of modifications to the abstract simulator that enable policy transfer to physical robots: simulation fidelity enhancements, training optimizations and simulation stochasticity. We then run an empirical study with extensive ablations to determine the value of each modification category for enabling policy transfer in cooperative robot soccer tasks. We also compare the performance of policies produced by our method with a well-tuned non-learning-based behavior architecture from the annual RoboCup competition and find that our approach leads to a similar level of performance. Broadly we show that MARL can be use to train cooperative physical robot behaviors using highly abstract models of the world.
Reinforcement Learning (RL) and Multi-Agent Reinforcement Learning (MARL) have emerged as promising methodologies for addressing challenges in automated cyber defence (ACD). These techniques offer adaptive decision-making capabilities in high-dimensional, adversarial environments. This report provides a structured set of guidelines for cybersecurity professionals and researchers to assess the suitability of RL and MARL for specific use cases, considering factors such as explainability, exploration needs, and the complexity of multi-agent coordination. It also discusses key algorithmic approaches, implementation challenges, and real-world constraints, such as data scarcity and adversarial interference. The report further outlines open research questions, including policy optimality, agent cooperation levels, and the integration of MARL systems into operational cybersecurity frameworks. By bridging theoretical advancements and practical deployment, these guidelines aim to enhance the effectiveness of AI-driven cyber defence strategies.
The manufacturing industry is undergoing a transformative shift, driven by cutting-edge technologies like 5G, AI, and cloud computing. Despite these advancements, effective system control, which is crucial for optimizing production efficiency, remains a complex challenge due to the intricate, knowledge-dependent nature of manufacturing processes and the reliance on domain-specific expertise. Conventional control methods often demand heavy customization, considerable computational resources, and lack transparency in decision-making. In this work, we investigate the feasibility of using Large Language Models (LLMs), particularly GPT-4, as a straightforward, adaptable solution for controlling manufacturing systems, specifically, mobile robot scheduling. We introduce an LLM-based control framework to assign mobile robots to different machines in robot assisted serial production lines, evaluating its performance in terms of system throughput. Our proposed framework outperforms traditional scheduling approaches such as First-Come-First-Served (FCFS), Shortest Processing Time (SPT), and Longest Processing Time (LPT). While it achieves performance that is on par with state-of-the-art methods like Multi-Agent Reinforcement Learning (MARL), it offers a distinct advantage by delivering comparable throughput without the need for extensive retraining. These results suggest that the proposed LLM-based solution is well-suited for scenarios where technical expertise, computational resources, and financial investment are limited, while decision transparency and system scalability are critical concerns.
Multi-Agent Reinforcement Learning (MARL) has shown promise in solving complex problems involving cooperation and competition among agents, such as an Unmanned Surface Vehicle (USV) swarm used in search and rescue, surveillance, and vessel protection. However, aligning system behavior with user preferences is challenging due to the difficulty of encoding expert intuition into reward functions. To address the issue, we propose a Reinforcement Learning with Human Feedback (RLHF) approach for MARL that resolves credit-assignment challenges through an Agent-Level Feedback system categorizing feedback into intra-agent, inter-agent, and intra-team types. To overcome the challenges of direct human feedback, we employ a Large Language Model (LLM) evaluator to validate our approach using feedback scenarios such as region constraints, collision avoidance, and task allocation. Our method effectively refines USV swarm policies, addressing key challenges in multi-agent systems while maintaining fairness and performance consistency.
Efficient path planning for unmanned aerial vehicles (UAVs) is crucial in remote sensing and information collection. As task scales expand, the cooperative deployment of multiple UAVs significantly improves information collection efficiency. However, collaborative communication and decision-making for multiple UAVs remain major challenges in path planning, especially in noisy environments. To efficiently accomplish complex information collection tasks in 3D space and address robust communication issues, we propose a multi-agent reinforcement learning (MARL) framework for UAV path planning based on the Counterfactual Multi-Agent Policy Gradients (COMA) algorithm. The framework incorporates attention mechanism-based UAV communication protocol and training-deployment system, significantly improving communication robustness and individual decision-making capabilities in noisy conditions. Experiments conducted on both synthetic and real-world datasets demonstrate that our method outperforms existing algorithms in terms of path planning efficiency and robustness, especially in noisy environments, achieving a 78\% improvement in entropy reduction.
Designing effective reward functions in multi-agent reinforcement learning (MARL) is a significant challenge, often leading to suboptimal or misaligned behaviors in complex, coordinated environments. We introduce Multi-agent Reinforcement Learning from Multi-phase Human Feedback of Mixed Quality (M3HF), a novel framework that integrates multi-phase human feedback of mixed quality into the MARL training process. By involving humans with diverse expertise levels to provide iterative guidance, M3HF leverages both expert and non-expert feedback to continuously refine agents' policies. During training, we strategically pause agent learning for human evaluation, parse feedback using large language models to assign it appropriately and update reward functions through predefined templates and adaptive weight by using weight decay and performance-based adjustments. Our approach enables the integration of nuanced human insights across various levels of quality, enhancing the interpretability and robustness of multi-agent cooperation. Empirical results in challenging environments demonstrate that M3HF significantly outperforms state-of-the-art methods, effectively addressing the complexities of reward design in MARL and enabling broader human participation in the training process.
Although multi-agent reinforcement learning (MARL) has shown its success across diverse domains, extending its application to large-scale real-world systems still faces significant challenges. Primarily, the high complexity of real-world environments exacerbates the credit assignment problem, substantially reducing training efficiency. Moreover, the variability of agent populations in large-scale scenarios necessitates scalable decision-making mechanisms. To address these challenges, we propose a novel framework: Sequential rollout with Sequential value estimation (SrSv). This framework aims to capture agent interdependence and provide a scalable solution for cooperative MARL. Specifically, SrSv leverages the autoregressive property of the Transformer model to handle varying populations through sequential action rollout. Furthermore, to capture the interdependence of policy distributions and value functions among multiple agents, we introduce an innovative sequential value estimation methodology and integrates the value approximation into an attention-based sequential model. We evaluate SrSv on three benchmarks: Multi-Agent MuJoCo, StarCraft Multi-Agent Challenge, and DubinsCars. Experimental results demonstrate that SrSv significantly outperforms baseline methods in terms of training efficiency without compromising convergence performance. Moreover, when implemented in a large-scale DubinsCar system with 1,024 agents, our framework surpasses existing benchmarks, highlighting the excellent scalability of SrSv.
In this paper, we consider the direct application of the relativistic extended thermodynamics theory of polyatomic gases developed in [Ann. Phys. 377 (2017) 414--445] to the relativistic BGK model proposed by Marle. We present the perturbed Marle model around the generalized J\"{u}ttner distribution and investigate the properties of the linear operator. Then we prove the global existence and large-time behavior of classical solutions when the initial data is sufficiently close to a global equilibrium.
This article presents the first field deployment of a multi-agent reinforcement learning (MARL) based variable speed limit (VSL) control system on Interstate 24 (I-24) near Nashville, Tennessee. We design and demonstrate a full pipeline from training MARL agents in a traffic simulator to a field deployment on a 17-mile segment of I-24 encompassing 67 VSL controllers. The system was launched on March 8th, 2024, and has made approximately 35 million decisions on 28 million trips in six months of operation. We apply an invalid action masking mechanism and several safety guards to ensure real-world constraints. The MARL-based implementation operates up to 98% of the time, with the safety guards overriding the MARL decisions for the remaining time. We evaluate the performance of the MARL-based algorithm in comparison to a previously deployed non-RL VSL benchmark algorithm on I-24. Results show that the MARL-based VSL control system achieves a superior performance. The accuracy of correctly warning drivers about slowing traffic ahead is improved by 14% and the response delay to non-recurrent congestion is reduced by 75%. The preliminary data shows that the VSL control system has reduced the crash rate by 26% and the secondary crash rate by 50%. We open-sourced the deployed MARL-based VSL algorithm at https://github.com/Lab-Work/marl-vsl-controller.
Mass casualty incidents (MCIs) are a growing concern, characterized by complexity and uncertainty that demand adaptive decision-making strategies. The victim tagging step in the emergency medical response must be completed quickly and is crucial for providing information to guide subsequent time-constrained response actions. In this paper, we present a mathematical formulation of multi-agent victim tagging to minimize the time it takes for responders to tag all victims. Five distributed heuristics are formulated and evaluated with simulation experiments. The heuristics considered are on-the go, practical solutions that represent varying levels of situational uncertainty in the form of global or local communication capabilities, showcasing practical constraints. We further investigate the performance of a multi-agent reinforcement learning (MARL) strategy, factorized deep Q-network (FDQN), to minimize victim tagging time as compared to baseline heuristics. Extensive simulations demonstrate that between the heuristics, methods with local communication are more efficient for adaptive victim tagging, specifically choosing the nearest victim with the option to replan. Analyzing all experiments, we find that our FDQN approach outperforms heuristics in smaller-scale scenarios, while heuristics excel in more complex scenarios. Our experiments contain diverse complexities that explore the upper limits of MARL capabilities for real-world applications and reveal key insights.
In cooperative multi-agent reinforcement learning (MARL), agents typically form a single grand coalition based on credit assignment to tackle a composite task, often resulting in suboptimal performance. This paper proposed a nucleolus-based credit assignment grounded in cooperative game theory, enabling the autonomous partitioning of agents into multiple small coalitions that can effectively identify and complete subtasks within a larger composite task. Specifically, our designed nucleolus Q-learning could assign fair credits to each agent, and the nucleolus Q-operator provides theoretical guarantees with interpretability for both learning convergence and the stability of the formed small coalitions. Through experiments on Predator-Prey and StarCraft scenarios across varying difficulty levels, our approach demonstrated the emergence of multiple effective coalitions during MARL training, leading to faster learning and superior performance in terms of win rate and cumulative rewards especially in hard and super-hard environments, compared to four baseline methods. Our nucleolus-based credit assignment showed the promise for complex composite tasks requiring effective subteams of agents.
In multi-robot exploration, a team of mobile robot is tasked with efficiently mapping an unknown environments. While most exploration planners assume omnidirectional sensors like LiDAR, this is impractical for small robots such as drones, where lightweight, directional sensors like cameras may be the only option due to payload constraints. These sensors have a constrained field-of-view (FoV), which adds complexity to the exploration problem, requiring not only optimal robot positioning but also sensor orientation during movement. In this work, we propose MARVEL, a neural framework that leverages graph attention networks, together with novel frontiers and orientation features fusion technique, to develop a collaborative, decentralized policy using multi-agent reinforcement learning (MARL) for robots with constrained FoV. To handle the large action space of viewpoints planning, we further introduce a novel information-driven action pruning strategy. MARVEL improves multi-robot coordination and decision-making in challenging large-scale indoor environments, while adapting to various team sizes and sensor configurations (i.e., FoV and sensor range) without additional training. Our extensive evaluation shows that MARVEL's learned policies exhibit effective coordinated behaviors, outperforming state-of-the-art exploration planners across multiple metrics. We experimentally demonstrate MARVEL's generalizability in large-scale environments, of up to 90m by 90m, and validate its practical applicability through successful deployment on a team of real drone hardware.
RouteRL is a novel framework that integrates multi-agent reinforcement learning (MARL) with a microscopic traffic simulation, facilitating the testing and development of efficient route choice strategies for autonomous vehicles (AVs). The proposed framework simulates the daily route choices of driver agents in a city, including two types: human drivers, emulated using behavioral route choice models, and AVs, modeled as MARL agents optimizing their policies for a predefined objective. RouteRL aims to advance research in MARL, transport modeling, and human-AI interaction for transportation applications. This study presents a technical report on RouteRL, outlines its potential research contributions, and showcases its impact via illustrative examples.
In cooperative multi-agent reinforcement learning (MARL), well-designed communication protocols can effectively facilitate consensus among agents, thereby enhancing task performance. Moreover, in large-scale multi-agent systems commonly found in real-world applications, effective communication plays an even more critical role due to the escalated challenge of partial observability compared to smaller-scale setups. In this work, we endeavor to develop a scalable communication protocol for MARL. Unlike previous methods that focus on selecting optimal pairwise communication links-a task that becomes increasingly complex as the number of agents grows-we adopt a global perspective on communication topology design. Specifically, we propose utilizing the exponential topology to enable rapid information dissemination among agents by leveraging its small-diameter and small-size properties. This approach leads to a scalable communication protocol, named ExpoComm. To fully unlock the potential of exponential graphs as communication topologies, we employ memory-based message processors and auxiliary tasks to ground messages, ensuring that they reflect global information and benefit decision-making. Extensive experiments on large-scale cooperative benchmarks, including MAgent and Infrastructure Management Planning, demonstrate the superior performance and robust zero-shot transferability of ExpoComm compared to existing communication strategies. The code is publicly available at https://github.com/LXXXXR/ExpoComm.
Cell-free (CF) massive multiple-input multiple-output (mMIMO) systems offer high spectral efficiency (SE) through multiple distributed access points (APs). However, the large number of antennas increases power consumption. We propose incorporating stacked intelligent metasurfaces (SIM) into CF mMIMO systems as a cost-effective, energy-efficient solution. This paper focuses on optimizing the joint power allocation of APs and the phase shift of SIMs to maximize the sum SE. To address this complex problem, we introduce a fully distributed multi-agent reinforcement learning (MARL) algorithm. Our novel algorithm, the noisy value method with a recurrent policy in multi-agent policy optimization (NVR-MAPPO), enhances performance by encouraging diverse exploration under centralized training and decentralized execution. Simulations demonstrate that NVR-MAPPO significantly improves sum SE and robustness across various scenarios.
Multiagent Reinforcement Learning (MARL) poses significant challenges due to the exponential growth of state and action spaces and the non-stationary nature of multiagent environments. This results in notable sample inefficiency and hinders generalization across diverse tasks. The complexity is further pronounced in relational settings, where domain knowledge is crucial but often underutilized by existing MARL algorithms. To overcome these hurdles, we propose integrating relational planners as centralized controllers with efficient state abstractions and reinforcement learning. This approach proves to be sample-efficient and facilitates effective task transfer and generalization.
Reinforcement learning (RL) emerges as a promising data-driven approach for adaptive traffic signal control (ATSC) in complex urban traffic networks, with deep neural networks substantially augmenting its learning capabilities. However, centralized RL becomes impractical for ATSC involving multiple agents due to the exceedingly high dimensionality of the joint action space. Multi-agent RL (MARL) mitigates this scalability issue by decentralizing control to local RL agents. Nevertheless, this decentralized method introduces new challenges: the environment becomes partially observable from the perspective of each local agent due to constrained inter-agent communication. Both centralized RL and MARL exhibit distinct strengths and weaknesses, particularly under heavy intersectional traffic conditions. In this paper, we justify that MARL can achieve the optimal global Q-value by separating into multiple IRL (Independent Reinforcement Learning) processes when no spill-back congestion occurs (no agent dependency) among agents (intersections). In the presence of spill-back congestion (with agent dependency), the maximum global Q-value can be achieved by using centralized RL. Building upon the conclusions, we propose a novel Dynamic Parameter Update Strategy for Deep Q-Network (DQN-DPUS), which updates the weights and bias based on the dependency dynamics among agents, i.e. updating only the diagonal sub-matrices for the scenario without spill-back congestion. We validate the DQN-DPUS in a simple network with two intersections under varying traffic, and show that the proposed strategy can speed up the convergence rate without sacrificing optimal exploration. The results corroborate our theoretical findings, demonstrating the efficacy of DQN-DPUS in optimizing traffic signal control.
Multi-agent reinforcement learning (MARL) faces challenges in coordinating agents due to complex interdependencies within multi-agent systems. Most MARL algorithms use the simultaneous decision-making paradigm but ignore the action-level dependencies among agents, which reduces coordination efficiency. In contrast, the sequential decision-making paradigm provides finer-grained supervision for agent decision order, presenting the potential for handling dependencies via better decision order management. However, determining the optimal decision order remains a challenge. In this paper, we introduce Action Generation with Plackett-Luce Sampling (AGPS), a novel mechanism for agent decision order optimization. We model the order determination task as a Plackett-Luce sampling process to address issues such as ranking instability and vanishing gradient during the network training process. AGPS realizes credit-based decision order determination by establishing a bridge between the significance of agents' local observations and their decision credits, thus facilitating order optimization and dependency management. Integrating AGPS with the Multi-Agent Transformer, we propose the Prioritized Multi-Agent Transformer (PMAT), a sequential decision-making MARL algorithm with decision order optimization. Experiments on benchmarks including StarCraft II Multi-Agent Challenge, Google Research Football, and Multi-Agent MuJoCo show that PMAT outperforms state-of-the-art algorithms, greatly enhancing coordination efficiency.
Recently testing of games via autonomous agents has shown great promise in tackling challenges faced by the game industry, which mainly relied on either manual testing or record/replay. In particular Reinforcement Learning (RL) solutions have shown potential by learning directly from playing the game without the need for human intervention. In this paper, we present cMarlTest, an approach for testing 3D games through curiosity driven Multi-Agent Reinforcement Learning (MARL). cMarlTest deploys multiple agents that work collaboratively to achieve the testing objective. The use of multiple agents helps resolve issues faced by a single agent approach. We carried out experiments on different levels of a 3D game comparing the performance of cMarlTest with a single agent RL variant. Results are promising where, considering three different types of coverage criteria, cMarlTest achieved higher coverage. cMarlTest was also more efficient in terms of the time taken, with respect to the single agent based variant.
Autonomous vehicles (AVs) using Multi-Agent Reinforcement Learning (MARL) for simultaneous route optimization may destabilize traffic environments, with human drivers possibly experiencing longer travel times. We study this interaction by simulating human drivers and AVs. Our experiments with standard MARL algorithms reveal that, even in trivial cases, policies often fail to converge to an optimal solution or require long training periods. The problem is amplified by the fact that we cannot rely entirely on simulated training, as there are no accurate models of human routing behavior. At the same time, real-world training in cities risks destabilizing urban traffic systems, increasing externalities, such as $CO_2$ emissions, and introducing non-stationarity as human drivers adapt unpredictably to AV behaviors. Centralization can improve convergence in some cases, however, it raises privacy concerns for the travelers' destination data. In this position paper, we argue that future research must prioritize realistic benchmarks, cautious deployment strategies, and tools for monitoring and regulating AV routing behaviors to ensure sustainable and equitable urban mobility systems.
Agents built on large language models (LLMs) have excelled in turn-by-turn human-AI collaboration but struggle with simultaneous tasks requiring real-time interaction. Latency issues and the challenge of inferring variable human strategies hinder their ability to make autonomous decisions without explicit instructions. Through experiments with current independent System 1 and System 2 methods, we validate the necessity of using Dual Process Theory (DPT) in real-time tasks. We propose DPT-Agent, a novel language agent framework that integrates System 1 and System 2 for efficient real-time simultaneous human-AI collaboration. DPT-Agent's System 1 uses a Finite-state Machine (FSM) and code-as-policy for fast, intuitive, and controllable decision-making. DPT-Agent's System 2 integrates Theory of Mind (ToM) and asynchronous reflection to infer human intentions and perform reasoning-based autonomous decisions. We demonstrate the effectiveness of DPT-Agent through further experiments with rule-based agents and human collaborators, showing significant improvements over mainstream LLM-based frameworks. DPT-Agent can effectively help LLMs convert correct slow thinking and reasoning into executable actions, thereby improving performance. To the best of our knowledge, DPT-Agent is the first language agent framework that achieves successful real-time simultaneous human-AI collaboration autonomously. Code of DPT-Agent can be found in https://github.com/sjtu-marl/DPT-Agent.
Offline Reinforcement Learning (RL) focuses on learning policies solely from a batch of previously collected data. offering the potential to leverage such datasets effectively without the need for costly or risky active exploration. While recent advances in Offline Multi-Agent RL (MARL) have shown promise, most existing methods either rely on large datasets jointly collected by all agents or agent-specific datasets collected independently. The former approach ensures strong performance but raises scalability concerns, while the latter emphasizes scalability at the expense of performance guarantees. In this work, we propose a novel scalable routine for both dataset collection and offline learning. Agents first collect diverse datasets coherently with a pre-specified information-sharing network and subsequently learn coherent localized policies without requiring either full observability or falling back to complete decentralization. We theoretically demonstrate that this structured approach allows a multi-agent extension of the seminal Fitted Q-Iteration (FQI) algorithm to globally converge, in high probability, to near-optimal policies. The convergence is subject to error terms that depend on the informativeness of the shared information. Furthermore, we show how this approach allows to bound the inherent error of the supervised-learning phase of FQI with the mutual information between shared and unshared information. Our algorithm, SCAlable Multi-agent FQI (SCAM-FQI), is then evaluated on a distributed decision-making problem. The empirical results align with our theoretical findings, supporting the effectiveness of SCAM-FQI in achieving a balance between scalability and policy performance.
Despite much progress in training distributed artificial intelligence (AI), building cooperative multi-agent systems with multi-agent reinforcement learning (MARL) faces challenges in sample efficiency, interpretability, and transferability. Unlike traditional learning-based methods that require extensive interaction with the environment, large language models (LLMs) demonstrate remarkable capabilities in zero-shot planning and complex reasoning. However, existing LLM-based approaches heavily rely on text-based observations and struggle with the non-Markovian nature of multi-agent interactions under partial observability. We present COMPASS, a novel multi-agent architecture that integrates vision-language models (VLMs) with a dynamic skill library and structured communication for decentralized closed-loop decision-making. The skill library, bootstrapped from demonstrations, evolves via planner-guided tasks to enable adaptive strategies. COMPASS propagates entity information through multi-hop communication under partial observability. Evaluations on the improved StarCraft Multi-Agent Challenge (SMACv2) demonstrate COMPASS achieves up to 30\% higher win rates than state-of-the-art MARL algorithms in symmetric scenarios.
Multi-agent reinforcement learning (MARL) lies at the heart of a plethora of applications involving the interaction of a group of agents in a shared unknown environment. A prominent framework for studying MARL is Markov games, with the goal of finding various notions of equilibria in a sample-efficient manner, such as the Nash equilibrium (NE) and the coarse correlated equilibrium (CCE). However, existing sample-efficient approaches either require tailored uncertainty estimation under function approximation, or careful coordination of the players. In this paper, we propose a novel model-based algorithm, called VMG, that incentivizes exploration via biasing the empirical estimate of the model parameters towards those with a higher collective best-response values of all the players when fixing the other players' policies, thus encouraging the policy to deviate from its current equilibrium for more exploration. VMG is oblivious to different forms of function approximation, and permits simultaneous and uncoupled policy updates of all players. Theoretically, we also establish that VMG achieves a near-optimal regret for finding both the NEs of two-player zero-sum Markov games and CCEs of multi-player general-sum Markov games under linear function approximation in an online environment, which nearly match their counterparts with sophisticated uncertainty quantification.
As a data-driven approach, offline MARL learns superior policies solely from offline datasets, ideal for domains rich in historical data but with high interaction costs and risks. However, most existing methods are task-specific, requiring retraining for new tasks, leading to redundancy and inefficiency. To address this issue, in this paper, we propose a task-efficient multi-task offline MARL algorithm, Skill-Discovery Conservative Q-Learning (SD-CQL). Unlike existing offline skill-discovery methods, SD-CQL discovers skills by reconstructing the next observation. It then evaluates fixed and variable actions separately and employs behavior-regularized conservative Q-learning to execute the optimal action for each skill. This approach eliminates the need for local-global alignment and enables strong multi-task generalization from limited small-scale source tasks. Substantial experiments on StarCraftII demonstrates the superior generalization performance and task-efficiency of SD-CQL. It achieves the best performance on $\textbf{10}$ out of $14$ task sets, with up to $\textbf{65%}$ improvement on individual task sets, and is within $4\%$ of the best baseline on the remaining four.