LLM-based agents have made significant advancements in interactive environments, such as mobile operations and web browsing, and other domains beyond computer using. Current multi-agent systems universally excel in performance, compared to single agents, but struggle with generalization across environments due to predefined roles and inadequate strategies for generalizing language agents. The challenge of achieving both strong performance and good generalization has hindered the progress of multi-agent systems for interactive environments. To address these issues, we propose CollabUIAgents, a multi-agent reinforcement learning framework with a novel multi-agent credit re-assignment (CR) strategy, assigning process rewards with LLMs rather than environment-specific rewards and learning with synthesized preference data, in order to foster generalizable, collaborative behaviors among the role-free agents' policies. Empirical results show that our framework improves both performance and cross-environment generalizability of multi-agent systems. Moreover, our 7B-parameter system achieves results on par with or exceed strong closed-source models, and the LLM that guides the CR. We also provide insights in using granular CR rewards effectively for environment generalization, and accommodating trained LLMs in multi-agent systems.
Scalability remains a challenge in multi-agent reinforcement learning and is currently under active research. A framework named mean-field reinforcement learning (MFRL) could alleviate the scalability problem by employing the Mean Field Theory to turn a many-agent problem into a two-agent problem. However, this framework lacks the ability to identify essential interactions under nonstationary environments. Causality contains relatively invariant mechanisms behind interactions, though environments are nonstationary. Therefore, we propose an algorithm called causal mean-field Q-learning (CMFQ) to address the scalability problem. CMFQ is ever more robust toward the change of the number of agents though inheriting the compressed representation of MFRL's action-state space. Firstly, we model the causality behind the decision-making process of MFRL into a structural causal model (SCM). Then the essential degree of each interaction is quantified via intervening on the SCM. Furthermore, we design the causality-aware compact representation for behavioral information of agents as the weighted sum of all behavioral information according to their causal effects. We test CMFQ in a mixed cooperative-competitive game and a cooperative game. The result shows that our method has excellent scalability performance in both training in environments containing a large number of agents and testing in environments containing much more agents.
Guiding the policy of multi-agent reinforcement learning to align with human common sense is a difficult problem, largely due to the complexity of modeling common sense as a reward, especially in complex and long-horizon multi-agent tasks. Recent works have shown the effectiveness of reward shaping, such as potential-based rewards, to enhance policy alignment. The existing works, however, primarily rely on experts to design rule-based rewards, which are often labor-intensive and lack a high-level semantic understanding of common sense. To solve this problem, we propose a hierarchical vision-based reward shaping method. At the bottom layer, a visual-language model (VLM) serves as a generic potential function, guiding the policy to align with human common sense through its intrinsic semantic understanding. To help the policy adapts to uncertainty and changes in long-horizon tasks, the top layer features an adaptive skill selection module based on a visual large language model (vLLM). The module uses instructions, video replays, and training records to dynamically select suitable potential function from a pre-designed pool. Besides, our method is theoretically proven to preserve the optimal policy. Extensive experiments conducted in the Google Research Football environment demonstrate that our method not only achieves a higher win rate but also effectively aligns the policy with human common sense.
Autonomous vehicles (AVs), possibly using Multi-Agent Reinforcement Learning (MARL) for simultaneous route optimization, may destabilize traffic networks, with human drivers potentially experiencing longer travel times. We study this interaction by simulating human drivers and AVs. Our experiments with standard MARL algorithms reveal that, both in simplified and complex networks, policies often fail to converge to an optimal solution or require long training periods. This 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 will adapt unpredictably to AV behaviors. In this position paper, we argue that city authorities must collaborate with the ML community to monitor and critically evaluate the routing algorithms proposed by car companies toward fair and system-efficient routing algorithms and regulatory standards.
Partially Controlled Multi-Agent Systems (PCMAS) are comprised of controllable agents, managed by a system designer, and uncontrollable agents, operating autonomously. This study addresses an optimal composition design problem in PCMAS, which involves the system designer's problem, determining the optimal number and policies of controllable agents, and the uncontrollable agents' problem, identifying their best-response policies. Solving this bi-level optimization problem is computationally intensive, as it requires repeatedly solving multi-agent reinforcement learning problems under various compositions for both types of agents. To address these challenges, we propose a novel hypernetwork-based framework that jointly optimizes the system's composition and agent policies. Unlike traditional methods that train separate policy networks for each composition, the proposed framework generates policies for both controllable and uncontrollable agents through a unified hypernetwork. This approach enables efficient information sharing across similar configurations, thereby reducing computational overhead. Additional improvements are achieved by incorporating reward parameter optimization and mean action networks. Using real-world New York City taxi data, we demonstrate that our framework outperforms existing methods in approximating equilibrium policies. Our experimental results show significant improvements in key performance metrics, such as order response rate and served demand, highlighting the practical utility of controlling agents and their potential to enhance decision-making in PCMAS.
The Mixed-Shelves Picker Routing Problem (MSPRP) is a fundamental challenge in warehouse logistics, where pickers must navigate a mixed-shelves environment to retrieve SKUs efficiently. Traditional heuristics and optimization-based approaches struggle with scalability, while recent machine learning methods often rely on sequential decision-making, leading to high solution latency and suboptimal agent coordination. In this work, we propose a novel hierarchical and parallel decoding approach for solving the min-max variant of the MSPRP via multi-agent reinforcement learning. While our approach generates a joint distribution over agent actions, allowing for fast decoding and effective picker coordination, our method introduces a sequential action selection to avoid conflicts in the multi-dimensional action space. Experiments show state-of-the-art performance in both solution quality and inference speed, particularly for large-scale and out-of-distribution instances. Our code is publicly available at http://github.com/LTluttmann/marl4msprp.
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's strong performance against state-of-the-art MARL baselines across both symmetric and asymmetric scenarios. Notably, in the symmetric Protoss 5v5 task, COMPASS achieved a 57\% win rate, representing a 30 percentage point advantage over QMIX (27\%). Project page can be found at https://stellar-entremet-1720bb.netlify.app/.
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.
The increasing deployment of unmanned surface vehicles (USVs) require computational support and coverage in applications such as maritime search and rescue. Unmanned aerial vehicles (UAVs) can offer low-cost, flexible aerial services, and ground stations (GSs) can provide powerful supports, which can cooperate to help the USVs in complex scenarios. However, the collaboration between UAVs and GSs for USVs faces challenges of task uncertainties, USVs trajectory uncertainties, heterogeneities, and limited computational resources. To address these issues, we propose a cooperative UAV and GS based robust multi-access edge computing framework to assist USVs in completing computational tasks. Specifically, we formulate the optimization problem of joint task offloading and UAV trajectory to minimize the total execution time, which is in the form of mixed integer nonlinear programming and NP-hard to tackle. Therefore, we propose the algorithm of generative artificial intelligence-enhanced heterogeneous agent proximal policy optimization (GAI-HAPPO). The proposed algorithm integrates GAI models to enhance the actor network ability to model complex environments and extract high-level features, thereby allowing the algorithm to predict uncertainties and adapt to dynamic conditions. Additionally, GAI stabilizes the critic network, addressing the instability of multi-agent reinforcement learning approaches. Finally, extensive simulations demonstrate that the proposed algorithm outperforms the existing benchmark methods, thus highlighting the potentials in tackling intricate, cross-domain issues in the considered scenarios.
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.
Power grid operation is becoming more complex due to the increase in generation of renewable energy. The recent series of Learning To Run a Power Network (L2RPN) competitions have encouraged the use of artificial agents to assist human dispatchers in operating power grids. However, the combinatorial nature of the action space poses a challenge to both conventional optimizers and learned controllers. Action space factorization, which breaks down decision-making into smaller sub-tasks, is one approach to tackle the curse of dimensionality. In this study, we propose a centrally coordinated multi-agent (CCMA) architecture for action space factorization. In this approach, regional agents propose actions and subsequently a coordinating agent selects the final action. We investigate several implementations of the CCMA architecture, and benchmark in different experimental settings against various L2RPN baseline approaches. The CCMA architecture exhibits higher sample efficiency and superior final performance than the baseline approaches. The results suggest high potential of the CCMA approach for further application in higher-dimensional L2RPN as well as real-world power grid settings.
Medication adherence is critical for the recovery of adolescents and young adults (AYAs) who have undergone hematopoietic cell transplantation (HCT). However, maintaining adherence is challenging for AYAs after hospital discharge, who experience both individual (e.g. physical and emotional symptoms) and interpersonal barriers (e.g., relational difficulties with their care partner, who is often involved in medication management). To optimize the effectiveness of a three-component digital intervention targeting both members of the dyad as well as their relationship, we propose a novel Multi-Agent Reinforcement Learning (MARL) approach to personalize the delivery of interventions. By incorporating the domain knowledge, the MARL framework, where each agent is responsible for the delivery of one intervention component, allows for faster learning compared with a flattened agent. Evaluation using a dyadic simulator environment, based on real clinical data, shows a significant improvement in medication adherence (approximately 3%) compared to purely random intervention delivery. The effectiveness of this approach will be further evaluated in an upcoming trial.
Developing intelligent agents for long-term cooperation in dynamic open-world scenarios is a major challenge in multi-agent systems. Traditional Multi-agent Reinforcement Learning (MARL) frameworks like centralized training decentralized execution (CTDE) struggle with scalability and flexibility. They require centralized long-term planning, which is difficult without custom reward functions, and face challenges in processing multi-modal data. CTDE approaches also assume fixed cooperation strategies, making them impractical in dynamic environments where agents need to adapt and plan independently. To address decentralized multi-agent cooperation, we propose Decentralized Adaptive Knowledge Graph Memory and Structured Communication System (DAMCS) in a novel Multi-agent Crafter environment. Our generative agents, powered by Large Language Models (LLMs), are more scalable than traditional MARL agents by leveraging external knowledge and language for long-term planning and reasoning. Instead of fully sharing information from all past experiences, DAMCS introduces a multi-modal memory system organized as a hierarchical knowledge graph and a structured communication protocol to optimize agent cooperation. This allows agents to reason from past interactions and share relevant information efficiently. Experiments on novel multi-agent open-world tasks show that DAMCS outperforms both MARL and LLM baselines in task efficiency and collaboration. Compared to single-agent scenarios, the two-agent scenario achieves the same goal with 63% fewer steps, and the six-agent scenario with 74% fewer steps, highlighting the importance of adaptive memory and structured communication in achieving long-term goals. We publicly release our project at: https://happyeureka.github.io/damcs.
We investigate the problem of distributed training under partial observability, whereby cooperative multi-agent reinforcement learning agents (MARL) maximize the expected cumulative joint reward. We propose distributed value decomposition networks (DVDN) that generate a joint Q-function that factorizes into agent-wise Q-functions. Whereas the original value decomposition networks rely on centralized training, our approach is suitable for domains where centralized training is not possible and agents must learn by interacting with the physical environment in a decentralized manner while communicating with their peers. DVDN overcomes the need for centralized training by locally estimating the shared objective. We contribute with two innovative algorithms, DVDN and DVDN (GT), for the heterogeneous and homogeneous agents settings respectively. Empirically, both algorithms approximate the performance of value decomposition networks, in spite of the information loss during communication, as demonstrated in ten MARL tasks in three standard environments.
State-of-the-art methods for Human-AI Teaming and Zero-shot Cooperation focus on task completion, i.e., task rewards, as the sole evaluation metric while being agnostic to how the two agents work with each other. Furthermore, subjective user studies only offer limited insight into the quality of cooperation existing within the team. Specifically, we are interested in understanding the cooperative behaviors arising within the team when trained agents are paired with humans -- a problem that has been overlooked by the existing literature. To formally address this problem, we propose the concept of constructive interdependence -- measuring how much agents rely on each other's actions to achieve the shared goal -- as a key metric for evaluating cooperation in human-agent teams. We interpret interdependence in terms of action interactions in a STRIPS formalism, and define metrics that allow us to assess the degree of reliance between the agents' actions. We pair state-of-the-art agents HAT with learned human models as well as human participants in a user study for the popular Overcooked domain, and evaluate the task reward and teaming performance for these human-agent teams. Our results demonstrate that although trained agents attain high task rewards, they fail to induce cooperative behavior, showing very low levels of interdependence across teams. Furthermore, our analysis reveals that teaming performance is not necessarily correlated with task reward, highlighting that task reward alone cannot reliably measure cooperation arising in a team.
Multi-agent reinforcement learning (MARL) often relies on \emph{parameter sharing (PS)} to scale efficiently. However, purely shared policies can stifle each agent's unique specialization, reducing overall performance in heterogeneous environments. We propose \textbf{Low-Rank Agent-Specific Adaptation (LoRASA)}, a novel approach that treats each agent's policy as a specialized ``task'' fine-tuned from a shared backbone. Drawing inspiration from parameter-efficient transfer methods, LoRASA appends small, low-rank adaptation matrices to each layer of the shared policy, naturally inducing \emph{parameter-space sparsity} that promotes both specialization and scalability. We evaluate LoRASA on challenging benchmarks including the StarCraft Multi-Agent Challenge (SMAC) and Multi-Agent MuJoCo (MAMuJoCo), implementing it atop widely used algorithms such as MAPPO and A2PO. Across diverse tasks, LoRASA matches or outperforms existing baselines \emph{while reducing memory and computational overhead}. Ablation studies on adapter rank, placement, and timing validate the method's flexibility and efficiency. Our results suggest LoRASA's potential to establish a new norm for MARL policy parameterization: combining a shared foundation for coordination with low-rank agent-specific refinements for individual specialization.
Communicating in natural language is a powerful tool in multi-agent settings, as it enables independent agents to share information in partially observable settings and allows zero-shot coordination with humans. However, most prior works are limited as they either rely on training with large amounts of human demonstrations or lack the ability to generate natural and useful communication strategies. In this work, we train language models to have productive discussions about their environment in natural language without any human demonstrations. We decompose the communication problem into listening and speaking. Our key idea is to leverage the agent's goal to predict useful information about the world as a dense reward signal that guides communication. Specifically, we improve a model's listening skills by training them to predict information about the environment based on discussions, and we simultaneously improve a model's speaking skills with multi-agent reinforcement learning by rewarding messages based on their influence on other agents. To investigate the role and necessity of communication in complex social settings, we study an embodied social deduction game based on Among Us, where the key question to answer is the identity of an adversarial imposter. We analyze emergent behaviors due to our technique, such as accusing suspects and providing evidence, and find that it enables strong discussions, doubling the win rates compared to standard RL. We release our code and models at https://socialdeductionllm.github.io/
Differential Case Marking (DCM) refers to the phenomenon where grammatical case marking is applied selectively based on semantic, pragmatic, or other factors. The emergence of DCM has been studied in artificial language learning experiments with human participants, which were specifically aimed at disentangling the effects of learning from those of communication (Smith & Culbertson, 2020). Multi-agent reinforcement learning frameworks based on neural networks have gained significant interest to simulate the emergence of human-like linguistic phenomena. In this study, we employ such a framework in which agents first acquire an artificial language before engaging in communicative interactions, enabling direct comparisons to human result. Using a very generic communication optimization algorithm and neural-network learners that have no prior experience with language or semantic preferences, our results demonstrate that learning alone does not lead to DCM, but when agents communicate, differential use of markers arises. This supports Smith and Culbertson (2020)'s findings that highlight the critical role of communication in shaping DCM and showcases the potential of neural-agent models to complement experimental research on language evolution.
The advancement of Large Language Models (LLMs) and their finetuning strategies has triggered the renewed interests in multi-agent reinforcement learning. In this paper, we introduce a focal diversity-optimized multi-agent reinforcement learning approach, coined as MARL-Focal, with three unique characteristics. First, we develop an agent-fusion framework for encouraging multiple LLM based agents to collaborate in producing the final inference output for each LLM query. Second, we develop a focal-diversity optimized agent selection algorithm that can choose a small subset of the available agents based on how well they can complement one another to generate the query output. Finally, we design a conflict-resolution method to detect output inconsistency among multiple agents and produce our MARL-Focal output through reward-aware and policy-adaptive inference fusion. Extensive evaluations on five benchmarks show that MARL-Focal is cost-efficient and adversarial-robust. Our multi-agent fusion model achieves performance improvement of 5.51\% compared to the best individual LLM-agent and offers stronger robustness over the TruthfulQA benchmark. Code is available at https://github.com/sftekin/rl-focal
Multi-Agent Reinforcement Learning (MARL) has recently emerged as a significant area of research. However, MARL evaluation often lacks systematic diversity, hindering a comprehensive understanding of algorithms' capabilities. In particular, cooperative MARL algorithms are predominantly evaluated on benchmarks such as SMAC and GRF, which primarily feature team game scenarios without assessing adequately various aspects of agents' capabilities required in fully cooperative real-world tasks such as multi-robot cooperation and warehouse, resource management, search and rescue, and human-AI cooperation. Moreover, MARL algorithms are mainly evaluated on low dimensional state spaces, and thus their performance on high-dimensional (e.g., image) observations is not well-studied. To fill this gap, this paper highlights the crucial need for expanding systematic evaluation across a wider array of existing benchmarks. To this end, we conduct extensive evaluation and comparisons of well-known MARL algorithms on complex fully cooperative benchmarks, including tasks with images as agents' observations. Interestingly, our analysis shows that many algorithms, hailed as state-of-the-art on SMAC and GRF, may underperform standard MARL baselines on fully cooperative benchmarks. Finally, towards more systematic and better evaluation of cooperative MARL algorithms, we have open-sourced PyMARLzoo+, an extension of the widely used (E)PyMARL libraries, which addresses an open challenge from [TBG++21], facilitating seamless integration and support with all benchmarks of PettingZoo, as well as Overcooked, PressurePlate, Capture Target and Box Pushing.