Adapting a single agent to a new multi-agent system brings challenges, necessitating adjustments across various tasks, environments, and interactions with unknown teammates and opponents. Addressing this challenge is highly complex, and researchers have proposed two simplified scenarios, Multi-agent reinforcement learning for zero-shot learning and Ad-Hoc Teamwork. Building on these foundations, we propose a more comprehensive setting, Agent Collaborative-Competitive Adaptation (ACCA), which evaluates an agent to generalize across diverse scenarios, tasks, and interactions with both unfamiliar opponents and teammates. In ACCA, agents adjust to task and environmental changes, collaborate with unseen teammates, and compete against unknown opponents. We introduce a new modeling approach, Multi-Retrieval and Dynamic Generation (MRDG), that effectively models both teammates and opponents using their behavioral trajectories. This method incorporates a positional encoder for varying team sizes and a hypernetwork module to boost agents' learning and adaptive capabilities. Additionally, a viewpoint alignment module harmonizes the observational perspectives of retrieved teammates and opponents with the learning agent. Extensive tests in benchmark scenarios like SMAC, Overcooked-AI, and Melting Pot show that MRDG significantly improves robust collaboration and competition with unseen teammates and opponents, surpassing established baselines. Our code is available at: https://github.com/vcis-wangchenxu/MRDG.git
Multi-agent reinforcement learning (MARL) has achieved strong performance in cooperative adversarial tasks. However, most existing methods typically train agents against fixed opponent strategies and rely on such meta-static difficulty conditions, which limits their adaptability to changing environments and often leads to suboptimal policies. Inspired by the success of curriculum learning (CL) in supervised tasks, we propose a dynamic CL framework for MARL that employs an self-adaptive difficulty adjustment mechanism. This mechanism continuously modulates opponent strength based on real-time agent training performance, allowing agents to progressively learn from easier to more challenging scenarios. However, the dynamic nature of CL introduces instability due to nonstationary environments and sparse global rewards. To address this challenge, we develop a Counterfactual Group Relative Policy Advantage (CGRPA), which is tightly coupled with the curriculum by providing intrinsic credit signals that reflect each agent's impact under evolving task demands. CGRPA constructs a counterfactual advantage function that isolates individual contributions within group behavior, facilitating more reliable policy updates throughout the curriculum. CGRPA evaluates each agent's contribution through constructing counterfactual action advantage function, providing intrinsic rewards that enhance credit assignment and stabilize learning under non-stationary conditions. Extensive experiments demonstrate that our method improves both training stability and final performance, achieving competitive results against state-of-the-art methods. The code is available at https://github.com/NICE-HKU/CL2MARL-SMAC.
Multi-agent reinforcement learning (MARL) methods have achieved state-of-the-art results on a range of multi-agent tasks. Yet, MARL algorithms typically require significantly more environment interactions than their single-agent counterparts to converge, a problem exacerbated by the difficulty in exploring over a large joint action space and the high variance intrinsic to MARL environments. To tackle these issues, we propose a novel algorithm that combines a decomposed centralized critic with decentralized ensemble learning, incorporating several key contributions. The main component in our scheme is a selective exploration method that leverages ensemble kurtosis. We extend the global decomposed critic with a diversity-regularized ensemble of individual critics and utilize its excess kurtosis to guide exploration toward high-uncertainty states and actions. To improve sample efficiency, we train the centralized critic with a novel truncated variation of the TD($\lambda$) algorithm, enabling efficient off-policy learning with reduced variance. On the actor side, our suggested algorithm adapts the mixed samples approach to MARL, mixing on-policy and off-policy loss functions for training the actors. This approach balances between stability and efficiency and outperforms purely off-policy learning. The evaluation shows our method outperforms state-of-the-art baselines on standard MARL benchmarks, including a variety of SMAC II maps.
In this paper we propose for the first time the hyperparameter optimization (HPO) algorithm POCAII. POCAII differs from the Hyperband and Successive Halving literature by explicitly separating the search and evaluation phases and utilizing principled approaches to exploration and exploitation principles during both phases. Such distinction results in a highly flexible scheme for managing a hyperparameter optimization budget by focusing on search (i.e., generating competing configurations) towards the start of the HPO process while increasing the evaluation effort as the HPO comes to an end. POCAII was compared to state of the art approaches SMAC, BOHB and DEHB. Our algorithm shows superior performance in low-budget hyperparameter optimization regimes. Since many practitioners do not have exhaustive resources to assign to HPO, it has wide applications to real-world problems. Moreover, the empirical evidence showed how POCAII demonstrates higher robustness and lower variance in the results. This is again very important when considering realistic scenarios with extremely expensive models to train.
Multi-agent reinforcement Learning (MARL) is often challenged by the sight range dilemma, where agents either receive insufficient or excessive information from their environment. In this paper, we propose a novel method, called Dynamic Sight Range Selection (DSR), to address this issue. DSR utilizes an Upper Confidence Bound (UCB) algorithm and dynamically adjusts the sight range during training. Experiment results show several advantages of using DSR. First, we demonstrate using DSR achieves better performance in three common MARL environments, including Level-Based Foraging (LBF), Multi-Robot Warehouse (RWARE), and StarCraft Multi-Agent Challenge (SMAC). Second, our results show that DSR consistently improves performance across multiple MARL algorithms, including QMIX and MAPPO. Third, DSR offers suitable sight ranges for different training steps, thereby accelerating the training process. Finally, DSR provides additional interpretability by indicating the optimal sight range used during training. Unlike existing methods that rely on global information or communication mechanisms, our approach operates solely based on the individual sight ranges of agents. This approach offers a practical and efficient solution to the sight range dilemma, making it broadly applicable to real-world complex environments.
Multi-agent reinforcement learning (MARL) has emerged as a promising solution for learning complex and scalable coordination behaviors in multi-robot systems. However, established MARL platforms (e.g., SMAC and MPE) lack robotics relevance and hardware deployment, leaving multi-robot learning researchers to develop bespoke environments and hardware testbeds dedicated to the development and evaluation of their individual contributions. The Multi-Agent RL Benchmark and Learning Environment for the Robotarium (MARBLER) is an exciting recent step in providing a standardized robotics-relevant platform for MARL, by bridging the Robotarium testbed with existing MARL software infrastructure. However, MARBLER lacks support for parallelization and GPU/TPU execution, making the platform prohibitively slow compared to modern MARL environments and hindering adoption. We contribute JaxRobotarium, a Jax-powered end-to-end simulation, learning, deployment, and benchmarking platform for the Robotarium. JaxRobotarium enables rapid training and deployment of multi-robot RL (MRRL) policies with realistic robot dynamics and safety constraints, supporting parallelization and hardware acceleration. Our generalizable learning interface integrates easily with SOTA MARL libraries (e.g., JaxMARL). In addition, JaxRobotarium includes eight standardized coordination scenarios, including four novel scenarios that bring established MARL benchmark tasks (e.g., RWARE and Level-Based Foraging) to a robotics setting. We demonstrate that JaxRobotarium retains high simulation fidelity while achieving dramatic speedups over baseline (20x in training and 150x in simulation), and provides an open-access sim-to-real evaluation pipeline through the Robotarium testbed, accelerating and democratizing access to multi-robot learning research and evaluation. Our code is available at https://github.com/GT-STAR-Lab/JaxRobotarium.
In real-world multi-agent systems (MASs), observation delays are ubiquitous, preventing agents from making decisions based on the environment's true state. An individual agent's local observation often consists of multiple components from other agents or dynamic entities in the environment. These discrete observation components with varying delay characteristics pose significant challenges for multi-agent reinforcement learning (MARL). In this paper, we first formulate the decentralized stochastic individual delay partially observable Markov decision process (DSID-POMDP) by extending the standard Dec-POMDP. We then propose the Rainbow Delay Compensation (RDC), a MARL training framework for addressing stochastic individual delays, along with recommended implementations for its constituent modules. We implement the DSID-POMDP's observation generation pattern using standard MARL benchmarks, including MPE and SMAC. Experiments demonstrate that baseline MARL methods suffer severe performance degradation under fixed and unfixed delays. The RDC-enhanced approach mitigates this issue, remarkably achieving ideal delay-free performance in certain delay scenarios while maintaining generalizability. Our work provides a novel perspective on multi-agent delayed observation problems and offers an effective solution framework. The source code is available at https://anonymous.4open.science/r/RDC-pymarl-4512/.
Learning cooperative multi-agent policy from offline multi-task data that can generalize to unseen tasks with varying numbers of agents and targets is an attractive problem in many scenarios. Although aggregating general behavior patterns among multiple tasks as skills to improve policy transfer is a promising approach, two primary challenges hinder the further advancement of skill learning in offline multi-task MARL. Firstly, extracting general cooperative behaviors from various action sequences as common skills lacks bringing cooperative temporal knowledge into them. Secondly, existing works only involve common skills and can not adaptively choose independent knowledge as task-specific skills in each task for fine-grained action execution. To tackle these challenges, we propose Hierarchical and Separate Skill Discovery (HiSSD), a novel approach for generalizable offline multi-task MARL through skill learning. HiSSD leverages a hierarchical framework that jointly learns common and task-specific skills. The common skills learn cooperative temporal knowledge and enable in-sample exploitation for offline multi-task MARL. The task-specific skills represent the priors of each task and achieve a task-guided fine-grained action execution. To verify the advancement of our method, we conduct experiments on multi-agent MuJoCo and SMAC benchmarks. After training the policy using HiSSD on offline multi-task data, the empirical results show that HiSSD assigns effective cooperative behaviors and obtains superior performance in unseen tasks.
We introduce Attentive VLM Agent (AVA), a multimodal StarCraft II agent 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 agent addresses this limitation by incorporating RGB visual inputs and natural language observations that more closely simulate human cognitive processes during gameplay. The AVA architecture 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. The experimental evaluation in our proposed AVACraft environment, which contains 21 multimodal StarCraft II scenarios, demonstrates that AVA 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.
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.
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.
Overestimation in single-agent reinforcement learning has been extensively studied. In contrast, overestimation in the multiagent setting has received comparatively little attention although it increases with the number of agents and leads to severe learning instability. Previous works concentrate on reducing overestimation in the estimation process of target Q-value. They ignore the follow-up optimization process of online Q-network, thus making it hard to fully address the complex multiagent overestimation problem. To solve this challenge, in this study, we first establish an iterative estimation-optimization analysis framework for multiagent value-mixing Q-learning. Our analysis reveals that multiagent overestimation not only comes from the computation of target Q-value but also accumulates in the online Q-network's optimization. Motivated by it, we propose the Dual Ensembled Multiagent Q-Learning with Hypernet Regularizer algorithm to tackle multiagent overestimation from two aspects. First, we extend the random ensemble technique into the estimation of target individual and global Q-values to derive a lower update target. Second, we propose a novel hypernet regularizer on hypernetwork weights and biases to constrain the optimization of online global Q-network to prevent overestimation accumulation. Extensive experiments in MPE and SMAC show that the proposed method successfully addresses overestimation across various tasks.
Inferring reward functions from demonstrations is a key challenge in reinforcement learning (RL), particularly in multi-agent RL (MARL), where large joint state-action spaces and complex inter-agent interactions complicate the task. While prior single-agent studies have explored recovering reward functions and policies from human preferences, similar work in MARL is limited. Existing methods often involve separate stages of supervised reward learning and MARL algorithms, leading to unstable training. In this work, we introduce a novel end-to-end preference-based learning framework for cooperative MARL, leveraging the underlying connection between reward functions and soft Q-functions. Our approach uses a carefully-designed multi-agent value decomposition strategy to improve training efficiency. Extensive experiments on SMAC and MAMuJoCo benchmarks show that our algorithm outperforms existing methods across various tasks.
Recent studies have shown that cooperative multi-agent deep reinforcement learning (c-MADRL) is under the threat of backdoor attacks. Once a backdoor trigger is observed, it will perform malicious actions leading to failures or malicious goals. However, existing backdoor attacks suffer from several issues, e.g., instant trigger patterns lack stealthiness, the backdoor is trained or activated by an additional network, or all agents are backdoored. To this end, in this paper, we propose a novel backdoor leverage attack against c-MADRL, BLAST, which attacks the entire multi-agent team by embedding the backdoor only in a single agent. Firstly, we introduce adversary spatiotemporal behavior patterns as the backdoor trigger rather than manual-injected fixed visual patterns or instant status and control the period to perform malicious actions. This method can guarantee the stealthiness and practicality of BLAST. Secondly, we hack the original reward function of the backdoor agent via unilateral guidance to inject BLAST, so as to achieve the \textit{leverage attack effect} that can pry open the entire multi-agent system via a single backdoor agent. We evaluate our BLAST against 3 classic c-MADRL algorithms (VDN, QMIX, and MAPPO) in 2 popular c-MADRL environments (SMAC and Pursuit), and 2 existing defense mechanisms. The experimental results demonstrate that BLAST can achieve a high attack success rate while maintaining a low clean performance variance rate.
The availability of challenging simulation environments is pivotal for advancing the field of Multi-Agent Reinforcement Learning (MARL). In cooperative MARL settings, the StarCraft Multi-Agent Challenge (SMAC) has gained prominence as a benchmark for algorithms following centralized training with decentralized execution paradigm. However, with continual advancements in SMAC, many algorithms now exhibit near-optimal performance, complicating the evaluation of their true effectiveness. To alleviate this problem, in this work, we highlight a critical issue: the default opponent policy in these environments lacks sufficient diversity, leading MARL algorithms to overfit and exploit unintended vulnerabilities rather than learning robust strategies. To overcome these limitations, we propose SMAC-HARD, a novel benchmark designed to enhance training robustness and evaluation comprehensiveness. SMAC-HARD supports customizable opponent strategies, randomization of adversarial policies, and interfaces for MARL self-play, enabling agents to generalize to varying opponent behaviors and improve model stability. Furthermore, we introduce a black-box testing framework wherein agents are trained without exposure to the edited opponent scripts but are tested against these scripts to evaluate the policy coverage and adaptability of MARL algorithms. We conduct extensive evaluations of widely used and state-of-the-art algorithms on SMAC-HARD, revealing the substantial challenges posed by edited and mixed strategy opponents. Additionally, the black-box strategy tests illustrate the difficulty of transferring learned policies to unseen adversaries. We envision SMAC-HARD as a critical step toward benchmarking the next generation of MARL algorithms, fostering progress in self-play methods for multi-agent systems. Our code is available at https://github.com/devindeng94/smac-hard.
This paper introduces a new environment LLM-PySC2 (the Large Language Model StarCraft II Learning Environment), a platform derived from DeepMind's StarCraft II Learning Environment that serves to develop Large Language Models (LLMs) based decision-making methodologies. This environment is the first to offer the complete StarCraft II action space, multi-modal observation interfaces, and a structured game knowledge database, which are seamlessly connected with various LLMs to facilitate the research of LLMs-based decision-making. To further support multi-agent research, we developed an LLM collaborative framework that supports multi-agent concurrent queries and multi-agent communication. In our experiments, the LLM-PySC2 environment is adapted to be compatible with the StarCraft Multi-Agent Challenge (SMAC) task group and provided eight new scenarios focused on macro-decision abilities. We evaluated nine mainstream LLMs in the experiments, and results show that sufficient parameters are necessary for LLMs to make decisions, but improving reasoning ability does not directly lead to better decision-making outcomes. Our findings further indicate the importance of enabling large models to learn autonomously in the deployment environment through parameter training or train-free learning techniques. Ultimately, we expect that the LLM-PySC2 environment can promote research on learning methods for LLMs, helping LLM-based methods better adapt to task scenarios.
Offline-to-Online Reinforcement Learning has emerged as a powerful paradigm, leveraging offline data for initialization and online fine-tuning to enhance both sample efficiency and performance. However, most existing research has focused on single-agent settings, with limited exploration of the multi-agent extension, i.e., Offline-to-Online Multi-Agent Reinforcement Learning (O2O MARL). In O2O MARL, two critical challenges become more prominent as the number of agents increases: (i) the risk of unlearning pre-trained Q-values due to distributional shifts during the transition from offline-to-online phases, and (ii) the difficulty of efficient exploration in the large joint state-action space. To tackle these challenges, we propose a novel O2O MARL framework called Offline Value Function Memory with Sequential Exploration (OVMSE). First, we introduce the Offline Value Function Memory (OVM) mechanism to compute target Q-values, preserving knowledge gained during offline training, ensuring smoother transitions, and enabling efficient fine-tuning. Second, we propose a decentralized Sequential Exploration (SE) strategy tailored for O2O MARL, which effectively utilizes the pre-trained offline policy for exploration, thereby significantly reducing the joint state-action space to be explored. Extensive experiments on the StarCraft Multi-Agent Challenge (SMAC) demonstrate that OVMSE significantly outperforms existing baselines, achieving superior sample efficiency and overall performance.
StarCraft Multi-Agent Challenge (SMAC) is one of the most commonly used experimental environments in multi-agent reinforcement learning (MARL), where the specific task is to control a set number of allied units to defeat enemy forces. Traditional MARL algorithms often require interacting with the environment for up to 1 million steps to train a model, and the resulting policies are typically non-interpretable with weak transferability. In this paper, we propose a novel approach to solving SMAC tasks called LLM-SMAC. In our framework, agents leverage large language models (LLMs) to generate decision tree code by providing task descriptions. The model is further self-reflection using feedback from the rewards provided by the environment. We conduct experiments in the SMAC and demonstrate that our method can produce high-quality, interpretable decision trees with minimal environmental exploration. Moreover, these models exhibit strong transferability, successfully applying to similar SMAC environments without modification. We believe this approach offers a new direction for solving decision-making tasks in the future.
This paper proposes a safety modulator actor-critic (SMAC) method to address safety constraint and overestimation mitigation in model-free safe reinforcement learning (RL). A safety modulator is developed to satisfy safety constraints by modulating actions, allowing the policy to ignore safety constraint and focus on maximizing reward. Additionally, a distributional critic with a theoretical update rule for SMAC is proposed to mitigate the overestimation of Q-values with safety constraints. Both simulation and real-world scenarios experiments on Unmanned Aerial Vehicles (UAVs) hovering confirm that the SMAC can effectively maintain safety constraints and outperform mainstream baseline algorithms.
With expansive state-action spaces, efficient multi-agent exploration remains a longstanding challenge in reinforcement learning. Although pursuing novelty, diversity, or uncertainty attracts increasing attention, redundant efforts brought by exploration without proper guidance choices poses a practical issue for the community. This paper introduces a systematic approach, termed LEMAE, choosing to channel informative task-relevant guidance from a knowledgeable Large Language Model (LLM) for Efficient Multi-Agent Exploration. Specifically, we ground linguistic knowledge from LLM into symbolic key states, that are critical for task fulfillment, in a discriminative manner at low LLM inference costs. To unleash the power of key states, we design Subspace-based Hindsight Intrinsic Reward (SHIR) to guide agents toward key states by increasing reward density. Additionally, we build the Key State Memory Tree (KSMT) to track transitions between key states in a specific task for organized exploration. Benefiting from diminishing redundant explorations, LEMAE outperforms existing SOTA approaches on the challenging benchmarks (e.g., SMAC and MPE) by a large margin, achieving a 10x acceleration in certain scenarios.