Large Language Models (LLMs) trained with reinforcement learning and verifiable rewards have achieved strong results on complex reasoning tasks. Recent work extends this paradigm to a multi-agent setting, where a meta-thinking agent proposes plans and monitors progress while a reasoning agent executes subtasks through sequential conversational turns. Despite promising performance, we identify a critical limitation: lazy agent behavior, in which one agent dominates while the other contributes little, undermining collaboration and collapsing the setup to an ineffective single agent. In this paper, we first provide a theoretical analysis showing why lazy behavior naturally arises in multi-agent reasoning. We then introduce a stable and efficient method for measuring causal influence, helping mitigate this issue. Finally, as collaboration intensifies, the reasoning agent risks getting lost in multi-turn interactions and trapped by previous noisy responses. To counter this, we propose a verifiable reward mechanism that encourages deliberation by allowing the reasoning agent to discard noisy outputs, consolidate instructions, and restart its reasoning process when necessary. Extensive experiments demonstrate that our framework alleviates lazy agent behavior and unlocks the full potential of multi-agent framework for complex reasoning tasks. Techniques such as chain-of-thought prompting (Wei et al., 2022; Kojima et al., 2022) and structured methods like Tree-of-Thoughts and Graph-of-Thoughts (Y ao et al., 2023; Besta et al., 2024) expand the space for deliberation. More recently, multi-agent frameworks enable LLMs with specialized roles to collaborate via planning, delegation, and debate, echoing human team dynamics (Li et al., 2023; Wu et al., 2024a; Chen et al., 2023; Du et al., 2023; Y uan & Xie). To support multi-agent and multi-turn reinforcement learning, multi-turn Group Relative Preference Optimization (GRPO) (Wan et al., 2025; Shi et al., 2025; Wei et al., 2025) and its variants (Guo et al., 2025b; Zhang et al., 2025c; Ning et al., 2025; Xue et al., 2025) compute advantages and importance ratios at the turn level, enabling finer-grained optimization and more precise credit assignment. Building on this foundation, ReMA (Wan et al., 2025) introduces a multi-agent LLM reasoning framework with two specialized roles: a meta-thinking agent, which decomposes tasks, sets intermediate goals, and adapts based on feedback, and a reasoning agent, which performs step-by-step 1 The agents alternate sequentially, but since only a final outcome reward is available, ReMA computes a group advantage following GRPO (Shao et al., 2024) and uniformly assigns this trajectory-level signal to every turn in the rollout.

No-regret learning dynamics play a central role in game theory, enabling decentralized convergence to equilibrium for concepts such as Coarse Correlated Equilibrium (CCE) or Correlated Equilibrium (CE). In this work, we improve the convergence rate to CCE in general-sum Markov games, reducing it from the previously best-known rate of $\mathcal{O}(\log^5 T / T)$ to a sharper $\mathcal{O}(\log T / T)$. This matches the best known convergence rate for CE in terms of $T$, number of iterations, while also improving the dependence on the action set size from polynomial to polylogarithmic-yielding exponential gains in high-dimensional settings. Our approach builds on recent advances in adaptive step-size techniques for no-regret algorithms in normal-form games, and extends them to the Markovian setting via a stage-wise scheme that adjusts learning rates based on real-time feedback. We frame policy updates as an instance of Optimistic Follow-the-Regularized-Leader (OFTRL), customized for value-iteration-based learning. The resulting self-play algorithm achieves, to our knowledge, the fastest known convergence rate to CCE in Markov games.

The gaze of a person tends to reflect their interest. This work explores what happens when this statement is taken literally and applied to robots. Here we present a robot system that employs a moving robot head with a screen-based eye model that can direct the robot's gaze to points in physical space and present a reflection-like mirror image of the attended region on top of each eye. We conducted a user study with 33 participants, who were asked to instruct the robot to perform pick-and-place tasks, monitor the robot's task execution, and interrupt it in case of erroneous actions. Despite a deliberate lack of instructions about the role of the eyes and a very brief system exposure, participants felt more aware about the robot's information processing, detected erroneous actions earlier, and rated the user experience higher when eye-based mirroring was enabled compared to non-reflective eyes. These results suggest a beneficial and intuitive utilization of the introduced method in cooperative human-robot interaction.

Scenario-based testing is essential for validating the performance of autonomous driving (AD) systems. However, such testing is limited by the scarcity of long-tailed, safety-critical scenarios in existing datasets collected in the real world. To tackle the data issue, we propose the Adv-BMT framework, which augments real-world scenarios with diverse and realistic adversarial traffic interactions. The core component of Adv-BMT is a bidirectional motion transformer (BMT) model to perform inverse traffic motion predictions, which takes agent information in the last time step of the scenario as input, and reconstructs the traffic in the inverse of chronological order until the initial time step. The Adv-BMT framework is a two-staged pipeline: it first conducts adversarial initializations and then inverse motion predictions. Different from previous work, we do not need any collision data for pretraining, and are able to generate realistic and diverse collision interactions. Our experimental results validate the quality of generated collision scenarios by Adv-BMT: training in our augmented dataset would reduce episode collision rates by 20%. Demo and code are available at: https://metadriverse.github.io/adv-bmt/.

The success of teams in robotics, nature, and society often depends on the division of labor among diverse specialists; however, a principled explanation for when such diversity surpasses a homogeneous team is still missing. Focusing on multi-agent task allocation problems, we study this question from the perspective of reward design: what kinds of objectives are best suited for heterogeneous teams? We first consider an instantaneous, non-spatial setting where the global reward is built by two generalized aggregation operators: an inner operator that maps the $N$ agents' effort allocations on individual tasks to a task score, and an outer operator that merges the $M$ task scores into the global team reward. We prove that the curvature of these operators determines whether heterogeneity can increase reward, and that for broad reward families this collapses to a simple convexity test. Next, we ask what incentivizes heterogeneity to emerge when embodied, time-extended agents must learn an effort allocation policy. To study heterogeneity in such settings, we use multi-agent reinforcement learning (MARL) as our computational paradigm, and introduce Heterogeneity Gain Parameter Search (HetGPS), a gradient-based algorithm that optimizes the parameter space of underspecified MARL environments to find scenarios where heterogeneity is advantageous. Across different environments, we show that HetGPS rediscovers the reward regimes predicted by our theory to maximize the advantage of heterogeneity, both validating HetGPS and connecting our theoretical insights to reward design in MARL. Together, these results help us understand when behavioral diversity delivers a measurable benefit.

The number of studies that combine Evolutionary Machine Learning and self-supervised learning has been growing steadily in recent years. Evolutionary Machine Learning has been shown to help automate the design of machine learning algorithms and to lead to more reliable solutions. Self-supervised learning, on the other hand, has produced good results in learning useful features when labelled data is limited. This suggests that the combination of these two areas can help both in shaping evolutionary processes and in automating the design of deep neural networks, while also reducing the need for labelled data. Still, there are no detailed reviews that explain how Evolutionary Machine Learning and self-supervised learning can be used together. To help with this, we provide an overview of studies that bring these areas together. Based on this growing interest and the range of existing works, we suggest a new sub-area of research, which we call Evolutionary Self-Supervised Learning and introduce a taxonomy for it. Finally, we point out some of the main challenges and suggest directions for future research to help Evolutionary Self-Supervised Learning grow and mature as a field.

In contrast, using only the current turn avoids this overhead but discards essential information. This trade-off limits the scalability of search agents. To address this challenge, we propose MemSearcher, an agent workflow that iteratively maintains a compact memory and combines the current turn with it. At each turn, MemSearcher fuses the user's question with the memory to generate reasoning traces, perform search actions, and update memory to retain only information essential for solving the task. This design stabilizes context length across multi-turn interactions, improving efficiency without sacrificing accuracy. To optimize this workflow, we introduce multi-context GRPO, an end-to-end RL framework that jointly optimize reasoning, search strategies, and memory management of MemSearcher Agents. Trained on the same dataset as Search-R1, MemSearcher achieves significant improvements over strong baselines on seven public benchmarks: +11% on Qwen2.5-3B-Instruct and +12% on Qwen2.5-7B-Instruct Notably, the 3B-based Mem-Searcher even outperforms 7B-based baselines, demonstrating that striking a balance between information integrity and efficiency yields both higher accuracy and lower computational overhead.

Training a team of agents from scratch in multi-agent reinforcement learning (MARL) is highly inefficient, much like asking beginners to play a symphony together without first practicing solo. Existing methods, such as offline or transferable MARL, can ease this burden, but they still rely on costly multi-agent data, which often becomes the bottleneck. In contrast, solo experiences are far easier to obtain in many important scenarios, e.g., collaborative coding, household cooperation, and search-and-rescue. To unlock their potential, we propose Solo-to-Collaborative RL (SoCo), a framework that transfers solo knowledge into cooperative learning. SoCo first pretrains a shared solo policy from solo demonstrations, then adapts it for cooperation during multi-agent training through a policy fusion mechanism that combines an MoE-like gating selector and an action editor. Experiments across diverse cooperative tasks show that SoCo significantly boosts the training efficiency and performance of backbone algorithms. These results demonstrate that solo demonstrations provide a scalable and effective complement to multi-agent data, making cooperative learning more practical and broadly applicable.

Large language models (LLMs) exhibit complementary strengths across domains and come with varying inference costs, motivating the design of multi-agent LLM systems where specialized models collaborate efficiently. Existing approaches predominantly rely on decentralized frameworks, which invoke multiple LLMs for every input and thus lead to substantial and uncontrolled inference costs. In this work, we introduce a centralized multi-LLM framework, where a controller LLM selectively coordinates a pool of expert models in a cost-efficient and cost-controllable manner. We formulate this coordination problem as reinforcement learning with dual objectives: maximizing task performance while minimizing the overall inference cost. In addition, we expect the multi-agent system to have adapted behavior with different budget conditions during inference. To this end, we propose CoRL, a reinforcement learning framework that optimizes the performance cost trade-off in a controllable multi-budget setting. Experiments on four diverse benchmarks demonstrate that CoRL enables a single system to surpass the best expert LLM under high-budget settings, while maintaining strong performance in more economical low-budget modes, highlighting the effectiveness of centralized coordination for scalable and cost-efficient multi-agent LLM systems.

The trajectory of AI development suggests that we will increasingly rely on agent-based systems composed of independently developed agents with different information, privileges, and tools. The success of these systems will critically depend on effective collaboration among these heterogeneous agents, even under partial observability. Despite intense interest, few empirical studies have evaluated such agent-agent collaboration at scale. We propose a collaborative maze-solving benchmark that (i) isolates collaborative capabilities, (ii) modulates problem complexity, (iii) enables scalable automated grading, and (iv) imposes no output-format constraints, preserving ecological plausibility. Using this framework, we evaluate 32 leading open- and closed-source models in solo, homogeneous, and heterogeneous pairings. Our results reveal a "collaboration gap": models that perform well solo often degrade substantially when required to collaborate. Collaboration can break down dramatically; for instance, small distilled models that solve mazes well alone may fail almost completely in certain pairings. We find that starting with the stronger agent often improves outcomes, motivating a "relay inference" approach where the stronger agent leads before handing off to the weaker one, closing much of the gap. Our findings argue for (1) collaboration-aware evaluation, (2) training strategies developed to enhance collaborative capabilities, and (3) interaction design that reliably elicits agents' latent skills, guidance that applies to AI-AI and human-AI collaboration.