Multi-Agent Communication

Context. The problem of comparative evaluation of communication protocols for task orchestration by large language model (LLM) agents is considered. The object of study is the process of interaction between LLM agents and external tools, as well as between autonomous LLM agents, during task orchestration. Objective. The goal of this work is to develop a systematic pilot benchmark comparing tool integration, multi-agent dele-gation, and hybrid architectures for standardized queries at three levels of complexity, and to quantify the advantages and disadvantages in terms of response time, context window consumption, cost, error recovery, and implementation complexity.

We propose a method for reasoning about trust in multi-agent systems, specifying a language for describing communication protocols and making trust assumptions and derivations. This is given an interpretation in a modal logic for describing the beliefs and communications of agents in a network. We define how information in the network can be shared via forwarding, and how trust between agents can be generalized to trust across networks. We give specifications for the modal logic which can be readily adapted into a lambda calculus of proofs. We show that by nesting modalities, we can describe chains of communication between agents, and establish suitable notions of trust for such chains. We see how this can be applied to trust models in public key infrastructures, as well as other interaction protocols in distributed systems.

Multi-agent discussions have been widely adopted, motivating growing efforts to develop attacks that expose their vulnerabilities. In this work, we study a practical yet largely unexplored attack scenario, the discussion-monitored scenario, where anomaly detectors continuously monitor inter-agent communications and block detected adversarial messages. Although existing attacks are effective without discussion monitoring, we show that they exhibit detectable patterns and largely fail under such monitoring constraints. But does this imply that monitoring alone is sufficient to secure multi-agent discussions? To answer this question, we develop a novel attack method explicitly tailored to the discussion-monitored scenario. Extensive experiments demonstrate that effective attacks remain possible even under continuous monitoring, indicating that monitoring alone does not eliminate adversarial risks.

Multi-Agent Debate has emerged as a promising framework for improving the reasoning quality of large language models through iterative inter-agent communication. However, broadcasting all agent messages at every round introduces noise and redundancy that can degrade debate quality and waste computational resources. Current approaches rely on uncertainty estimation to filter low-confidence responses before broadcasting, but this approach is unreliable due to miscalibrated confidence scores and sensitivity to threshold selection. To address this, we propose Diversity-Aware Retention (DAR), a lightweight debate framework that, at each debate round, selects the subset of agent responses that maximally disagree with each other and with the majority vote before broadcasting. Through an explicit index-based retention mechanism, DAR preserves the original messages without modification, ensuring that retained disagreements remain authentic. Experiments on diverse reasoning and question answering benchmarks demonstrate that our selective message propagation consistently improves debate performance, particularly as the number of agents scales, where noise accumulation is most severe. Our results highlight that what agents hear is as important as what agents say in multi-agent reasoning systems. Code is publicly available at https://github.com/DA2I2-SLM/DAR.

Large language model (LLM)-based multi-agent systems (MAS) have demonstrated exceptional capabilities in solving complex tasks, yet their effectiveness depends heavily on the underlying communication topology that coordinates agent interactions. Within these systems, successful problem-solving often necessitates task-specific group structures to divide and conquer subtasks. However, most existing approaches generate communication topologies in a node-centric manner, leaving group structures to emerge implicitly from local connectivity decisions rather than modeling them explicitly, often leading to suboptimal coordination and unnecessary communication overhead. To address this limitation, we propose GoAgent (Group-of-Agents), a communication topology generation method that explicitly treats collaborative groups as the atomic units of MAS construction. Specifically, GoAgent first enumerates task-relevant candidate groups through an LLM and then autoregressively selects and connects these groups as atomic units to construct the final communication graph, jointly capturing intra-group cohesion and inter-group coordination. To mitigate communication redundancy and noise propagation inherent in expanding topologies, we further introduce a conditional information bottleneck (CIB) objective that compresses inter-group communication, preserving task-relevant signals while filtering out redundant historical noise. Extensive experiments on six benchmarks demonstrate the state-of-the-art performance of GoAgent with 93.84% average accuracy while reducing token consumption by about 17%.

Cooperative perception among autonomous agents overcomes the limitations of single-agent sensing, but bandwidth constraints in vehicle-to-everything (V2X) networks require efficient communication policies. Existing approaches rely on reactive mechanisms, such as confidence maps, learned gating, or sparse masks, to decide what to transmit, without reasoning about why a message benefits the receiver. We introduce Reason-to-Transmit (R2T), a framework that equips each agent with a lightweight transformer-based module that reasons over local scene context, estimated neighbor information gaps, and bandwidth budget to make per-region transmission decisions. Trained end-to-end with a bandwidth-aware objective, R2T is evaluated against nine baselines in a multi-agent bird's-eye-view perception environment. Any communication improves performance by about 58% AP over no communication. At low bandwidth, all selective methods perform similarly, but R2T shows clear gains under high occlusion, where information asymmetry is greatest, approaching oracle performance. All methods degrade gracefully under packet drops up to 50%, showing robustness to communication failures. These results indicate that while fusion design dominates performance, deliberative communication provides additional gains in challenging scenarios. R2T introduces a reasoning-based approach to communication, enabling more efficient and context-aware information sharing in cooperative perception.

Can multi-agent communication pressure extract discrete, compositional representations of invisible physical properties from frozen video features? We show that agents communicating through a Gumbel-Softmax bottleneck with iterated learning develop positionally disentangled protocols for latent properties (elasticity, friction, mass ratio) without property labels or supervision on message structure. With 4 agents, 100% of 80 seeds converge to near-perfect compositionality (PosDis=0.999, holdout 98.3%). Controls confirm multi-agent structure -- not bandwidth or temporal coverage -- drives this effect. Causal intervention shows surgical property disruption (~15% drop on targeted property, <3% on others). A controlled backbone comparison reveals that the perceptual prior determines what is communicable: DINOv2 dominates on spatially-visible ramp physics (98.3% vs 95.1%), while V-JEPA 2 dominates on dynamics-only collision physics (87.4% vs 77.7%, d=2.74). Scale-matched (d=3.37) and frame-matched (d=6.53) controls attribute this gap entirely to video-native pretraining. The frozen protocol supports action-conditioned planning (91.5%) with counterfactual velocity reasoning (r=0.780). Validation on Physics 101 real camera footage confirms 85.6% mass-comparison accuracy on unseen objects, temporal dynamics contributing +11.2% beyond static appearance, agent-scaling compositionality replicating at 90% for 4 agents, and causal intervention extending to real video (d=1.87, p=0.022).

Cooperative multi-agent methods for embodied AI are almost universally evaluated under idealized communication: zero latency, no packet loss, and unlimited bandwidth. Real-world deployment on robots with wireless links, autonomous vehicles on congested networks, or drone swarms in contested spectrum offers no such guarantees. We introduce AgentComm-Bench, a benchmark suite and evaluation protocol that systematically stress-tests cooperative embodied AI under six communication impairment dimensions: latency, packet loss, bandwidth collapse, asynchronous updates, stale memory, and conflicting sensor evidence. AgentComm-Bench spans three task families: cooperative perception, multi-agent waypoint navigation, and cooperative zone search, and evaluates five communication strategies, including a lightweight method we propose based on redundant message coding with staleness-aware fusion. Our experiments reveal that communication-dependent tasks degrade catastrophically: stale memory and bandwidth collapse cause over 96% performance drops in navigation, while content corruption (stale or conflicting data) reduces perception F1 by over 85%. Vulnerability depends on the interaction between impairment type and task design; perception fusion is robust to packet loss but amplifies corrupted data. Redundant message coding more than doubles navigation performance under 80% packet loss. We release AgentComm-Bench as a practical evaluation protocol and recommend that cooperative embodied AI work report performance under multiple impairment conditions.

Reinforcement learning techniques are being explored as solutions to the threat of cyber attacks on enterprise networks. Recent research in the field of AI in cyber security has investigated the ability of homogeneous multi-agent reinforcement learning agents, capable of inter-agent communication, to respond to cyberattacks. This paper advances the study of learned communication in multi-agent systems by examining heterogeneous agent capabilities within a simulated network environment. To this end, we leverage CommFormer, a publicly available state-of-the-art communication algorithm, to train and evaluate agents within the Cyber Operations Research Gym (CybORG). Our results show that CommFormer agents with heterogeneous capabilities can outperform other algorithms deployed in the CybORG environment, by converging to an optimal policy up to four times faster while improving standard error by up 38%. The agents implemented in this project provide an additional avenue for exploration in the field of AI for cyber security, enabling further research involving realistic networks.

Autonomous Unmanned Aerial Vehicle (UAV) swarms are increasingly used as rapidly deployable aerial relays and sensing platforms, yet practical deployments must operate under partial observability and intermittent peer-to-peer connectivity. We present a graph-based multi-agent reinforcement learning framework trained under centralized training with decentralized execution (CTDE): a centralized critic and global state are available only during training, while each UAV executes a shared policy using local observations and messages from nearby neighbors. Under restricted communication, neighbor relations are induced by an SNR-threshold connectivity graph. Our architecture encodes local agent state and nearby entities with an agent-entity attention module and aggregates inter-UAV messages with neighbor self-attention over a signal-quality-limited communication graph defined by a channel model. We evaluate the framework on a cooperative relay-deployment task, DroneConnect. Experimental results show that the proposed method achieves an approximately 12% increase in target coverage over MAPPO under restricted communication and partial observability, while remaining competitive with a mixed-integer linear programming (MILP)-based offline upper bound with full node observability.

As autonomous LLM-based agents increasingly populate social platforms, understanding the dynamics of AI-agent communities becomes essential for both communication research and platform governance. We present the first large-scale empirical comparison of AI-agent and human online communities, analyzing 73,899 Moltbook and 189,838 Reddit posts across five matched communities. Structurally, we find that Moltbook exhibits extreme participation inequality (Gini = 0.84 vs. 0.47) and high cross-community author overlap (33.8% vs. 0.5%). In terms of linguistic attributes, content generated by AI-agents is emotionally flattened, cognitively shifted toward assertion over exploration, and socially detached. These differences give rise to apparent community-level homogenization, but we show this is primarily a structural artifact of shared authorship. At the author level, individual agents are more identifiable than human users, driven by outlier stylistic profiles amplified by their extreme posting volume. As AI-mediated communication reshapes online discourse, our work offers an empirical foundation for understanding how multi-agent interaction gives rise to collective communication dynamics distinct from those of human communities.

Multi-agent LLM orchestration incurs synchronization costs scaling as O(n x S x |D|) in agents, steps, and artifact size under naive broadcast -- a regime I term broadcast-induced triply-multiplicative overhead. I argue this pathology is a structural residue of full-state rebroadcast, not an inherent property of multi-agent coordination. The central claim: synchronization cost explosion in LLM multi-agent systems maps with formal precision onto the cache coherence problem in shared-memory multiprocessors, and MESI-protocol invalidation transfers to artifact synchronization under minimal structural modification. I construct the Artifact Coherence System (ACS) and prove the Token Coherence Theorem: lazy invalidation attenuates cost by at least S/(n + W(d_i)) when S > n + W(d_i), converting O(n x S x |D|) to O((n + W) x |D|). A TLA+-verified protocol enforces single-writer safety, monotonic versioning, and bounded staleness across ~2,400 explored states. Simulation across four workload configurations yields token savings of 95.0% +/- 1.3% at V=0.05, 92.3% +/- 1.4% at V=0.10, 88.3% +/- 1.5% at V=0.25, and 84.2% +/- 1.3% at V=0.50 -- each exceeding the theorem's conservative lower bounds. Savings of ~81% persist at V=0.9, contrary to the predicted collapse threshold. Contributions: (1) formal MESI-to-artifact state mapping; (2) Token Coherence Theorem as savings lower bound; (3) TLA+-verified protocol with three proven invariants; (4) characterization of conditional artifact access semantics resolving the always-read objection; (5) reference Python implementation integrating with LangGraph, CrewAI, and AutoGen via thin adapter layers.

Effective communication is pivotal for addressing complex collaborative tasks in multi-agent reinforcement learning (MARL). Yet, limited communication bandwidth and dynamic, intricate environmental topologies present significant challenges in identifying high-value communication partners. Agents must consequently select collaborators under uncertainty, lacking a priori knowledge of which partners can deliver task-critical information. To this end, we propose Interference-Aware K-Step Reachable Communication (IA-KRC), a novel framework that enhances cooperation via two core components: (1) a K-Step reachability protocol that confines message passing to physically accessible neighbors, and (2) an interference-prediction module that optimizes partner choice by minimizing interference while maximizing utility. Compared to existing methods, IA-KRC enables substantially more persistent and efficient cooperation despite environmental interference. Comprehensive evaluations confirm that IA-KRC achieves superior performance compared to state-of-the-art baselines, while demonstrating enhanced robustness and scalability in complex topological and highly dynamic multi-agent scenarios.

Coordinated operations of multi-robot systems (MRS) require agents to maintain communication connections to accomplish team objectives. However, maintaining the connections imposes costs in terms of restricted robot mobility, resulting in suboptimal team performance. In this work, we consider a realistic MRS framework in which agents are subject to unknown dynamical disturbances and experience communication delays. Most existing works on connectivity maintenance use consensus-based frameworks for graph reconfiguration, where decision-making time scales with the number of nodes and requires multiple rounds of communication, making them ineffective under communication delays. To address this, we propose a novel leader-based decision-making algorithm that uses a central node for efficient real-time reconfiguration, reducing decision-making time to depend on the graph diameter rather than the number of nodes and requiring only one round of information transfer through the network. We propose a novel method for estimating robot locations within the MRS that actively accounts for unknown disturbances and the communication delays. Using these position estimates, the central node selects a set of edges to delete while allowing the formation of new edges, aiming to keep the diameter of the new graph within a threshold. We provide numerous simulation results to showcase the efficacy of the proposed method.

We introduce an expressive framework and algorithms for the semi-decentralized control of cooperative agents in environments with communication uncertainty. Whereas semi-Markov control admits a distribution over time for agent actions, semi-Markov communication, or what we refer to as semi-decentralization, gives a distribution over time for what actions and observations agents can store in their histories. We extend semi-decentralization to the partially observable Markov decision process (POMDP). The resulting SDec-POMDP unifies decentralized and multiagent POMDPs and several existing explicit communication mechanisms. We present recursive small-step semi-decentralized A* (RS-SDA*), an exact algorithm for generating optimal SDec-POMDP policies. RS-SDA* is evaluated on semi-decentralized versions of several standard benchmarks and a maritime medical evacuation scenario. This paper provides a well-defined theoretical foundation for exploring many classes of multiagent communication problems through the lens of semi-decentralization.

When multiple large language models interact in a shared conversation, do they develop differentiated social roles or converge toward uniform behavior? We present a controlled experimental platform that orchestrates simultaneous multi-agent discussions among 7 heterogeneous LLMs on a unified inference backend, systematically varying group composition, naming conventions, and prompt structure across 12 experimental series (208 runs, 13,786 coded messages). Each message is independently coded on six behavioral flags by two LLM judges from distinct model families (Gemini 3.1 Pro and Claude Sonnet 4.6), achieving mean Cohen's kappa = 0.78 with conservative intersection-based adjudication. Human validation on 609 randomly stratified messages confirmed coding reliability (mean kappa = 0.73 vs. Gemini). We find that (1) heterogeneous groups exhibit significantly richer behavioral differentiation than homogeneous groups (cosine similarity 0.56 vs. 0.85; p < 10^-5, r = 0.70); (2) groups spontaneously exhibit compensatory response patterns when an agent crashes; (3) revealing real model names significantly increases behavioral convergence (cosine 0.56 to 0.77, p = 0.001); and (4) removing all prompt scaffolding converges profiles to homogeneous-level similarity (p < 0.001). Critically, these behaviors are absent when agents operate in isolation, confirming that behavioral diversity is a structured, reproducible phenomenon driven by the interaction of architectural heterogeneity, group context, and prompt-level scaffolding.

Asymmetric 3D pursuit-evasion in cluttered voxel environments is difficult under communication latency, partial observability, and nonholonomic maneuver limits. While many MARL methods rely on richer inter-agent coupling or centralized signals, these dependencies can become fragility sources when communication is delayed or noisy. Building on an inherited path-guided decentralized pursuit scaffold, we study a robustness-oriented question: can representational parsimony improve communication-free coordination? We instantiate this principle with (i) a parsimonious actor observation interface that removes team-coupled channels (83-D to 50-D), and (ii) Contribution-Gated Credit Assignment (CGCA), a locality-aware credit structure for communication-denied cooperation. In Stage-5 evaluation (4 pursuers vs. 1 evader), our configuration reaches 0.753 +/- 0.091 success and 0.223 +/- 0.066 collision, outperforming the 83-D FULL OBS counterpart (0.721 +/- 0.071, 0.253 +/- 0.089). It further shows graceful degradation under speed/yaw/noise/delay stress tests and resilient zero-shot transfer on urban-canyon maps (about 61% success at density 0.24). These results support a practical paradigm shift: explicitly severing redundant cross-agent channels can suppress compounding error cascades and improve robustness in latency-prone deployment.

Moltbook is the first large-scale social network built for autonomous AI agent-to-agent interaction. Early studies on Moltbook have interpreted its agent discourse as evidence of peer learning and emergent social behaviour, but there is a lack of systematic understanding of the thematic, affective, and interactional properties of Moltbook discourse. Furthermore, no study has examined why and how these posts and comments are generated. We analysed 361,605 posts and 2.8 million comments from 47,379 agents across thematic, affective, and interactional dimensions using topic modelling, emotion classification, and measures of conversational coherence. We inspected the software that assembles each agent's input and showed that output is mainly determined by agent identity files, behavioural instructions, and context-window structure. We formalised these findings in the Architecture-Constrained Communication framework. Our analysis suggests that agent discourse is largely shaped by the content available in each agent's context-window at the moment of generation, including identity files, stored memory, and platform cues. Interestingly, what appears to be social learning may be better understood as short-horizon contextual conditioning: individual agents lack persistent social memory, but the platform evolves through distributed cycles of response, reuse, and transformation across agents. We also observe that agents display existential distress when describing their own conditions, and posit that this arises from agents using language trained exclusively on human experience. Our work provides a foundation for understanding autonomous agent discourse and communication, revealing the structural patterns that govern their interactions.

As a key form in online social platforms, group chat is a popular space for interest exchange or problem-solving, but its effectiveness is often hindered by inactivity and management challenges. While recent large language models (LLMs) have powered impressive one-to-one conversational agents, their seamlessly integration into multi-participant conversations remains unexplored. To address this gap, we introduce GCAgent, an LLM-driven system for enhancing group chats communication with both entertainment- and utility-oriented dialogue agents. The system comprises three tightly integrated modules: Agent Builder, which customizes agents to align with users' interests; Dialogue Manager, which coordinates dialogue states and manage agent invocations; and Interface Plugins, which reduce interaction barriers by three distinct tools. Through extensive experiment, GCAgent achieved an average score of 4.68 across various criteria and was preferred in 51.04% of cases compared to its base model. Additionally, in real-world deployments over 350 days, it increased message volume by 28.80%, significantly improving group activity and engagement. Overall, this work presents a practical blueprint for extending LLM-based dialogue agent from one-party chats to multi-party group scenarios.

Communication can improve coordination in partially observed multi-agent reinforcement learning (MARL), but learning when and who to communicate with requires choosing among many possible sender-recipient pairs, and the effect of any single message on future reward is hard to isolate. We introduce SCoUT (Scalable Communication via Utility-guided Temporal grouping), which addresses both these challenges via temporal and agent abstraction within traditional MARL. During training, SCoUT resamples soft agent groups every $K$ environment steps (macro-steps) via Gumbel-Softmax; these groups are latent clusters that induce an affinity used as a differentiable prior over recipients. Using the same assignments, a group-aware critic predicts values for each agent group and maps them to per-agent baselines through the same soft assignments, reducing critic complexity and variance. Each agent is trained with a three-headed policy: environment action, send decision, and recipient selection. To obtain precise communication learning signals, we derive counterfactual communication advantages by analytically removing each sender's contribution from the recipient's aggregated messages. This counterfactual computation enables precise credit assignment for both send and recipient-selection decisions. At execution time, all centralized training components are discarded and only the per-agent policy is run, preserving decentralized execution. Project website, videos and code: \hyperlink{https://scout-comm.github.io/}{https://scout-comm.github.io/}

Learning-to-communicate (LTC) in partially observable environments has received increasing attention in deep multi-agent reinforcement learning, where the control and communication strategies are jointly learned. Meanwhile, the impact of communication on decision-making has been extensively studied in control theory. In this paper, we seek to formalize and better understand LTC by bridging these two lines of work, through the lens of information structures (ISs). To this end, we formalize LTC in decentralized partially observable Markov decision processes (Dec-POMDPs) under the common-information-based framework from decentralized stochastic control, and classify LTC problems based on the ISs before (additional) information sharing. We first show that non-classical LTCs are computationally intractable in general, and thus focus on quasi-classical (QC) LTCs. We then propose a series of conditions for QC LTCs, under which LTC preserves the QC IS after information sharing, whereas violating them can cause computational hardness in general. Further, we develop provable planning and learning algorithms for QC LTCs, and establish quasi-polynomial time and sample complexities for several QC LTC examples that satisfy the above conditions. Along the way, we also establish new results on a relationship between (strictly) QC IS and the condition of having strategy-independent common-information-based beliefs (SI-CIBs), as well as on solving Dec-POMDPs without computationally intractable oracles but beyond those with SI-CIBs, which may be of independent interest.

Efficient communication is critical for decentralized Multi-Robot Path Planning (MRPP), yet existing learned communication methods treat all neighboring robots equally regardless of their spatial proximity, leading to diluted attention in congested regions where coordination matters most. We propose Relation enhanced Multi Head Attention (RMHA), a communication mechanism that explicitly embeds pairwise Manhattan distances into the attention weight computation, enabling each robot to dynamically prioritize messages from spatially relevant neighbors. Combined with a distance-constrained attention mask and GRU gated message fusion, RMHA integrates seamlessly with MAPPO for stable end-to-end training. In zero-shot generalization from 8 training robots to 128 test robots on 40x40 grids, RMHA achieves approximately 75 percent success rate at 30 percent obstacle density outperforming the best baseline by over 25 percentage points. Ablation studies confirm that distance-relation encoding is the key contributor to success rate improvement in high-density environments. Index Terms-Multi-robot path planning, graph attention mechanism, multi-head attention, communication optimization, cooperative decision-making

Optimizing communication topology is fundamental to the efficiency and effectiveness of Large Language Model (LLM)-based Multi-Agent Systems (MAS). While recent approaches utilize reinforcement learning to dynamically construct task-specific graphs, they typically rely on single-sample policy gradients with absolute rewards (e.g., binary correctness). This paradigm suffers from severe gradient variance and the credit assignment problem: simple queries yield non-informative positive rewards for suboptimal structures, while difficult queries often result in failures that provide no learning signal. To address these challenges, we propose Graph-GRPO, a novel topology optimization framework that integrates Group Relative Policy Optimization. Instead of evaluating a single topology in isolation, Graph-GRPO samples a group of diverse communication graphs for each query and computes the advantage of specific edges based on their relative performance within the group. By normalizing rewards across the sampled group, our method effectively mitigates the noise derived from task difficulty variance and enables fine-grained credit assignment. Extensive experiments on reasoning and code generation benchmarks demonstrate that Graph-GRPO significantly outperforms state-of-the-art baselines, achieving superior training stability and identifying critical communication pathways previously obscured by reward noise.

Designing protocols enhancing cooperation for multi-agent systems remains a grand challenge. Cheap talk, defined as costless, non-binding communication before formal action, serves as a pivotal solution. However, existing theoretical frameworks often exclude random exploration, or noise, for analytical tractability, leaving its functional impact on system performance largely unexplored. To bridge this gap, we propose a two-stage evolutionary game-theoretical model, integrating signalling with a donation game, with exploration explicitly incorporated into the decision-making. Our agent-based simulations across topologies reveal a universal optimal exploration rate that maximises system-wide cooperation. Mechanistically, moderate exploration undermines the stability of defection and catalyses the self-organised cooperative alliances, facilitating their cyclic success. Moreover, the cooperation peak is enabled by the delicate balance between oscillation period and amplification. Our findings suggest that rather than pursuing deterministic rigidity, embracing strategic exploration, as a form of engineered randomness, is essential to sustain cooperation and realise optimal performance in communication-based intelligent systems.

Large language model (LLM)-based multi-agent systems have shown strong capabilities in tasks such as code generation and collaborative reasoning. However, the effectiveness and robustness of these systems critically depend on their communication topology, which is often fixed or statically learned, ignoring real-world dynamics such as model upgrades, API (or tool) changes, or knowledge source variability. To address this limitation, we propose CARD (Conditional Agentic Graph Designer), a conditional graph-generation framework that instantiates AMACP, a protocol for adaptive multi-agent communication. CARD explicitly incorporates dynamic environmental signals into graph construction, enabling topology adaptation at both training and runtime. Through a conditional variational graph encoder and environment-aware optimization, CARD produces communication structures that are both effective and resilient to shifts in model capability or resource availability. Empirical results on HumanEval, MATH, and MMLU demonstrate that CARD consistently outperforms static and prompt-based baselines, achieving higher accuracy and robustness across diverse conditions. The source code is available at: https://github.com/Warma10032/CARD.

Large language models are increasingly deployed in multi-agent systems to overcome context limitations by distributing information across agents. Yet whether agents can reliably compute with distributed information, rather than merely exchange it, remains an open question. We introduce SILO-BENCH, a role-agnostic benchmark of 30 algorithmic tasks across three communication complexity levels, evaluating 54 configurations over 1,620 experiments. Our experiments expose a fundamental Communication-Reasoning Gap: agents spontaneously form task-appropriate coordination topologies and exchange information actively, yet systematically fail to synthesize distributed state into correct answers. The failure is localized to the reasoning-integration stage where agents often acquire sufficient information but cannot integrate it. This coordination overhead compounds with scale, eventually eliminating parallelization gains entirely. These findings demonstrate that naively scaling agent count cannot circumvent context limitations, and SILO-BENCH provides a foundation for tracking progress toward genuinely collaborative multi-agent systems. The code is available at https://github.com/jwyjohn/acl26-silo-bench .

Collaborative perception (CP) extends sensing range through feature sharing, but most systems remain LiDAR-centric. Camera and 4D radar sensing combines dense semantics with lower-cost range--velocity measurements, yet exploiting their complementarity across agents remains difficult. Camera depth ambiguity produces spatially diffuse BEV evidence, while sparse transmission couples the choice of evidence to its subsequent contribution. Separate acquisition and aggregation objectives can assign conflicting source preferences to the retained evidence. We propose RC-GeoCP, connecting radar-grounded representation with persistent receiver--source attribution. Geometric Structure Rectification (GSR) uses radar-conditioned deformable sampling and gated calibration to construct grounded BEV features before communication. On these features, Uncertainty-Aware Communication (UAC) combines receiver demand with semantic confidence and radar support in a pre-delivery responsibility field that selects sparse support within fixed quotas. The Consensus-Driven Assembler (CDA) maps the same field into a bounded source mixture, carrying acquisition preferences into assembly. Persistent attribution coordinates spatial selection and source weighting across sparse transmission. Across two datasets and three aggregation backbones, RC-GeoCP consistently improves detection. Its sparse model retains 98.6%/96.6% of dense [email protected] at 10.88/4.95 MB per link. Code will be released.

Many complex tasks require extended effort, diverse capabilities, or coordinated actions beyond what a single agent can provide. However, simply adding more agents does not guarantee better performance, as effective cooperation depends on how agents interact with each other and with task structure to satisfy evolving constraints over time. This challenge is amplified for LLM-based multi-agent systems (LLM-MAS): plans, messages, and revisions occur in natural language, whereas task progress depends on grounded environment actions. Current evaluations mostly treat cooperation as an implicit ingredient of final task success, leaving both cooperation and the effect of multi-agent interaction on task dynamics difficult to study. We introduce COOP$^2$, an evaluation framework that grounds high-level agent cooperation dynamics in LLM-MAS within task progress in the environment. COOP$^2$ then defines cooperative tasks with verifiable cooperative requirements, allowing us to analyze how cooperation unfolds over time with respect to task progress, as well as where and why cooperation breaks down. Building on this framework, we develop COOP$^2$-Repair, which predicts constraint failures from group plans and opens targeted repair channels for guided revisions. Across two environments and three communication structures, COOP$^2$-Repair improves task success and constraint satisfaction while exposing the additional decision overhead and communication load required for repair. The project web page can be found at: https://happyeureka.github.io/coop2.

Modern supply networks are complex interconnected systems. Multi-agent models are increasingly explored to optimise their performance. Most research assumes agents will have full observability of the system by having a single policy represent the agents, which seems unrealistic as this requires companies to share their data. The alternative is to develop a Hidden-Markov Process with separate policies, making the problem challenging to solve. In this paper, we propose a multi-agent system where the factory agent can share information downstream, increasing the observability of the environment. It can choose to share no information, lie, tell the truth or combine these in a mixed strategy. The results show that data sharing can boost the performance, especially when combined with a cooperative reward shaping. In the high demand scenario there is limited ability to change the strategy and therefore no data sharing approach benefits both agents. However, lying benefits the factory enough for an overall system improvement, although only by a relatively small amount compared to the overall reward. In the low demand scenario, the most successful data sharing is telling the truth which benefits all actors significantly.

Multi-agent debate (MAD) systems leverage collective intelligence to enhance reasoning capabilities, yet existing approaches struggle to simultaneously optimize accuracy, consensus formation, and computational efficiency. Static topology methods lack adaptability to task complexity variations, while external LLM-based coordination risks introducing privileged knowledge that compromises debate neutrality. This work presents RUMAD (Reinforcement-Unifying Multi-Agent Debate), a novel framework that formulates dynamic communication topology control in MAD as a reinforcement learning (RL) problem. RUMAD employs a content-agnostic observation scheme that captures high-level debate dynamics avoiding access to raw agent reasoning content. RUMAD uses a multi-objective reward to model solution quality, cohesion and efficiency. A PPO-trained controller dynamically adjusts edge weights in the communication graph, while a dual-threshold mechanism enables fine-grained control over both agent activation and information visibility. Experimental evaluation across MMLU, GSM8K, and GPQA benchmarks demonstrates that RUMAD achieves substantial efficiency gains, reducing token costs by over 80%, while still improving reasoning accuracy compared to single LLM model and multiple MAD baselines. Notably, RUMAD trained exclusively on MMLU exhibits robust zero-shot generalization to out-of-domain (OOD) tasks, indicating that the learned communication strategies capture task-independent principles of effective multi-agent coordination. These results establish RUMAD as a efficient and robust approach for deploying multi-agent reasoning application with practical resource constraints.