communication-efficient map approximation

This work addresses collaborative navigation for heterogeneous, resource-constrained robot teams operating in unknown environments, focusing on efficient map communication under severe bandwidth constraints. We propose a task-driven, communication-aware iterative compression framework: (i) a Kalman filter–based map estimation decoder for robust reconstruction; (ii) a lightweight communication constraint model; and (iii) a dynamically scalable set of compression templates to adapt to environmental complexity. Crucially, we introduce the first sequential compression selection strategy that jointly optimizes communication cost and navigation task performance. Evaluated on Mars slope terrain and real-world terrestrial maps, our method reduces communication volume by 98% while achieving path planning accuracy comparable to full-map transmission. Moreover, computational latency is significantly lower than that of state-of-the-art approaches, demonstrating both efficacy and efficiency for bandwidth-limited multi-robot systems.

In resource-constrained multi-agent systems (MAS), conventional communication protocols suffer from low efficiency and misalignment between transmitted information and task objectives. Method: This paper proposes a goal-oriented communication paradigm that transcends traditional approaches focused solely on signal fidelity or bandwidth optimization. Its core innovation is a task-goal-anchored importance evaluation mechanism, integrating information theory, communication theory, and machine learning into a unified, learnable communication framework. Contribution/Results: The framework enables end-to-end trainable communication policies, emergent cooperative protocol synthesis, and robust multi-agent coordination under communication constraints. We systematically formalize the theoretical foundations of goal-oriented communication and delineate its application pathways—along with key challenges—in swarm robotics, federated learning, and edge intelligence. This work establishes a novel task-driven paradigm for intelligent communication in distributed autonomous systems.

This work addresses the robust co-synthesis of joint action and communication policies for stochastic multi-agent systems under communication constraints, aiming to maximize the probability of reaching a common reach-avoid objective. We propose a novel information-overhead cost function, enabling— for the first time—the joint robust synthesis of action and communication policies. Our approach models the system as a stochastic game and integrates symbolic policy synthesis, constrained optimization, probabilistic reachability analysis, and quantitative information-flow metrics. The method rigorously guarantees performance bounds under dynamic bandwidth constraints and establishes both the existence and computational tractability of feasible policies. Evaluated on multiple benchmark tasks, the synthesized policies achieve over 92% of the unconstrained optimal reach-avoid probability while satisfying strict communication limits, thereby significantly advancing both the practical applicability and theoretical completeness of resource-constrained multi-agent coordination.

To address delays caused by sudden agent failures in multi-agent path finding (MAPF), this paper proposes a dynamic scheduling adaptation framework that avoids global replanning. The method introduces a distributed, fault-resilient coordination mechanism comprising two novel communication protocols: one guarantees that the makespan increase is strictly bounded by *k* timesteps (*k* being the number of failed agents), while the other offloads computational load to networked infrastructure nodes to reduce on-agent overhead. It further integrates localized conflict resolution with online path adjustment and provides theoretical bounds on solution quality and runtime complexity. Experiments demonstrate that the approach significantly reduces replanning cost, maintains near-optimal scheduling efficiency, and scales effectively with agent count—showcasing strong practical deployability for real-world MAPF systems.

This work addresses task-driven multi-robot exploration in unknown environments, where mobile sensor robots collaboratively assist a primary robot to efficiently reach a target. To handle communication-constrained scenarios, we propose a task-oriented uncertainty metric as the reward function—marking the first explicit incorporation of map compression distortion into exploration decision-making. We design a scalable map compression mechanism integrating sparse coding with the information bottleneck principle, and develop a distributed communication–action coordination framework that unifies multi-agent reinforcement learning with distributed consensus optimization. Experiments on realistic map simulations demonstrate significant improvements: target arrival time is substantially reduced, communication overhead decreases by 37%, and performance consistently surpasses baseline methods including information gain and random exploration.

This study addresses the absence of a joint graph-structure design framework in existing decentralized optimization by proposing GATE. This method introduces a novel "hybrid-tearing" paradigm that decomposes the global problem into local subproblems via message passing, jointly designing constraint representations, dual variable blocks, and connected clusters to transcend conventional gossip communication limitations. Efficient collaborative computation is achieved through tree-recursive updates combined with a lightweight surrogate model (GATE-S). Theoretically, the algorithm is proven to attain a linear convergence rate under explicit dependencies on network topology and function regularity. Empirical evaluations further validate its significant advantages in reducing both communication and computational overhead.

This work addresses the challenge of bandwidth-constrained multi-agent reinforcement learning (MARL), where conventional approaches suffer performance degradation due to the entanglement of communication and policy representations, causing compression to adversely affect policy efficacy. To overcome this limitation, the authors propose a decoupled architecture that separates communication from policy learning via dedicated communication channels and introduces a normalized bandwidth budget β, enabling, for the first time, an isolated analysis of communication overhead and policy capacity. The method employs a lightweight SLIM design, end-to-end training, and explicit modeling of partially observable environments. Evaluated across multiple MARL benchmarks, it achieves state-of-the-art performance while maintaining strong robustness and scalability even under severe bandwidth compression.

This work addresses the problem of approximating optimal utility in multi-agent coordination with minimal communication. It introduces a novel approach based on the Frieze–Kannan weak regularity lemma, which avoids strong assumptions such as informational substitutability. By coarsening the observation space into a constant-sized partition, the original game is transformed into a coarse game that can be solved efficiently. The resulting protocol runs in time polynomial in the number of agents $n$, actions $m$, and $1/\varepsilon$, while incurring only $2^{O(CC_\alpha(G))}/\varepsilon^2$ bits of communication. Under the assumption that P≠NP, this communication complexity nearly matches the theoretical lower bound, marking the first efficient communication protocol for this setting that does not rely on strong structural assumptions.

This work addresses the challenge of limited communication bandwidth in real-world multi-agent reinforcement learning systems, which often hinders collaborative efficiency. The authors propose a novel communication compression method that integrates information bottleneck theory with vector quantization to discretely encode messages under information-theoretic principles. A dynamic gating mechanism is further introduced to adaptively determine when agents should communicate, enabling selective and efficient information exchange. This approach represents the first integration of the information bottleneck principle with vector quantization for multi-agent communication. Experimental results demonstrate that the method reduces bandwidth usage by 41.4% while improving task performance by 181.8% over a no-communication baseline. Moreover, it achieves superior performance on the success rate–bandwidth Pareto frontier, with an area under the curve (AUC) of 0.198 compared to 0.142 for existing methods.

This work addresses the suboptimality and scalability limitations in large-scale multi-agent path finding (MAPF) stemming from insufficient coordination by proposing LC-MAPF, a novel framework that formulates MAPF as a decentralized partially observable Markov decision process with a learnable local communication mechanism. Integrating imitation learning and reinforcement learning, LC-MAPF employs graph neural networks to iteratively aggregate neighborhood information through multiple rounds of local message passing and generate coordinated actions. Experimental results demonstrate that LC-MAPF significantly outperforms existing learning-based solvers across diverse unseen scenarios, achieving higher success rates and superior path quality while maintaining strong scalability.