Prompt Description

With the rapid development of smart city emergency rescue technology, the collaborative rescue of Embodied Swarms in extreme disaster areas where communication infrastructure is damaged is limited by semantic redundancy and the spatiotemporal non-stationarity of multi-dimensional resources. Aiming at the problems of low task completion rate of traditional bit-level transmission schemes under limited bandwidth, and the control failure caused by the curse of dimensionality of large-scale embodied swarms in hybrid action spaces, this paper proposes a Semantic-Awareness-Driven Embodied Swarm Diffusive Collaborative Computing and 3C Closed-Loop Optimization Framework (SD-HEC). This framework includes three core mechanisms: 1. Constructing a semantic-aware dynamic hypergraph model (SH-Sens) that uses hyperedges to capture high-order group coupling relationships between heterogeneous nodes, and introduces a lightweight semantic encoder to quantify the "semantic importance" of tasks, realizing high-fidelity and low-bandwidth mapping from physical space to digital space; 2. Proposing a disturbance-resistant diffusion model generative multi-agent decision algorithm (Diff-MAS): Aiming at the problems of non-full rank information perception and time delay jitter caused by communication constraints, the generative characteristics of the diffusion model are used to perform "semantic reconstruction" and denoising of the missing cluster state under partially observable conditions. By learning the joint action distribution, the strategy misalignment caused by communication disconnection is alleviated, and the collaborative consistency of large-scale heterogeneous clusters under non-ideal communication links is guaranteed; 3. Establishing a 3C closed-loop on-demand flow control mechanism based on information value (V-RFC): Aiming at the non-stationarity of communication resources and computing load, the "semantic contribution" of perceived data to the control target and the perceived freshness (AoI) are quantified in real time, and a reverse adjustment mechanism is dynamically triggered to prioritize the transmission of key control instructions when communication is extremely scarce, realizing entropy reduction optimization of the entire "perception-transmission-calculation-control" link under limited resources. Based on the (AirSim+NS3) high-fidelity communication-constrained simulation environment, SD-HEC improves the task semantic fidelity by 41.2% and the system energy efficiency ratio by 33.5% in large-scale heterogeneous clusters compared with the traditional bit-level transmission scheme. Compared with baseline methods such as Lyapunov optimization and JCO-Deep, the Diff-MAS algorithm significantly improves the sample efficiency of model convergence by about 1.8 times; under extreme conditions where the communication link packet loss rate is as high as 40%, based on the 3C closed-loop semantic flow control mechanism, the system's key task completion rate remains above 89.6%. This research breaks through the constraints of traditional communication on swarm intelligence, verifies the engineering feasibility of generative AI and semantic communication in extremely constrained environments, and provides theoretical basis and technical support for building a high-resilience adaptive disaster emergency response system. Keywords: Embodied Swarm; Semantic Communication; Diffusion Model; Dynamic Hypergraph; 3C Closed-Loop Optimization; Flow Control Mechanism