Vehicle–Road–Cloud Collaborative Perception: Resource and Intelligence Optimization

Vehicle–road–cloud collaborative perception improves perception performance via multi-agent information sharing and data fusion, but it faces coupled trade-offs among perception accuracy, computing resources, and communication bandwidth. Optimizing agents’ intelligence or underlaying resources alone fails to resolve this conflict, limiting collaboration efficiency. We propose C4I-JO, a joint resource and intelligence optimization method for vehicle–road–cloud collaborative perception. We employ slimmable networks to achieve intelligent elasticity. Based on these, C4I-JO jointly optimizes four key dimensions to minimize resource consumption while meeting accuracy and latency constraints, including collaborative mechanisms to cut redundant communication, resource allocation to avoid supply–demand bottlenecks, intelligent elasticity to balance accuracy and resources, and computation offloading to reduce local burden. We propose a two-layer iterative decoupling algorithm that addresses the optimization problem. Specifically, the outer level leverages Second-Order Cone Programming (SOCP) and the interior-point method, while the inner level utilizes a Genetic Algorithm (GA). Simulations show that C4I-JO outperforms baselines in both resource efficiency and perception quality.