CoSDH: Communication-Efficient Collaborative Perception via Supply-Demand Awareness and Intermediate-Late Hybridization

TL;DR

CoSDH tackles the bandwidth-accuracy trade-off in collaborative 3D perception by modeling supply-demand relationships to selectively share information and by adopting an intermediate-late hybrid fusion strategy. It introduces supply-demand-aware information selection, multi-scale feature compression, and confidence-aware late fusion to preserve accuracy while dramatically reducing communication volume. Across simulated and real-world datasets (OPV2V, V2XSim, DAIR-V2X), CoSDH achieves state-of-the-art detection performance under realistic bandwidth constraints, delivering favorable accuracy-bandwidth trade-offs and demonstrating practical applicability. The approach shows robustness to latency and provides clear avenues for deployment in real-world V2X scenarios with limited communication budgets.

Abstract

Multi-agent collaborative perception enhances perceptual capabilities by utilizing information from multiple agents and is considered a fundamental solution to the problem of weak single-vehicle perception in autonomous driving. However, existing collaborative perception methods face a dilemma between communication efficiency and perception accuracy. To address this issue, we propose a novel communication-efficient collaborative perception framework based on supply-demand awareness and intermediate-late hybridization, dubbed as \mymethodname. By modeling the supply-demand relationship between agents, the framework refines the selection of collaboration regions, reducing unnecessary communication cost while maintaining accuracy. In addition, we innovatively introduce the intermediate-late hybrid collaboration mode, where late-stage collaboration compensates for the performance degradation in collaborative perception under low communication bandwidth. Extensive experiments on multiple datasets, including both simulated and real-world scenarios, demonstrate that \mymethodname~ achieves state-of-the-art detection accuracy and optimal bandwidth trade-offs, delivering superior detection precision under real communication bandwidths, thus proving its effectiveness and practical applicability. The code will be released at https://github.com/Xu2729/CoSDH.

Figures (9)

• Figure 1: Issues of existing communication-efficient collaborative perception methods.
• Figure 2: The overall architecture of CoSDH. The Supply-Demand-Aware Information Selection module selects sparse but important information, which is then further compressed by the Message Compression module to achieve efficient communication. Confidence-Aware Late Fusion compensates for the intermediate fusion detection results to improve accuracy.
• Figure 3: Comparison of our multi-scale compression and fusion with other methods. It can achieve thorough fusion with smaller communication volume.
• Figure 4: Our confidence-aware late fusion. It filters detection results based on confidence and suppresses suboptimal results from collaborative agents, improving overall detection accuracy.
• Figure 5: Comparison of the trade-off between detection accuracy and bandwidth of different methods on OPV2V xu2022opv2v, V2XSim li2022v2x and DAIR-V2X yu2022dair datasets, CoSDH achieves the best accuracy-bandwidth trade-off. The real-world limitation refers to the total bandwidth limit of 27 Mbps, which means that each collaborative agent does not exceed 6.75 Mbps.