From Features to Reference Points: Lightweight and Adaptive Fusion for Cooperative Autonomous Driving

TL;DR

RefPtsFusion introduces a lightweight, interpretable cooperative driving framework that exchanges compact reference points instead of large feature maps, significantly reducing communication bandwidth while maintaining perception accuracy.

Contribution

The paper proposes a novel reference point-based fusion method that is sensor- and model-independent, enhancing efficiency and robustness in cooperative autonomous driving.

Findings

Reduces communication from hundreds of MB/s to a few KB/s at 5 FPS.

Maintains stable perception performance with minimal bandwidth.

Demonstrates robustness and scalability in diverse driving scenarios.

Abstract

We present RefPtsFusion, a lightweight and interpretable framework for cooperative autonomous driving. Instead of sharing large feature maps or query embeddings, vehicles exchange compact reference points, e.g., objects' positions, velocities, and size information. This approach shifts the focus from "what is seen" to "where to see", creating a sensor- and model-independent interface that works well across vehicles with heterogeneous perception models while greatly reducing communication bandwidth. To enhance the richness of shared information, we further develop a selective Top-K query fusion that selectively adds high-confidence queries from the sender. It thus achieves a strong balance between accuracy and communication cost. Experiments on the M3CAD dataset show that RefPtsFusion maintains stable perception performance while reducing communication overhead by five orders of magnitude, dropping from hundreds of MB/s to only a few KB/s at 5 FPS (frame per second), compared to traditional feature-level fusion methods. Extensive experiments also demonstrate RefPtsFusion's strong robustness and consistent transmission behavior, highlighting its potential for scalable, real-time cooperative driving systems.