Reproduce “Defer to Plan: Adaptive Multi-Agent Fusion for End-to-End V2X Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nuoran Li, Zhang Zhang, Yueran Zhao, Tianze Wang, Chao Sun
The paper introduces "Defer to Plan," an end-to-end V2X collaborative driving system that defers multi-agent feature fusion to the planning stage. This approach uses an autoregressive decoder with adaptive weights to integrate ego-vehicle and roadside unit data, achieving a 3.33% higher Driving Score and 7.32% higher Infraction Score compared to the CoDriving baseline on the V2Xverse simulation platform.
Key insights
- Deferring multi-agent feature fusion to the planning stage, coupled with adaptive weighting, can significantly improve overall driving performance and safety compared to fixed-weight fusion at the perception level.
- Specialized processing for heterogeneous V2X data, including purifying roadside features and employing MoE experts for different driving maneuvers, enhances feature representation for diverse scenarios.
- An autoregressive decoder with a hybrid causal mask enables dynamic information aggregation and coherent trajectory generation, demonstrating robustness to real-world V2X challenges like pose noise and communication latency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.