Reproduce “HMAD: Advancing E2E Driving with Anchored Offset Proposals and Simulation-Supervised Multi-target Scoring” 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
Bin Wang, Pingjun Li, Jinkun Liu, Jun Cheng, Hailong Lei, Yinze Rong, Huan-ang Gao, Kangliang Chen and 2 more
The HMAD framework proposes an end-to-end autonomous driving system that combines an anchored offset proposal mechanism for diverse trajectory generation with a simulation-supervised multi-target scoring module for robust path selection. The system achieves a 44.5% driving score on the CVPR 2025 private test set and an EPDMS of 65.94% on the 'warmup two stage' benchmark, demonstrating improved adaptability and safety in complex scenarios.
Key insights
- Learning to evaluate trajectories through simulation supervision, rather than relying on fixed heuristics, enables adaptive and context-dependent trade-offs between conflicting driving criteria.
- An anchored offset proposal mechanism effectively supports the generation of diverse yet stable trajectory candidates, crucial for handling multimodal driving scenarios.
- Targeted data augmentation via hard case mining is vital for enhancing model robustness and safety in critical, rare 'long-tail' driving situations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.