HMAD: Advancing E2E Driving with Anchored Offset Proposals and Simulation-Supervised Multi-target Scoring
Bin Wang, Pingjun Li, Jinkun Liu, Jun Cheng, Hailong Lei, Yinze Rong, Huan-ang Gao, Kangliang Chen, Xing Pan, Weihao Gu
TL;DR
HMAD tackles end-to-end driving multimodality by decoupling diverse trajectory generation from safety-aware evaluation. It introduces a BEV-based perception backbone with anchored, dictionary-derived trajectory queries refined via iterative, BEV-aware offsets to produce multiple candidate paths, paired with a simulation-supervised, multi-target scorer that predicts interpretable metrics such as $EPDMS$ for ranking. Hard-case mining improves robustness in long-tail urban scenarios, while a 2D post-processing filter grounds proposals against visual evidence. On NAVSIM/OpenScene, HMAD achieves strong results, including a competitive $EPDMS$ of 65.94 and a 44.5% driving score on the CVPR 2025 private test set, demonstrating the value of integrating differentiable, simulator-informed scoring with BEV-based trajectory generation for end-to-end planning.
Abstract
End-to-end autonomous driving faces persistent challenges in both generating diverse, rule-compliant trajectories and robustly selecting the optimal path from these options via learned, multi-faceted evaluation. To address these challenges, we introduce HMAD, a framework integrating a distinctive Bird's-Eye-View (BEV) based trajectory proposal mechanism with learned multi-criteria scoring. HMAD leverages BEVFormer and employs learnable anchored queries, initialized from a trajectory dictionary and refined via iterative offset decoding (inspired by DiffusionDrive), to produce numerous diverse and stable candidate trajectories. A key innovation, our simulation-supervised scorer module, then evaluates these proposals against critical metrics including no at-fault collisions, drivable area compliance, comfortableness, and overall driving quality (i.e., extended PDM score). Demonstrating its efficacy, HMAD achieves a 44.5% driving score on the CVPR 2025 private test set. This work highlights the benefits of effectively decoupling robust trajectory generation from comprehensive, safety-aware learned scoring for advanced autonomous driving.