ResAD: Normalized Residual Trajectory Modeling for End-to-End Autonomous Driving
2510.08562v1
cs.CV, cs.RO
2025-10-11
Авторы:
Zhiyu Zheng, Shaoyu Chen, Haoran Yin, Xinbang Zhang, Jialv Zou, Xinggang Wang, Qian Zhang, Lefei Zhang
Abstract
End-to-end autonomous driving (E2EAD) systems, which learn to predict future
trajectories directly from sensor data, are fundamentally challenged by the
inherent spatio-temporal imbalance of trajectory data. This imbalance creates a
significant optimization burden, causing models to learn spurious correlations
instead of causal inference, while also prioritizing uncertain, distant
predictions, thereby compromising immediate safety. To address these issues, we
propose ResAD, a novel Normalized Residual Trajectory Modeling framework.
Instead of predicting the future trajectory directly, our approach reframes the
learning task to predict the residual deviation from a deterministic inertial
reference. The inertial reference serves as a counterfactual, forcing the model
to move beyond simple pattern recognition and instead identify the underlying
causal factors (e.g., traffic rules, obstacles) that necessitate deviations
from a default, inertially-guided path. To deal with the optimization imbalance
caused by uncertain, long-term horizons, ResAD further incorporates Point-wise
Normalization of the predicted residual. It re-weights the optimization
objective, preventing large-magnitude errors associated with distant, uncertain
waypoints from dominating the learning signal. Extensive experiments validate
the effectiveness of our framework. On the NAVSIM benchmark, ResAD achieves a
state-of-the-art PDMS of 88.6 using a vanilla diffusion policy with only two
denoising steps, demonstrating that our approach significantly simplifies the
learning task and improves model performance. The code will be released to
facilitate further research.
Ссылки и действия
Дополнительные ресурсы: