Mimir: Hierarchical Goal-Driven Diffusion with Uncertainty Propagation for End-To-End Autonomous Driving
Zebin Xing, Yupeng Zheng, Qichao Zhang, Zhixing Ding, Pengxuan Yang, Songen Gu, zhongpu xia, Dongbin Zhao
AI summary
Problem
Existing hierarchical end-to-end driving methods suffer from error propagation due to deterministic goal-point predictions and suffer from slow inference speeds caused by heavy high-level guidance modules.
Approach
Mimir employs a fast-slow dual-system framework that predicts goal points with Laplace-modeled uncertainty and extrapolates extended goals, which are then injected into a diffusion-based trajectory planner to guide robust motion generation.
Key results
- Laplace-distribution uncertainty modeling for goal points
- Multi-rate guidance mechanism with extended goal extrapolation
- 20% improvement in driving score on Navhard and Navtest benchmarks
- 1.6× faster high-level module inference without accuracy loss
Why it matters
Enables safer and more deployable real-time autonomous driving systems by mitigating error propagation and computational bottlenecks in hierarchical planning architectures.
Abstract
End-to-end autonomous driving has emerged as a pivotal direction in the field of autonomous systems. Recent works have demonstrated impressive performance by incorporating high- level guidance signals to steer low-level trajectory planners. How- ever, their potential is often constrained by inaccurate high-level guidance and the computational overhead of complex guidance modules. To address these limitations, we propose Mimir, a novel hierarchical dual-system framework capable of generating robust trajectories relying on goal points with uncertainty estimation: (1) Unlike previous approaches that deterministically model, we estimate goal point uncertainty with a Laplace distribution to enhance robustness; (2) To overcome the slow inference speed of the guidance system, we introduce a multi-rate guidance mecha- nism that predicts extended goal points in advance. Validated on challenging Navhard and Navtest benchmarks, Mimir surpasses previous state-of-the-art methods with a 20% improvement in the driving score EPDMS, while achieving 1.6× improvement in high- level module inference speed without compromising accuracy. The code and models will be released soon to promote reproducibility and further development.