Hi-Drive: Hierarchical POMDP Planning for Safe Autonomous Driving in Diverse Urban Environments
Xuanjin Jin, Chendong Zeng, Shengfa Zhu, Chunxiao Liu, Panpan Cai
AI summary
Problem
Autonomous planners struggle to simultaneously handle uncertainties in high-level behavioral intentions and low-level driving styles due to computational intractability, often compromising safety in complex urban environments.
Approach
Hi-Drive uses a hierarchical POMDP framework that tracks exo-agents' intentions and styles via Bayesian filtering, searches optimal high-level behaviors with a belief tree, and refines low-level trajectories through importance sampling.
Key results
- Competitive nuPlan benchmark performance without training
- Improved safety and task completion on real-world datasets
- Real-time planning in complex urban scenes
- Transparent hierarchical belief tracking and reasoning
Why it matters
Provides a training-free, computationally efficient planning framework that enhances safety and robustness for autonomous vehicles navigating unpredictable urban traffic.
Abstract
Uncertainties in dynamic road environments pose significant challenges for behavior and trajectory planning in autonomous driving. This paper introduces Hi-Drive, a hierarchi- cal planning algorithm addressing uncertainties at both behavior and trajectory levels using a hierarchical Partially Observable Markov Decision Process (POMDP) formulation. Hi-Drive em- ploys driver models to represent uncertain behavioral intentions of other vehicles and uses their parameters to infer hidden driving styles. By treating driver models as high-level decision- making actions, our approach effectively manages the exponential complexity inherent in POMDPs. To further enhance safety and robustness, Hi-Drive integrates a trajectory optimization based on importance sampling, refining trajectories using a compre- hensive analysis of critical agents. Evaluations on real-world urban driving datasets demonstrate that Hi-Drive significantly outperforms state-of-the-art planning-based and learning-based methods across diverse urban driving situations in real-world benchmarks.