Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hi-Drive: Hierarchical POMDP Planning for Safe Autonomous Driving in Diverse Urban Environments

Xuanjin Jin, Chendong Zeng, Shengfa Zhu, Chunxiao Liu, Panpan Cai

TL;DR

Hi-Drive addresses the dual uncertainty of other drivers' intentions and driving styles in urban autonomy by formulating a hierarchical POMDP that uses driver-models as high-level actions. It combines a multi-model belief tracker (with a hierarchical Bayesian filter), an online belief-tree planner (DESPOT-based), and a cross-scenario trajectory optimizer using importance sampling to refine safety- and robustness-critical trajectories. The approach yields real-time performance, scales to urban driving with many agents, and provides interpretable reasoning, achieving state-of-the-art results on large benchmarks without requiring training. This work significantly advances safe, robust autonomous driving by tightly integrating high-level intention reasoning with low-level trajectory optimization in a general urban setting.

Abstract

Uncertainties in dynamic road environments pose significant challenges for behavior and trajectory planning in autonomous driving. This paper introduces Hi-Drive, a hierarchical planning algorithm addressing uncertainties at both behavior and trajectory levels using a hierarchical Partially Observable Markov Decision Process (POMDP) formulation. Hi-Drive employs 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 comprehensive 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.

Hi-Drive: Hierarchical POMDP Planning for Safe Autonomous Driving in Diverse Urban Environments

TL;DR

Hi-Drive addresses the dual uncertainty of other drivers' intentions and driving styles in urban autonomy by formulating a hierarchical POMDP that uses driver-models as high-level actions. It combines a multi-model belief tracker (with a hierarchical Bayesian filter), an online belief-tree planner (DESPOT-based), and a cross-scenario trajectory optimizer using importance sampling to refine safety- and robustness-critical trajectories. The approach yields real-time performance, scales to urban driving with many agents, and provides interpretable reasoning, achieving state-of-the-art results on large benchmarks without requiring training. This work significantly advances safe, robust autonomous driving by tightly integrating high-level intention reasoning with low-level trajectory optimization in a general urban setting.

Abstract

Uncertainties in dynamic road environments pose significant challenges for behavior and trajectory planning in autonomous driving. This paper introduces Hi-Drive, a hierarchical planning algorithm addressing uncertainties at both behavior and trajectory levels using a hierarchical Partially Observable Markov Decision Process (POMDP) formulation. Hi-Drive employs 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 comprehensive 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.
Paper Structure (28 sections, 4 equations, 2 figures, 3 tables)

This paper contains 28 sections, 4 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Overview
  4. Hierarchical POMDP Modeling
  5. Driver Models
  6. POMDP Model
  7. State and Observation
  8. Action Space
  9. State Transition and Observation Models
  10. Reward Function
  11. Belief Tracking with Multi-Model Inference
  12. Independence Assumptions
  13. Hierarchical Bayesian Filter
  14. Low-Level Particle Filter
  15. High-Level Exact Filter
  16. ...and 13 more sections

Figures (2)

  • Figure 1: Overview of Hi-Drive: (a) The multi-model inference tracks beliefs about exo-agents' behavioral intentions and driving styles based on real-time observations. (b) The belief tree search determines the ego-vehicle’s optimal policy over high-level behaviors. (c) Trajectory optimization resamples representative scenarios using importance sampling, cross-evaluates low-level trajectories generated by those scenarios, and selects the one with the best driving performance. See detailed explanations of the driving scene visualization in Section \ref{['interpretation']}.
  • Figure 2: The hierarchical belief tracker. Green arrows represent the process of updating the belief. (a) The belief over behavioral intention $m$. (b) The belief over physical states $x^t$ and driving styles $\theta$.