Abstract:Uncertainties in dynamic road environments pose significant challenges for behavior and trajectory planning in autonomous driving. This paper introduces BoT-Drive, a planning algorithm that addresses uncertainties at both behavior and trajectory levels within a Partially Observable Markov Decision Process (POMDP) framework. BoT-Drive employs driver models to characterize unknown behavioral intentions and utilizes their model parameters to infer hidden driving styles. By also treating driver models as decision-making actions for the autonomous vehicle, BoT-Drive effectively tackles the exponential complexity inherent in POMDPs. To enhance safety and robustness, the planner further applies importance sampling to refine the driving trajectory conditioned on the planned high-level behavior. Evaluation on real-world data shows that BoT-Drive consistently outperforms both existing planning methods and learning-based methods in regular and complex urban driving scenes, demonstrating significant improvements in driving safety and reliability.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges brought by uncertainty to the behavior and trajectory planning of autonomous vehicles in dynamic road environments. Specifically, these uncertainties mainly stem from the unpredictability of the driving intentions and styles of other traffic participants, as well as the impact of this uncertainty on the performance of planning algorithms in complex interaction scenarios. Traditional planning methods often reduce computational complexity by simplifying uncertainty handling, but this method may sacrifice the safety and efficiency of autonomous vehicles. For this reason, the paper proposes the BoT - Drive algorithm, aiming to simultaneously solve the uncertainty problems at the behavior and trajectory levels through hierarchical planning in the framework of Partially Observable Markov Decision Process (POMDP), thereby improving the safety and reliability of autonomous driving in complex urban environments. The core of the BoT - Drive algorithm lies in: 1. **Hierarchical Planning**: By modeling the behavioral intentions and driving styles of other traffic participants as hidden states and using them as high - level decision - making actions for autonomous vehicles, the planning time window of POMDP is effectively reduced, and the exponential complexity problem is solved. 2. **Importance Sampling**: In order to further enhance the safety and robustness of driving trajectories, BoT - Drive applies importance sampling technology at a low level to optimize driving trajectories, especially the coverage in high - risk events. 3. **Efficient Uncertainty Modeling**: Using Dynamic Bayesian Networks (DBN) and multi - model inference methods to dynamically track the intentions and styles of other traffic participants, which improves the decision - making ability of the algorithm in complex driving scenarios. Through these innovations, BoT - Drive can achieve safer and more reliable autonomous driving behavior and trajectory planning while ensuring real - time performance.

BoT-Drive: Hierarchical Behavior and Trajectory Planning for Autonomous Driving using POMDPs

A Confidence-based Allocation Approach for USV Trajectory Tracking Via Human-Robot Co-Driving

PEP: Policy-Embedded Trajectory Planning for Autonomous Driving

Learning Hierarchical Behavior and Motion Planning for Autonomous Driving.

Learning Online Belief Prediction for Efficient POMDP Planning in Autonomous Driving

Task-Motion Planning for Safe and Efficient Urban Driving

Hierarchical Game-Theoretic Planning for Autonomous Vehicles

Integrated Decision Making and Trajectory Planning for Autonomous Driving Under Multimodal Uncertainties: A Bayesian Game Approach

LeTS-Drive: Driving in a Crowd by Learning from Tree Search

Behavior and Interaction-aware Motion Planning for Autonomous Driving Vehicles based on Hierarchical Intention and Motion Prediction

Autonomous Driving at Intersections: A Critical-Turning-Point Approach for Left Turns

Reasoning Multi-Agent Behavioral Topology for Interactive Autonomous Driving

PODDP: Partially Observable Differential Dynamic Programming for Latent Belief Space Planning

GTP-UDrive: Unified Game-Theoretic Trajectory Planner and Decision-Maker for Autonomous Driving in Mixed Traffic Environments

Policy Iteration Based Approximate Dynamic Programming Toward Autonomous Driving in Constrained Dynamic Environment

PPAD: Iterative Interactions of Prediction and Planning for End-to-end Autonomous Driving

Hierarchical MPC-based Motion Planning for Automated Vehicles in Parallel Autonomy

Situation-aware decision making for autonomous driving on urban road using online POMDP

An Integrating Comprehensive Trajectory Prediction with Risk Potential Field Method for Autonomous Driving

PLOP: Probabilistic poLynomial Objects trajectory Planning for autonomous driving

Model Predictive Control for Aggressive Driving Over Uneven Terrain