Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DrPlanner: Diagnosis and Repair of Motion Planners for Automated Vehicles Using Large Language Models

Yuanfei Lin, Chenran Li, Mingyu Ding, Masayoshi Tomizuka, Wei Zhan, Matthias Althoff

TL;DR

DrPlanner introduces an autonomous framework that leverages large language models to diagnose and repair motion planners for automated vehicles. It constructs a structured multimodal description of planners, uses in-context demonstrations, and employs a closed-loop iterative prompting process to generate patches that improve trajectory costs measured by $J_{\text{SM1}}$. Empirical evaluation on open-source CommonRoad planners shows high pass rates and substantial reductions in trajectory cost, with ablations confirming the importance of few-shot prompts and feedback. This work demonstrates that LLM-driven diagnosis and patch generation can meaningfully enhance complex motion-planning systems, enabling scalable, data-driven improvement of automated-vehicle planners while maintaining safety considerations.

Abstract

Motion planners are essential for the safe operation of automated vehicles across various scenarios. However, no motion planning algorithm has achieved perfection in the literature, and improving its performance is often time-consuming and labor-intensive. To tackle the aforementioned issues, we present DrPlanner, the first framework designed to automatically diagnose and repair motion planners using large language models. Initially, we generate a structured description of the planner and its planned trajectories from both natural and programming languages. Leveraging the profound capabilities of large language models, our framework returns repaired planners with detailed diagnostic descriptions. Furthermore, our framework advances iteratively with continuous feedback from the evaluation of the repaired outcomes. Our approach is validated using both search- and sampling-based motion planners for automated vehicles; experimental results highlight the need for demonstrations in the prompt and show the ability of our framework to effectively identify and rectify elusive issues.

DrPlanner: Diagnosis and Repair of Motion Planners for Automated Vehicles Using Large Language Models

TL;DR

DrPlanner introduces an autonomous framework that leverages large language models to diagnose and repair motion planners for automated vehicles. It constructs a structured multimodal description of planners, uses in-context demonstrations, and employs a closed-loop iterative prompting process to generate patches that improve trajectory costs measured by . Empirical evaluation on open-source CommonRoad planners shows high pass rates and substantial reductions in trajectory cost, with ablations confirming the importance of few-shot prompts and feedback. This work demonstrates that LLM-driven diagnosis and patch generation can meaningfully enhance complex motion-planning systems, enabling scalable, data-driven improvement of automated-vehicle planners while maintaining safety considerations.

Abstract

Motion planners are essential for the safe operation of automated vehicles across various scenarios. However, no motion planning algorithm has achieved perfection in the literature, and improving its performance is often time-consuming and labor-intensive. To tackle the aforementioned issues, we present DrPlanner, the first framework designed to automatically diagnose and repair motion planners using large language models. Initially, we generate a structured description of the planner and its planned trajectories from both natural and programming languages. Leveraging the profound capabilities of large language models, our framework returns repaired planners with detailed diagnostic descriptions. Furthermore, our framework advances iteratively with continuous feedback from the evaluation of the repaired outcomes. Our approach is validated using both search- and sampling-based motion planners for automated vehicles; experimental results highlight the need for demonstrations in the prompt and show the ability of our framework to effectively identify and rectify elusive issues.
Paper Structure (20 sections, 1 equation, 4 figures, 1 algorithm)

This paper contains 20 sections, 1 equation, 4 figures, 1 algorithm.

Table of Contents

  1. Automated Software Repair
  2. Language Models for Motion Planning
  3. Contributions
  4. Preliminaries
  5. Motion Planning for Automated Vehicles
  6. Prompt Engineering for LLMs
  7. DrPlanner
  8. Overall Algorithm
  9. Diagnostic Description
  10. Instructions
  11. Motion Planner
  12. Planned Trajectory
  13. Few-Shots
  14. LLM Querying and Iterative Prompting
  15. Evaluation
  16. ...and 5 more sections

Figures (4)

  • Figure 1: An example usage of $\mathtt{DrPlanner}$: In a critical scenario, our imperfect motion planner plans a trajectory. The description of the trajectory and the planner is then fed into $\mathtt{DrPlanner}$. By harnessing the strengths of LLMs, we adeptly diagnose and repair the deficiencies within the planner.
  • Figure 2: Exemplary motion planning problem, where the ego vehicle needs to travel from its initial state to reach the goal region safely and efficiently.
  • Figure 3: Overview of the $\mathtt{DrPlanner}$ framework. The process starts with obtaining a planned trajectory for the planning problem with the given motion planner. Then, the planned trajectory is evaluated by the objective function. Afterwards, the description for the planner is generated and used to prompt an off-the-shelf LLM to generate the diagnoses and prescriptions for the planner, along with the patched programs. After applying the patches, the evaluation of the updated planner is incorporated back into the prompt as feedback to continuously enhance the diagnostic performance (marked by dashed arrows).
  • Figure : diagnoseAndRepairPlanner