Robust Real-Time Sampling-Based Motion Planner for Autonomous Vehicles in Narrow Environments
Minsoo Kim, Arthur Esquerre-Pourtère, Jaeheung Park
AI summary
Problem
Fixed biasing ratios in learned sampling-based planners often cause failures in narrow environments due to inherent neural network prediction inaccuracies, degrading planning performance and robustness.
Approach
The LA3T* algorithm dynamically adjusts the ratio of learned to uniform samples based on the network's confidence and replaces the single goal pose with a target tree of pre-defined path segments to reduce computational load in cluttered areas.
Key results
- Adaptive biasing ratio dynamically balances exploitation and exploration
- Target tree integration reduces computation in narrow goal regions
- Multi-head predictor enables real-time confidence-aware sampling
- Significantly increases success rates and reduces path length in simulation and real-world tests
Why it matters
Enables reliable, real-time navigation for autonomous vehicles and robots in tight, dynamic environments where conventional learning-based planners typically fail.
Abstract
Real-time sampling-based planners increasingly use learned sampling distributions for faster planning in autonomous vehicles. These planners employ a neural network to predict the optimal path and bias some samples toward the path. However, inherent prediction inaccuracies of the network often lead to suboptimal paths, especially in narrow spaces. Learned samples should be used carefully based on accuracy, as inaccurate samples can degrade planning performance. To address this problem, this paper proposes Learned Adaptive Anytime TargetTree-RRT* (LA3T*) algorithm. The proposed planner introduces the adap- tive biasing ratio. The approach learns to assess the reliability of the learned distribution using the network’s confidence. This confidence approximates a proper ratio of learned samples used, thereby adaptively maximizing planning performance while considering a level of prediction accuracy. Furthermore, the LA3T* algorithm incorporates the target tree algorithm. The goal pose is replaced with a set (target tree) of pre-defined optimal path segments, reducing computational efforts in narrow regions. Experiments in various driving tasks explore the benefits of each component through ablation studies. The proposed algorithm significantly increases the success rate and reduces the path length in simulated and real-world scenarios compared to other sampling-based methods. Note to Practitioners—This work was motivated by the need to develop a practical real-time motion planning algorithm for autonomous vehicles in narrow environments. Although learning-based sampling methods have been actively explored, their planning performance often degrades due to inaccurate predictions from the learned model, limiting their practicability and robustness. The Learned Adaptive Anytime TargetTree-RRT* (LA3T*) algorithm addresses this problem by adaptively lever- aging the learned model based on its reliability and integrating it with the target tree algorithm. Experiments conducted in var- ious driving and parking scenarios demonstrate its practicality Received 12 August 2024; revised 22 January 2025 and 3 April 2025; accepted 19 May 2025. Date of publication 29 May 2025; date of current version 12 June 2025. This article was recommended for publication by Associate Editor K. Yu and Editor J. Yi upon evaluation of the reviewers’ comments. This work was supported by the Institute of Information and Communications Technology Planning and Evaluation (IITP) grants funded by Korean Government [Ministry of Science and ICT (MSIT)] through the Core Technology Development of Open Simulator for Software-Defined Everything (SDx) Intelligent Application Development under Grant RS-2024-00459435 and through the Artificial Intelligence Graduate School Program at Seoul National University under Grant RS-2021-II211343. (Corresponding author: Jaeheung Park.) Minsoo Kim and Arthur Esquerre-Pourt`ere are with the Department of Intelligence and Information, Graduate School of Convergence Science and Technology, Seoul National University, Seoul 08826, Republic of Korea (e-mail: msk930512@snu.ac.kr; tutur263@snu.ac.kr). Jaeheung Park is with the Department of Intelligence and Information, Graduate School of Convergence Science and Technology, and ASRI, AIIS, Seoul National University, Seoul 08826, Republic of Korea, and also with the Advanced Institute of Convergence Technology, Suwon 16229, Republic of Korea (e-mail: park73@snu.ac.kr). Digital Object Identifier 10.1109/TASE.2025.3574262 and robustness, notably improving success rates in worst-case situations and reducing path length. Furthermore, experiments in sensor noise and dynamic environments—particularly when combined with existing methods—highlight the algorithm’s extensibility. LA3T* can be applied to various autonomous systems needing efficient, real-time motion planning in narrow spaces, such as other autonomous vehicles or robots. The algo- rithm has not yet been investigated in highly out-of-distribution scenarios. Future work will focus on improving performance in such scenarios and better handling more dynamic environments.