Wang, ChengGang (National University of Defense Technology), Wu, Tao (National University of Defense Technology), zhang, pengnian (College of Intelligence Science and Technology，National U), Li, Junxiang (National University of Defense Technology)

Elevation-Aware Path Planning with Learning Optimal Probability for Autonomous Driving

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on April 2, 2026

Abstract

Path planning is one of the core technologies for achieving safe and efficient navigation in autonomous vehicles, especially in complex terrains where elevation changes can significantly affect vehicle stability and energy efficiency. Traditional heuristic path planning algorithms often rely on 2D grid map inputs, which struggle to adapt to the impact of elevation changes on path quality in real-world terrain scenarios. This paper proposes a learnable path planning framework for autonomous vehicles that considers elevation, innovatively integrating elevation constraints with learnable heuristic search to enhance spatial path smoothness and planning efficiency simultaneously. Specifically, this study presents two key improvements: First, by incorporating terrain elevation constraints into the cost function, we optimize the vertical continuity of paths, effectively reducing the elevation cost of planned paths in complex terrain environments, thereby enhancing the terrain passability of autonomous vehicles. Second, we design a lightweight network for learning the optimal probability heuristic, addressing the challenges of low search efficiency and poor generalization in traditional A* algorithms caused by reliance on static heuristic functions (such as Euclidean and Manhattan distances) in complex obstacle environments. Experimental results demonstrate that our planner outperforms existing methods in three key aspects: (1) achieving lower elevation costs in complex terrains compared to other planners; (2) reducing search costs relative to traditional heuristic planners while maintaining superior efficiency across diverse mapping environments; (3) attaining comparable path planning performance to similar learnable planners with only about 50% model parameters.