Potential Field Path Planning for Quadruped Robots in Stepped Terrain Using Elevation Maps

Quadruped robots provide better mobility than wheeled robots in complex environments such as stairs, slopes, and uneven terrain. However, stable and efficient navigation in such environments requires path planning that accurately incorporates elevation information. Among various path planning algorithms, the PF (Potential Field) method is well-suited for real-time applications owing to its simple computational structure and inherently high path straightness. Conventional PF methods, however, do not capture elevation changes accurately, often misclassifying traversable stepped terrain as obstacles. This misclassification results in excessive repulsive forces, unnecessary detours, and even failure to reach the goal. This study proposes an improved PF path planning method that integrates elevation map analysis to classify stepped terrain. A real-time elevation map was generated using LiDAR sensors and a GPU-accelerated processing library. Elevation difference and slope thresholds have been applied to distinguish traversable steps from true obstacles based on the physical walking limits of the robot. Repulsive force computation was omitted for traversable steps, thereby improving both path straightness and goal-reaching performance. Additionally, a cubic spline-based path smoothing process was applied to enhance control signal continuity and walking stability. Experiments conducted in the ICROS 2025 Quadruped Robot Challenge environment demonstrated that the proposed method achieved a higher goal-reaching rate and improved path quality compared with conventional PF, particularly in stepped terrain sections where the conventional PF failed to generate feasible paths. Path smoothing had little effect on path length but reduced average curvature in some courses, improving overall smoothness. However, maximum curvature increased in certain sections, potentially triggering more in-place turns under the 30-degree turning threshold of the robot. These results indicate that even though smoothing improves general path continuity, curvature constraints must be considered to optimize stability. The proposed elevation map-based PF approach significantly improves the autonomous navigation performance of quadrupedal robots in complex terrains with frequent elevation changes. It shows strong potential for deployment in disaster response, industrial site monitoring, and rugged terrain exploration.