Gradient-Based Local Traversability Analysis Using Structured 3D Distance Data
Abstract
This paper presents a computationally efficient method for local traversability analysis using gradient-based processing of structured 3D distance data from LiDAR and stereo/depth sensors. Local terrain cost is computed directly from the z-component of surface normals estimated between neighboring points in the row-column structured 3D data. Point-wise costs are projected onto a 2D robot-centric grid and aggregated per cell to produce a traversability costmap suitable for the Nav2 stack in ROS2. The approach avoids explicit elevation-map generation and supports fusion of multiple 3D sensors into a single costmap. The method was implemented as a ROS2 node and deployed on a CSIR mobile robot platform equipped with a 32-beam LiDAR and stereo cameras. On a Jetson Xavier AGX, average processing times were 2.3 ms for 32-beam LiDAR scans and under 2 ms for 448x256 stereo point clouds, demonstrating real-time capability.