Geometric-Based Collision Filtering and Stability Check for Autonomous Robot Navigation in Uneven Terrain

2026-08-10T13:55:33+07:00

This paper presents a geometric-based collision-filtering and stability-checking method for autonomous navigation of a tracked rescue robot over uneven terrain. The proposed approach integrates a two-phase collision filter and a Center-of-Gravity (CoG)-based stability check into a motion-primitive local planner. In the first phase, obstacle samples from a 2.5D elevation map are collected within a search radius around the robot. In the second phase, each obstacle sample is projected into each candidate primitive's body frame and tested against a precomputed voxel correspondence table; primitives whose blocked-cell count exceeds a threshold are discarded. The surviving primitives then undergo waypoint-by-waypoint static stability verification: the support plane beneath each of the robot's four corners is estimated, and a primitive is rejected if the plane's tilt exceeds a tilt limit or the gap between the CoG and the support plane exceeds a fall threshold. The method was validated in a hole-avoidance scenario in both Gazebo simulation and a real-world laboratory environment, using a 77 kg tracked robot equipped with dual Livox MID-360 LiDAR sensors. With the collision and stability checks enabled, the success rate improved from 80% to 100% in simulation and reached 100% in real-world trials. The wall safety ratio increased from 92.55% to 99.50% in simulation and reached 98.94% in the real-world environment, at a marginal efficiency cost of 2.4% in path length and 2.2% in travel time.

The Journal of King Mongkut’s University of Technology North Bangkok

Geometric-Based Collision Filtering and Stability Check for Autonomous Robot Navigation in Uneven Terrain