Terrain-Adaptive Planning of a Mobile Robot With a Multiaxis Gimbal System for Stable SLAM

Abstract

Simultaneous localization and mapping (SLAM) for robots or automobiles in flat terrain or urban environments has been successfully addressed by mature solutions. However, navigating rugged terrain or off-road environments poses a significant challenge for SLAM due to high-frequency vibrations and rapid pitching motions induced by the terrain, leading to frequent SLAM failures. To address this challenge, we introduce a hybrid multi-axis gimbal system equipped with rotation mechanisms (pitch and yaw) and a translation mechanism (Z-axis). Our approach involves simulating a virtual suspension system using a virtual model control algorithm, enabling precise Z-axis stabilization control. This critical step ensures sensor stability even in rugged terrains. Furthermore, we propose a terrain-adaptive planning algorithm to adjust the virtual suspension parameters and traversal speed of the robot, ensuring the sensor data quality across diverse terrains. Specifically, we formulate the dynamic equation of the gimbal system and establish a coupled optimization problem to select optimal suspension control parameters and traversal speed. To efficiently and accurately solve this optimization problem, we propose a data-driven parameter prediction network. This network analyzes the sensed terrain data, and then automatically adjusts the virtual suspension parameters and traversal speed of the robot for diverse terrains, ensuring superior sensor data quality across all terrains. Finally, simulations and experiments are conducted to validate the improvements achieved by the designed gimbal system and the proposed terrain-adaptive planning algorithm. These improvements are demonstrated through enhanced sensor stability, as well as improved SLAM robustness and accuracy in rugged terrains.