月球南极巡视器多模态感知与避障方法

Multimodal Perception and Obstacle Avoidance Method for Lunar South Pole Rover

月球南极地区地形复杂,太阳高度角极低导致阴影区域变化大,为巡视器的自主导航提出了巨大的挑战。本工作提出一种面向月球南极的巡视器多模态感知与避障路径规划方案。一方面通过RGB相机获得全局的场景影像,进行初步的障碍物检测;另一方面利用深度相机获取实时的环境深度,通过简化的SLAM算法构建局部地图,即实时生成点云图;根据点云信息,计算路面粗糙度、坡度、阶跃信息等物理环境信息。利用上述多模态数据结合地图数据实时更新巡视器位姿和周围环境,结合使用多模态局部算法和障碍检测算法进行障碍物感知和避障决策,使巡视器能够动态调整路径。仿真结果表明该自主导航系统显著提高巡视器在复杂未知环境中的自主导航能力。

The terrain of the lunar south pole is complex,and the extremely low solar altitude angle causes large changes in the shadow area,which poses a huge challenge to the autonomous navigation of the rover.This work proposes a multi-modal sensing and obstacle avoidance path planning scheme for the rover facing the lunar south pole.On the one hand,the RGB camera is used to obtain the global scene image for preliminary obstacle detection;on the other hand,the depth camera is used to obtain the real-time environment depth,and a local map is constructed through a simplified SLAM algorithm,that is,a point cloud map is generated in real time;and based on the point cloud information,to calculate physical environment information such as road surface roughness,slope,step information,etc.The above-mentioned multimodal data combined with map data are used to update the posture and surrounding environment of the patrol vehicle in real time,and a combination of multi-modal local algorithms and obstacle detection algorithms are used for obstacle perception and obstacle avoidance decision-making,so that the patrol unit can dynamically adjust its path.The simulation results show that the autonomous navigation system significantly improves the autonomous navigation capability of the patrol unit in complex and unknown environments.