星表探索漫游器地形通过风险的快速感知方法  

Traversability Analysis for Autonomous Planetary Exploration Rovers:A Learning⁃based Approach

作　　者：谢宗武[1] 姬一明 刘阳[1] 马博宇 XIE Zongwu;JI Yiming;LIU Yang;MA Boyu(State Key Laboratory of Robotics and Systems,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学机器人技术与系统全国重点实验室,哈尔滨150001

出　　处：《载人航天》2024年第5期659-665,共7页Manned Spaceflight

摘　　要：星表漫游器是一类在地外行星表面执行探索、取样等任务的移动机器人,漫游器的高自主性、高可靠性依赖于其对自身周围地形环境的准确感知。针对漫游器自主导航对感知系统的要求,提出一种针对星表地形环境进行快速通过风险感知的方法。基于开源的星表地貌仿真系统生成逼真的虚拟数据集,同时考虑漫游器自身物理和运动学参数对地形通过风险的影响。实验结果表明:所提方法将单帧RGB图片的风险感知计算时间降低至0.1 s,相对于根据深度数据计算真实的通过风险,提出方法虽然略微损失了计算精度,但却带来计算速度的显著提升。基于神经网络的架构也赋予该方法处理跨尺度图像输入的能力,对不同分辨率的RGB图片输入都能保持较高的计算速度。A planetary rover is a type of mobile robot that performs tasks such as exploration and sampling on the surface of extraterrestrial planets.The high autonomy and reliability of a rover de⁃pend on its accurate perception of the surrounding terrain environment.To meet the requirements of the rover’s autonomous navigation perception system,a fast terrain traversability perception method for planetary terrain was proposed in this paper.An open⁃source planetary terrain simulation system was utilized to generate realistic virtual datasets,and the rover’s physical and kinematic parameters that influence terrain traversability were considered.The experiments demonstrated that the proposed method reduced the risk perception calculation time for a single⁃frame RGB image to 0.1 s.Al⁃though it slightly sacrificed the computational accuracy compared to calculating the actual travers⁃ability based on depth data,our method achieved a substantial increase in computation speed.The neural network architecture also enabled the method to handle multi⁃scale image inputs,and main⁃tained high computation speed for RGB images with different resolutions.

关 键 词：通过性分析 星表探索 地形重建 深度学习 

分 类 号：V11[航空宇航科学与技术—人机与环境工程]