面向降频感知平台的视觉-惯性-轮式里程估计算法  

Estimation Algorithm of Visual⁃inertial⁃wheel Odometry for Down⁃frequency Platforms

作　　者：贾慎涵 谢心如 黄煌[2] 王越[1] 熊蓉[1] JIA Shenhan;XIE Xinru;HUANG Huang;WANG Yue;XIONG Rong(Zhejiang University,Hangzhou 310027,China;Beijing Institute of Control Engineering,Beijing 100094,China)

机构地区：[1]浙江大学,杭州310027 [2]北京控制工程研究所,北京100094

出　　处：《载人航天》2024年第5期676-683,共8页Manned Spaceflight

摘　　要：受地外能源稀缺性的影响,探测器通常降低传感器频率以限制功耗。针对降频传感里程估计系统中存在的惯性先验退化问题,进行了基于轮速观测的先验融合以及面向低速平台的视觉算法改进。基于相机频率降为1 Hz,惯性测量单元(IMU)频率降为4 Hz的探测器地面样机展开实验。为实现降频感知平台的高精度里程估计,提出一种基于多状态约束卡尔曼滤波(MSC⁃KF)的视觉-惯性-轮式多传感器融合算法,并通过相机外参在线标定算法实现探测器长时间运行中自动修正系统参数,最后使用零速检测和状态保持等方法进一步抑制机器人静止模式的状态退化。在公开数据集进行充分验证后,基于探测器地面样机在地外复杂环境模拟场地进行了实验,结果表明:可以实现1.2%以内的累计误差,在降频感知平台上实现了高于现有最优方法的里程估计精度,能有效应用于地外探测器的自主导航功能实现。High⁃precision odometry is essential for autonomous rover exploration on planetary sur⁃faces and it is a current research hotspot in robotics.Due to the scarcity of extraterrestrial energy,rovers usually reduce sensor frequencies to limit the power consumption.To address the degradation problem of inertial priors in low⁃frequency odometry systems,we studied the fusion of wheel speed observations and the improved visual algorithms for low⁃speed platforms.Experiments on a rover pro⁃totype with a reduced camera frequency of 1 Hz and IMU frequency of 4 Hz showed that the inertial sensor priors degraded.To achieve high⁃precision odometry on a low⁃frequency platform,a Multi⁃State Constraint Kalman Filter(MSC⁃KF)based multi⁃sensor fusion algorithm integrating visual,in⁃ertial,and wheel data was proposed.An online camera extrinsic calibration algorithm automatically corrected the system parameters during prolonged operations.Additionally,zero⁃velocity detection and state maintenance further mitigated the state degradation in stationary modes.Extensive valida⁃tion on public datasets and numerous experiments in a simulated extraterrestrial environment using the rover prototype demonstrated that the proposed method achieved an accumulated error within 1.2%,which surpassed the existing methods and could effectively support the autonomous naviga⁃tion for extraterrestrial rovers.

关 键 词：里程估计 降频感知 多传感器融合 卡尔曼滤波 

分 类 号：V19[航空宇航科学与技术—人机与环境工程]