基于RGB-D相机的四足机器人环境感知与静步态规划研究

RESEARCH ON ENVIRONMENT PERCEPTION AND STATIC GAIT PLANNING OF QUADRUPED ROBOT BASED ON RGB-D CAMERA

王祺鑫

11849007

硕士

机械工程领域工程

柯文德

2020-06-03

2020-07-20

哈尔滨工业大学

深圳

足式机器人依靠离散支撑点运动，相较于轮式、蠕动式、履带式等其它移动方式，对行走路面的要求更低，灵活性更强。足式机器人中，四足机器人在保证稳定性的同时避免了机构的冗余，是机器人研究领域的热点方向之一。本文针对四足机器人在室内环境下的应用需求，选用RGB-D相机为主要外部环境传感器，研究四足机器人的环境感知与步态规划算法。主要研究内容如下：首先，基于视觉SLAM算法分析了室内环境的特点，提出结合视觉特征点与环境中的平面特征的视觉前端，研究点、面特征的提取、匹配以及运动求解算法。根据视觉里程计特点，选用基于位姿图的后端优化算法，设计回环检测，完成RGB-D SLAM算法设计，通过实验验证了点、面特征结合能有效地增强SLAM算法的鲁棒性。其次，研究四足机器人的步态规划，分析其运动学模型，构建起四足机器人的正、逆运动学方程。以稳定裕度为衡量基准提出了一种四足机器人的非连续静步态，包括机器人的重心轨迹规划与摆动足端轨迹规划，通过实验验证了四足机器人静步态运动时的稳定性。再次，将视觉SLAM算法生成的点云地图与四足机器人静步态结合，提出了基于点云地图的机器人路径规划算法。将点云地图转化为高程栅格地图，区分机器人的可移动区域。从地形特征对落足质量影响的角度量化地形特征，设计地形代价函数，将点云地图转换为地形代价地图，设计地形代价地图上的机器人体心搜索算法，基于稳定性约束规划机器人移动路径，通过实验验证了机器人路径规划算法的可行性。

Legged robots rely on discrete support point movements. Compared with wheeled, creeping, crawler and other movement methods, legged robots have lower requirements on the ground and more flexible. Among the legged robots, the quadruped robot avoids the redundancy of the mechanism while ensuring stability, which is one of the hotspots in robot research. In this paper, the RGB-D camera is selected as the main external environmental sensor for the application requirements of the quadruped robot in the indoor environment, to study the environmental perception and gait planning algorithm of the quadruped robot. The main research contents are as follows:First, based on the visual SLAM algorithm, the characteristics of the indoor environment are analyzed, and a visual front end combining visual feature points and planar features in the environment is proposed to study the extraction and matching of point and surface features, as well as motion solving algorithms. According to the characteristics of the visual odometer, the back-end optimization algorithm based on the pose map is selected to design the loop closing and the RGB-D SLAM algorithm design is completed. The experiments verify that the combination of point and surface features can effectively enhance the robustness of the SLAM algorithm.Second, study the gait planning of the quadruped robot, analyze its kinematics model, and construct the forward and inverse kinematic equations of the quadruped robot. Taking the stability margin as the benchmark, a discontinuous static gait of a four-legged robot is proposed, including the trajectory planning of the center of gravity of the robot and the trajectory planning of the swinging foot end. The stability of the four-legged robot during the static gait movement is verified through experiments.Thirdly, combining the point cloud map generated by the visual SLAM algorithm with the static gait of the quadruped robot, a robot path planning algorithm based on the point cloud map is proposed. Convert the point cloud map to an elevation grid map to distinguish the robot's movable area. Quantify terrain features from the perspective of the impact of terrain features on footing quality, design terrain cost functions, convert point cloud maps to terrain cost maps, design robot body center search algorithms on terrain cost maps, plan robot movement paths based on stability constraints, The feasibility of the robot path planning algorithm is verified through experiments.

四足机器人 RGB-D SLAM 静步态规划 路径规划

quadruped robot RGB-D SLAM static gait planning path planning

中文

联合培养

南方科技大学

王祺鑫. 基于RGB-D相机的四足机器人环境感知与静步态规划研究[D]. 深圳. 哈尔滨工业大学,2020.