Research On Pedestrian Perception And Navigation For Mobile Manipulation Robot

Mobile operation robots take into account both mobility and operability.They not only have functions such as mobile robot construction and navigation,but also can use robotic arms to perform tasks such as grasping.This is a hot topic in the field of robotics.However,for complex tasks in complex environments,it is difficult to achieve good results with a single sensor,and the combined use of multiple sensors can better complete the task.This article focuses on mobile operating robots,using laser and vision sensors to conduct research on mapping,sensing,and navigation.First,a set of integrated SLAM framework was established.Based on the ORB-SLAM2,scanning matching using laser was used to provide a priori motion information for the visual front end,and a point cloud thread was added.Compared with the original visual SLAM,the positioning accuracy during the mapping process is improved,and the obtained grid and octree map can be used for subsequent navigation and motion planning.Then using laser detection method based on Adaboost and visual detection method based on YOLO v3,a set of pedestrian detection framework in the environment is constructed,and the position and speed of pedestrians within the sensor's visible range can be obtained.Tested in the simulation environment and the actual environment respectively,and verified its good perception effect.Furthermore,based on the comfort space of the human body,a value-appraisal rule around pedestrians is proposed.Using the customized pedestrian perception layer as the carrier,the pedestrian detection results are integrated into the multi-layered cost map for navigation.Experiments have proved that the robot can treat humans and ordinary obstacles differently,and the generated trajectories are safer.Finally,in order to solve the problem that the target pose of the mobile operation and the navigation pose of the mobile base are difficult to match,a method using a reachable map is comprehensively considered to calculate the manipulator's capability score to determine the goal of the navigation.The comparison with empirical methods has proved its superiority.