Simulation Research On Autonomous Door Opening Strategy Of Mobile Manipulator

With the development of mobile manipulator technology,it will play a more and more important role in the human living environment,such as family service,medical care and so on,and the application of robots in these fields inevitably needs to solve a problem is to open the door.Therefore,this paper will focus on the target recognition,control and motion planning of mobile manipulator in the working environment.The type of the door is ordinary single door for household and office use.There is no automatic closing device such as spring on the door.The door handle is a long horizontal bar,and the door is opened by pushing the doorFirstly,the virtual hardware and software system of mobile manipulator are built in ROS and Gazebo.The base has three degrees of freedom equipped with laser range finder and RGBD camera,UR5 six-degree-of-freedom manipulator and equipped with six-axis force/torque sensor and two-finger gripper.At the same time,ROS is used to manage the software system of the robot,including sensing,control,navigation,task planning,etc.,which is responsible for the communication and interaction between modules.Secondly,the outline of the door is extracted according to the point cloud data obtained by the laser range finder,and on this basis,the handle is positioned in the outline of the door.Using the environmental information sensed by the laser range finder,the point cloud is segmented and clustered through the point cloud processing algorithm,a set of possible candidate doors are generated and geometric attributes are calculated,and candidate doors that do not meet the requirements are eliminated.On the basis of the detected candidate doors,the detection range is reduced,and the handle is detected based on the double distribution of point cloud intensity and curvature.In order to improve the robustness of the handle detection,the RGBD camera is used in parallel to detect the handle based on the YOLOv3 target detection algorithm.Take the average of the handle position coordinates obtained by the two methods when the requirements are met.Then,using hybrid position/force control,the autonomous door opening task is completed through coordinated motion planning of the manipulator and the base.Since the end effector of the manipulator always grasps the handle when the door is opened by the manipulator,the robot's degree of freedom is restricted by the door and requires compliance,this paper chooses the hybrid position/force control.Due to the limited working space of the manipulator,in order to open the door to an opening that allows the robot to pass through,the base and the manipulator need to move together.This requires the base to move forward while the manipulator pushes the door.In this paper,the corresponding motion planning and control methods are given.Finally,on the built simulation platform,using the environmental information sensed by the laser range finder and RGBD camera,the door and handle detection algorithms are used to complete the door and horizontal strip handle recognition and the simulation experiment of pushing the door autonomously.The obtained door and handle detection results and the robot's autonomous door opening behavior verify the effectiveness of the door and handle perception algorithm and door opening strategy.