Research On Motion Control And Vision Grasping Algorithm Of Wheeled Mobile Robot Equipped With Robot Arm

Robotic grasping is a very advanced and important research direction in the field of robots.The grasping of mobile robots combines the achievements of mobile robots' motion control,visual servo,deep learning and human-computer interaction,etc.,which enables robots making intelligent operation,intelligent perception and intelligent decision-making.The tasks of moving,manipulating,grasping and handling can be completed with high quality in more complex real scenes,which is an important manifestation of robot intelligence.In this paper,the self-built wheeled mobile robot equipped with a robotic arm is taken as the research object.The system construction,motion control and grasp prediction algorithm of the robot are mainly studied,including the following contents:Aiming at the imperfect omnidirectional mobile grasping robot system in the industry,this paper introduces the self-built complete system of the wheeled mobile robot equipped with a robot arm,including the body mechanical structure module,function module and drive module,power supply module and software control module,combined with the actual mobile grasping requirements for speed,torque,accuracy and endurance,etc.,parameterize the components in each corresponding module,select the most appropriate model,and integrate each module in a unified software control system,a set of high-performance mobile grasping robot system is formed.In order to solve the problem of the cooperative control of the robot system and the robot arm,this paper divides the motion control of the mobile grasping into the motion control of the moving part and the motion control of the grasping part.In the motion control of the moving part,it is mainly based on Mecanum wheel to obtain the kinematics equations of the AGV,and then designs a speed controller to control the AGV.In the motion control of the grasping part,the position control of the end of the robot arm and the opening and closing control of the gripper are mainly developed based on ROS,and the coordinated control of the robot movement and the robot arm grasping is realized under the same system.In order to solve the real-time grasping problem of robots and reduce the time of visual processing,this paper designs a network structure of encoding-decoding based on convolutional neural network.The designed network architecture is simple,the training parameters are few,and the operating efficiency of the visual system is improved.Different network parameter combinations were trained in the Cornell grasping dataset after data enhancement,and the model with the best training effect was selected.The model achieved a prediction accuracy rate of 81% in the Cornell grasping test-set.And it can be used in grasping experiments of real environment.To verify the effectiveness of the proposed grasping prediction algorithm,this paper designs a grasping experiment in a real scene.The grasping experiment uses 15 different objects common in daily life for multiple tests,and records the number of successful grasp.The overall successful rate of grasping is more than 75%,and the grasping successful rate of objects of individual shapes can reach more than 90%.The speed of grasping is fast,and the effect of 30 FPS can be achieved.The real-time performance is guaranteed,reflecting the grasping prediction algorithm,and the excellent performance.It also proves the feasibility of the algorithm in the real environment,providing data proof for the promotion of the algorithm.