Research On Safety Strategy Of Human-robot Cooperation Based On Human Motion Recognition?

With the development of industrial robots,more and more robots begin to walk into factories to take place of manually and highly repetitive work.The new model of human-robot collaboration will break the boundary between traditional industrial robots and workers,becoming the future direction.One of the major problems to be solved when collaborative robots are introd uced into actual production scenarios is safety.In this paper,an observation system is established for the collaborative environment based on the Kinect camera.By fusing the information from multi-kinect sensors,an accurate skeleton model of people in the environment can be performed.According to the skeleton model performed before,we design a human-robot autonomous collision avoidance model.At last,we establish dynamic security strategies based on the security collaboration standard ISO/TS 15066 and make a study of the method of autonomous collision avoidance.The human-robot collaboration system is a typical non-standard system,and different collaboration scenarios correspond to different sensor configuration s.According to the experimental scenarios defined in this article,an appropriate multi-Kinect configuration method is designed.Based on the parameter characteristics and drive requirements of the hardware sensors,we build a multi-camera information communication system with client + server mode using the UDP communication protocol.In order to unify the observed coordinates of the object from different cameras,we use improved ICP algorithm to calibrate the exterior parameters of the multi-kinect system.Considering the human-robot interaction,we complete the conversion of coordinate of different kinect camera and the registration of the camera coordinate and the robot coordinate to ensure the real-time collection of human posture in the scene.The purpose of introducing multiple cameras is to solve the problem of low accuracy of joint position recognition caused by the problems of single camera shake,human body occlusion,human-robot mutual occlusion,etc.Therefore,this paper studies the problems of single camera recognition of human mot ion,improves the recognition accuracy of human joint by fusing information from muti-kinect sensors.First,assign weights to the acquired joint position information through the tracking status of the camera's built-in SDK and the relative positional relationship between the human and the camera.Secondly,we make the constraint to the joint position information by human physiology and kinematics,mainly including the invariant characteristics of human bone length and the limited range of joint motion.Finally,the Kalman filter algorithm combined with bone length invariance and joint motion limitation is used to predict the position of untracked joint points.Considering that the improvement of the efficiency of collaborative robots should be based on the dynamic interaction between robots and human,this paper studies the safety standard of collaborative robots ISO/TS 15066.Based on the safety collaboration model of speed and separation distance determined in this standard,we build a cylinder-enclosed human-robot collision avoidance model,and propose a collision detection algorithm.We simplify the calculation formula of the minimum safe cooperation distance determined by the cooperation standard,Combined with the RRT algorithm(rapid random expansion tree motion planning algorithm),we propose a safety strategy for the intraction between robot and human in the collaborative space to avoid collisions.Finally,the performance of the multi-sensor data fusion of the vision system is verified through experiments,and the effectiveness and dynamic capabilities of the safety strategy are verified in the simulation environment ROS.