Research On Obstacle Avoidance Method Of Manipulator Based On Multi-view Sensor Network For Human-Robot Interaction

Ensuring human-robot safety is the primary problem that needs to be solved in the process of human-robot interaction and human-robot collaboration.Among them,the real-time obstacle avoidance of multi-degree-of-freedom manipulators has always been one of the research hotspots in this field.The existing anti-collision algorithms of manipulators are highly dependent on prior knowledge of the working environment,lack the ability to intelligently perceive and adapt to the working environment,and cannot affect the human body or others in unstructured and dynamic scenes.Changes in the movement state of obstacles make timely and effective avoidance.At the same time,with the rapid development of visual sensing and image technology,obstacle avoidance systems based on visual inspection have gradually emerged.For single-view obstacle detection systems,there are often problems such as limited vision and self-occlusion of the human body,and cannot completely reconstruct the working space environment of the robotic arm,and cannot provide accurate environmental information feedback for the real-time obstacle avoidance algorithm.For this reason,this article proposes an improved artificial potential field method for the dynamic scene of human-robot interaction and collaboration,and proposes a robot arm obstacle avoidance algorithm,and uses a multi-view detection system to perform real-time detection and fusion processing of dynamic targets in the working environment to achieve efficient and safe obstacle avoidance.The main research contents of this paper are as follows:1)In this paper,we propose an artificial potential field path planning method that combines adaptive planning step-size,spatial location sampling and velocity repulsion function evaluation.First,perform spatial path sampling on the position of the robotic arm at the next moment to achieve an advance judgment of the local minimum area,and combine the potential field characteristics of the potential field method to set an appropriate evaluation function for the sampling position to select the best location point.Secondly,for dynamic scenarios,an improved speed repulsion function is introduced,and the speed repulsion is dynamically adjusted through speed and distance influence factors,so that the algorithm can choose different obstacle avoidance strategies according to the obstacle movement state.Finally,through real-time environmental information feedback,the environmental complexity factor is introduced.and an adaptive planning step size mechanism is proposed.Compared with other classic solutions,the improved algorithm ensures the safety of obstacle avoidance while improving the efficiency of path planning and reducing problems such as path oscillation.In this paper,simulation verification and Tiago physical platform verification of the algorithm are carried out,and the results prove that the proposed obstacle avoidance algorithm has better safety and efficiency in path planning selection.2)For human pose estimation problem in the human-robot interaction process,this paper builds a distributed multi-view human pose detection system based on the visual depth sensor,and proposes a distributed human pose estimation method based on interactive multi-models.First of all,each camera node is independently detected by the existing human pose estimation and detection method,and the results of different sensor nodes are realized by the information weight consistency filtering(ICF)algorithm to achieve data and information fusion,which solves the problem of limited single-view visual range.Secondly,in order to solve the information fluctuation caused by the estimation errors of different views in the process of multi-view fusion,this paper uses interactive multi-model(IMM)to filter and improve the estimation results of human skeleton joints after fusion,which improves the stability of the detection results.Finally,this paper uses the same human action recognition method to conduct a behavior classification comparison experiment on two different skeleton joint point data sets.The experimental results show that the behavior recognition accuracy based on the ICF+IMM algorithm is higher,which proves that the method has the higher accuracy of human body pose detection.3)In this paper,an experimental environment for obstacle avoidance of a robotic arm based on a multi-view sensor network is built.Each visual sensor node in the experimental system independently estimates the posture of the human body,and uses the hand-eye calibration results to convert the coordinate system.Finally,information fusion and filtering are performed in the base coordinate system of the robotic arm as input to the subsequent obstacle avoidance algorithm of the robotic arm.And then it performed the obstacle avoidance path planning of the manipulator.The final result of the experiment proved the effectiveness,safety and accuracy of the composite system.