Study On Formation Control And Obstacle Avoidance Method Based On Vision Guidance

In recent years,robotic technology has been developing with each passing day.Various new types of robots have emerged.No matter industrial robots or service robots have made considerable progress,just as the previous years of Boston Dynamic Robots Atlas have achieved flip and triple jump.And robotics is attracting generations of research because of its unique charm.In the robot field,mobile robots are the most common types of robots.They are also widely used.A single robot can perform the set work well in many cases,but in the specific case like target search in the fixed region,the single robot would carry out the task very inefficient,so the exploration of swarm robots has become the hot research in recent years.The formation task as the basic work of swarm robots is also receiving much attention.For the formation tasks of swarm robots systems,the technologies can usually be divided into three parts: environment perception,planning decision and motion control.The environment perception for the robot is to use the sensor to obtain state information of other robots and obstacles around itself.Meanwhile,it is also called positioning technology.For the planning decision-making level,it mainly involves how to process the perceived information to obtain the next decision of the current robot.In the formation task,it includes two aspects of work.One is to process the position information of the neighboring robots.The formation algorithm is designed to maintain the required formation shape.And the second is how to achieve effective obstacle avoidance when the robot senses the presence of obstacles.The motion control part is mainly the implementation of the upper layer decision on the underlying hardware platform.This thesis focuses on the above two aspects of environmental awareness planning decision-making.Firstly,for the spatial localization problem of robots,this thesis adopts the visual method to obtain the position information of robots and obstacles in three-dimensional space.We designed the mobile beacons and segment it by the canny operator.The P4 P problem consisting of four corner points in the beacon is solved to the spatial position of the current robot by using EPnP algorithm.In addition,the AutoEncoder is used to reduce the dimensionality of the beacon image.The KNN(k Nearest Neighbor)algorithm is used to identify the information in the beacon image.Secondly,in the design of group formation algorithm,this thesis further designs the linear formation based on the study of group behavior and the relative position information of the nearest neighbors,and expands to the design of formation control on any topological shape on the basis of this with the application of Hungarian algorithm.Finally,for the robot obstacle avoidance task,the deep reinforcement learning algorithm is developed in recent years to design the robot obstacle avoidance decision layer.In this thesis the simulated virtual environment is set up.The neural network model is under the Actor-Critic framework,and the learning of obstacle avoidance tasks is carried out by means of DDPG algorithm.Finally,with four mobile robots,we carried out the experiments on positioning,formation control and obstacle avoidance.The validity of the designed model and algorithm is further verified.