Font Size: a A A

Path Planning And Target Detection And Localization Of Mecanum Robots In Unknown Environment

Posted on:2021-04-01Degree:MasterType:Thesis
Country:ChinaCandidate:C L LuFull Text:PDF
GTID:2428330632457497Subject:Engineering
Abstract/Summary:PDF Full Text Request
Service robots execute tasks in the environment that target location may change.The robot is required to acquire environmental data autonomously in a changing environment,and then performs path planning which is critical for mobile robots to complete the task.The intelligent mobile robot applied in path planning is mainly the differential mobile robot at present.When approaching the target point,the differential mobile robot has a slower speed and lower efficiency in the transverse error adjustment.Because of the omni-directional mobility of the Mecanum wheel robot,the efficiency of path planning can be effectively improved.In the local environment,the robot can only plan its own path according to the environmental data obtained by some sensors.How to effectively use data for path planning and target detection in the unknown environment becomes a key issue.In this thesis,the robot is trained to carry out path planning without maps by means of reinforcement learning,and to detect and locate specific targets after reaching the target point.Firstly,the Mecanum wheel mobile robot system is established,and kinematics simulation are conducted in the V-REP physical simulation environment,along with physical experiments to verify the accuracy and reliability of the model.The shape and kinematics model of the mobile robot are built based on V-REP,and the visual sensor,radar,speed odometer and other sensors are added in the model.The sensors' data collected by the robot in real time through remote API(Application Programming Interface)are used as the robot's environment information.The above contents establish the foundation for path planning and target detection and localization.Secondly,a model for reinforcement learning is established.In this thesis,the A3C(Asynchronous Advantage Actor-critic)algorithm is used to train the motion decisions of the mobile robot.According to the characteristics of the Mecanum wheel mobile robot,the local environment radar data of the robot is taken as input,and the speed,angle and angular velocity of the robot are taken as outputs,which establishes an end-to-end training model.Experiments are carried out in the environment built by V-REP to verify the effectiveness of the algorithm.A training method based on the minimum distance is adopted,which solves the problem that mobile robots waste most of their training in non-obstacle avoidance.The training method also reduces the training time,optimizes the establishment process of state space,and improves the training efficiency.Compared with the common reinforcement learning method,the minimum distance training method is more efficient under the same training times.An experiment on a homemade Mecanum wheel robot is carried out to verify the effectiveness of the algorithm in path planning in a real unknown environment.Thirdly,the images of target objects are collected in the simulation environment and the real environment respectively,and the images are expanded.An improved YOLOv3-Tiny network is proposed to ensure the detection speed and improve the detection accuracy.The models of the target objects are trained to obtain category and position in the image.Finally,based on the attitude and position of the car body,convolutional neural network detection results and radar data,this thesis proposes an algorithm to calculate the target position in the world coordinate system.According to the calculated results,the experiment about distance and error is performed,and the experimental results show that shortening the distance between the robot and the target can reduce the positioning error.
Keywords/Search Tags:unknown environment, mecanum wheel robot, path planning, target detection, reinforcement learning
PDF Full Text Request
Related items