Path Planning Of Meal Delivery Robot Based On Visual Servo

The closed-loop control of the robot is completed by using the information obtained by the visual sensor,which is called the visual servo control of the robot.Robotic visual servoing is a new way of robot positioning and navigation.There are three main methods of visual servoing,one is image-based visual servoing,the other is position-based visual servoing,and finally the visual servoing is a mixture of the two methods.This paper studies an image-based visual servo system.The food delivery robot is equipped with a visual sensor for real-time image acquisition and processing,and the environmental information is fed back to the robot control system.meal destination.The research content includes the following parts:First,the software and hardware platform of the food delivery robot visual servo was built.The hardware platform of the visual servo of the food delivery robot includes the main control module of the embedded main controller,the power module,the motor drive module,the pose detection module and the image acquisition module;the software platform includes the overall scheduling module of the software system,the human-computer interaction module,and the image processing module.and path planning module.Secondly,in order to ensure the reliability of the data,the vision sensor is first calibrated by Zhang’s calibration method.Then use the corrected image for image processing.The image processing steps are mainly divided into two parts.In the first part,the experiments compare and analyze three mainstream deep learning target detection algorithms: Mask R-CNN,SSD,and YOLOv5.The experimental results show that the YOLOv5 target detection algorithm has the highest detection accuracy and relatively fast detection speed.Considering the design comprehensively,an obstacle target detection algorithm based on YOLOv5 is proposed.Combined with Kalman filter,real-time state prediction is realized for the position of obstacles.The second part is the detection of the driving area of the food delivery robot.For the collected images,the Sobel edge detection algorithm is used to detect the edges of the lane lines of the campus driving roads.Then,after a series of operations such as perspective transformation and positioning and fitting are performed on the detected lane lines,the extraction and identification of the lane lines are completed.Finally,the surrounding environment map of the food delivery robot is constructed by combining the detection results of obstacle objects and the detection results of road sidelines.Then,the food delivery robot path planning system was designed independently.According to the surrounding real-time changing environment,the dynamic path planning of the food delivery robot is realized.In order to choose a suitable path planning algorithm.The experiments compared and analyzed common path planning algorithms such as Dijkstra,A* algorithm,curve difference method,artificial potential field method,D* Lite,etc.Finally,D* Lite algorithm was adopted as the real-time dynamic path planning algorithm of the food delivery robot.Experiments show that the designed food delivery robot path planning system based on the D* Lite algorithm can perform dynamic path planning according to the surrounding real-time changing environment.Finally,a food delivery experiment of a visual servo food delivery robot was carried out.In the experiment,a simulated meal delivery environment was constructed in the school playground.The simulated meal delivery experiment of the meal delivery robot was carried out by using the simulated meal delivery environment.The experiment tests the obstacle recognition module and path planning module of the food delivery robot.Experiments show that the designed visual servo food delivery robot can be competent for the campus food delivery task.