Intelligent Takeout Robot System Research

In recent years,the volume of goods and logistics in the new retail mode has been escalating.In this paper,a low-cost solution is designed for takeout distribution industry,which is based on Robotics and driverless technology.The core research and work of this paper include:(1)VGG-16-FCN-8s full-convolution codec semantics segmentation network is designed to detect the current road,with an average accuracy of 86.38%.Its generalization ability is far superior to the lane detection algorithm using statistical probability Hough transform and linear regression.(2)The realization of Faster R-CNN target detection network based on VGG-16 is studied.The average accuracy of 80.62% is achieved on the test set,and the detection speed of 5fps is achieved on the existing hardware of the vehicle.The application of Lab color model in traffic light recognition is compared with the traditional HSV detection algorithm in sunny,cloudy and rainy days.(3)The data sets of lane semantics segmentation and obstacle target detection are built,and the existing large open data sets are combined to train the neural network.Compared with the single data set,the accuracy of the multi-segment test road is improved by 0.8%~2%.(4)The vehicle model,thermal insulation box and unlocking device are designed.For software,Web ordering,WeChat and message notification,task management and key frame image object storage system are developed.Camera and 2D lidar data are used for environmental sensing,ultrasonic and laser rangefinders are used for obstacle avoidance perception.After the actual campus and surrounding multiple short-distance road verification,the whole system designed in this paper can complete the distribution process autonomously and return to the starting point.