Design and Fabrication Warehouse Robot Delivery Device with Machine Learning for Obstacle Avoidance

Abstract

Increased use of Industry 4.0 and smart logistics technologies has forced the warehouses to adopt autonomous transport mechanisms for material transport and minimizing the human effort and labor. This paper presents the development of a six-wheeled, all-terrain Warehouse Delivery Robot using six 12 V, 60 RPM Johnson geared DC motors designed to carry goods through a warehouse floor without encountering obstacles. An ESP32 microcontroller is employed as the main component of the embedded controller responsible for motor drive control, communication with a computer and sensor data integration. A Python application developed on the computer is used to command destination and movement of the robot via a Wi-Fi network. Obstacles are recognized using a combination of TensorFlow/YOLO object recognition algorithm (pre-processed using OpenCV library) that recognizes pallets, boxes, employees and other objects from images captured from the onboard cameras along with ultrasonic range detection mechanism which helps the robot to slow down, turn or change direction without stopping due to an obstacle ahead. Path planning is also done by an occupancy grid map of a SLAM style created using ultrasonic sweep and odometry which are then analyzed using A* shortest path search algorithm to find the shortest way from the robot to a desired location before taking over with the reactive part of the system. The bench and floor tests of the developed robot showed successful obstacle recognition, closed loop performance of motors and comfortable torque margin for 10 kg total weight of the robot and payload