Using Robotics to Create a Cleaner Environment

Littering is a prevalent issue that disrupts natural environments, endangers wildlife, and stunts plant growth. Due to the scale of litter that is not properly discarded, autonomous trash collection can make current litter collection efforts more effective. In this project, we designed an autonomous robotic system for a Jackal robot to detect, retrieve, and deposit discarded cans. The approach involved AI image detection and inverse kinematics. The Jackal robot and its functions are compatible with Robot Operating System (ROS), facilitating communication between the Intel® RealSense™ camera and the inverse kinematics ROS topic. This allows the robotic arm to locate and grasp the Coca-Cola cans. We implemented object detection by building a Coca-Cola can image database in Roboflow and training a machine learning model for object detection in YOLOv5, which reached 89.9% accuracy. Many diverse environments and types of Coca-Cola cans were used in the YOLOv5 training data, allowing the model to detect many types of Coca-Cola cans in different backgrounds. The Coca-Cola can images were taken by the Intel® RealSense™ camera which has two sensors: the RGB camera and the depth sensor. The RGB camera measured the X and Y coordinates of the cans on a 2D plane, and the depth sensor found the Z coordinates to calculate the distance to the Coca-Cola Cans. The data was then published to a ROS topic, where an inverse kinematics script calculated the desired angles of the arm joints. The arm sits on top of a turret base plate that allows the arm to rotate in 5 degrees of freedom and can reach up to 40 cm away from the Jackal. There were many factors to consider while making the robot arm. It needed to lift and deposit Coca-Cola cans from the ground to the trash bin attached to the top of the Jackal. The cans could be grasped in any given orientation or state, such as upright or lying down, crushed or not crushed. Additionally, the arm segments needed to be long enough to reach the cans and dispose of them in the trash bin and light enough for the servo motors to withstand their weight. We first did static analysis to determine the worst-case torque calculation at each joint to identify the appropriate arm segment lengths. With dimensions established, 3D models of each part were designed using SolidWorks and printed using PLA and PETG filament. Due to manufacturing defects, the Jackal could not move around autonomously, and we were not able to implement and test autonomous search algorithms. However, we were able to have the YOLOv5 model, inverse kinematics script, and the robotics arm work together to detect and grab Coca-Cola cans given the arm's range of freedom and information from our ROS system.