The thesis addresses the development of a robotic cell for the purpose of separating packaging waste. The robotic cell consists of a robotic manipulator, various grippers, a conveyor belt, and a camera for determining object grasping points and classifying the packaging. The robotic manipulator used in the application is a collaborative robot UR10e, capable of exchanging two different grippers: a classic two-finger gripper by Robotiq and a soft gripper with a vacuum suction. The conveyor belt is used to move objects from the camera to the reach of the robotic manipulator. It is controlled by a frequency driver, managed via a programmable logic controller (PLC).
Our work involved setting up and controlling the robotic manipulator, integrating and controlling the conveyor belt via the PLC, determining grasping points with the help of a camera and various methods, and combining all individual functionalities into a fully operational system. For successful object separation, it was necessary to classify the objects based on different material types. This classification was developed by an external company and integrated into the system to provide information on which bin the robot should place the gripped object.
In the following sections, we describe the individual functionalities integrated into the main computer system. We used the Robot Operating System (ROS), which allowed a modular design of the entire system. Each functionality was implemented as a separate node, communicating with other nodes via standard protocols with predefined message structures. Hardware integration was carried out using the TCP/IP transport layer.
The robotic manipulator is controlled via the \emph{ur-rtde} software library, which uses RTDE (Real-Time Data Exchange) to control the robot in real time. The library was developed at the University of Southern Denmark (SDU), where the robotic cell development also took place. To integrate the library into the system, we developed our own class, serving as an integrator of the library with additional functionalities customized to our system. This class was then included in the node responsible for robot control.
The PLC controlling the conveyor belt communicated with the main computer via the OPC UA protocol. The conveyor belt's movement was monitored with an incremental encoder, providing information on the relative distance traveled by the belt. We developed our own class for the communication protocol, reading and writing data to the PLC, which was implemented in an independent ROS node similar to the robot control.
In the last part of the thesis, we present and test the methods used to determine object grasping points. For system development, we started with a manual method for determining grasping points, which was later upgraded with two automatic methods. The manual and baseline automatic method allowed the use of both grippers, while the second method, based on a deep neural network, only supported the classic two-finger gripper. We concluded that the manual method was completely successful as it includes the human factor for selecting grasping points. The automatic methods were tested on various objects commonly found in packaging waste. We conducted the same number of test grips with the gripper that proved to be more successful for each method to evaluate them.
In conclusion, we summarize the results of the work, evaluate the set goals, and mention the advantages and disadvantages of the system. We also touch on possible improvements to address system issues and limitations, and finally highlight the potential for further development.
|
