MIE.41 – Proposal for Automated Handling of Radioactive Accelerator Target Material

• Murad Basrawi
• Diego Da Silva Gomes
• Camila Maeses
• Carlos Martinez
• Alan Tieng

The handling of radioactive and other hazardous materials poses concerns for health and safety of personnel. Inspection of radioactive target material at Fermilab has not been approved at present due to those concerns. The use of collaborative robots (COBOTs) is a way to move and inspect material without direct human intervention. Current teleoperated systems require specialized personnel because direct manipulation is a difficult and slow process. Our goal was to develop a method of inspection that addressed the safety concerns of current handling methods, was automated, and was adaptable to future Fermilab robotics projects. Here we designed and simulated an automated pick and place system by combining a 6-DOF robot arm, computer visualization, and machine learning. Using Robot Operating System (ROS) as the base, we have scripted a control system that is modular and can be adapted to multiple robot platforms and configurations. Unity, a physics capable game engine, was used to both generate synthetic datasets for machine learning and simulate the pick and place process. Domain randomization was used to randomize our scenario when generating our training dataset for a more robust model. A model for identifying the position and orientation of the material was then trained in Pytorch. The model and control scripts were then adapted for use in ROS on real hardware: a Universal Robots model UR3e equipped with a mechanical gripper and wrist camera. Our methodology will allow Fermilab to train and test robot systems in simulated scenarios and easily transfer work over to ROS with real hardware. The proposed robot system can be used to inspect target material to determine if its lifecycle can be extended. The robot system can also reduce costs for fabrication, storage, and disposal. Automation of the material handling also reduced the time of the process and eliminated the need for specialized training to perform. Most importantly, personnel will never directly interact with the material, improving safety during handling.