2019 Abstract from Molly and Alex

Target Hall 3D Visualization Robot Project

Student Researchers: Molly Wehrenberg (Northside College Prep, Chicago)

and Alex Jang (Naperville North High School)

Scientist Mentor: Katsuya Yonehara

As part of the Neutrino Main Injector facility at Fermilab, the Target Hall is located approximately 40 meters underground and is the access point to the Pre-Target tunnel and decay pipe walkway of the NuMI beamline. Target Hall, itself, houses the target and focusing horns which convert a high energy proton beam into a focused neutrino beam. The Target Hall is a high radiation area, especially when in use. Thus, required maintenance and monitoring of the hall is dangerous and often impossible for humans to complete. To prevent unnecessary exposure to radiation and increase laboratory efficiency, a 3D visualization robot that could autonomously perform maintenance tasks, such as tightening screws, was proposed to us this summer.

In order to simulate the functions of the proposed robot, we used an NVIDIA Jetson TX2— a computing device that has a comprehensive AI development platform. The software packages from the NVIDIA development kit make it possible for the TX2 to perform basic AI functions, such as image classification, object detection, and segmentation. In order to effectively recognize and locate a screw in the Target Hall, we focused on developing a custom object detection model for screws. Using NVIDIA DIGITS, a deep learning GPU training system, the TX2 learns to identify screws by processing through an image database. Using a live camera feed with the trained TX2, the computer can not only identify screws within its sight, but it can also return the exact coordinate location of each screw. In the future, the Target Hall robot will process these coordinates for moving and performing tasks.