FRC 2473 Presents at NVIDIA
One of 2473’s primary goals for the 2016-17 season was to effectively integrate various AI features into our robot. In pursuit of this goal, we added computer vision to our robot by way of a Raspberry Pi. We saw varying degrees of success over the year, but the final product worked moderately well. As offseason approached, we put effort into seeing how it could be improved.
Over the past summer, three of our members (Abhinav, Shruthi, and I) were selected to intern at NVIDIA. The company has historically been known for building powerful gaming GPUs but as of late have been venturing into robotics as well as the emerging AI field. Over the course of the summer, we all got familiar with NVIDIA’s latest top-of-the line processor; the Jetson, intended for running neural networks as well as intense computational tasks.. We gained valuable experience and used this to replicate the success we had on the Raspberry Pi; only this time, we were working purely on the Jetson. As many of you know, the FRC autonomous challenge last year included a task where the robot could autonomously drive and place a gear on a peg (FRC Steamworks 2016-2017). We used vision to auto align to the peg and accurately drive towards it. Since we were only using a Pi last year, we were forced to only analyze one frame and drive the robot to the peg in one shot since the Pi was so slow at doing the computations to run our Computer Vision. The powerful Jetson allowed us to analyze frames right as they came in, so we could process our data in real time and have our robot adjust to nearly anything it ran into.
What you just read was what we presented at NVIDIA yesterday. We showed them our robot and how exactly we implemented the Jetson on it. Employees stopping by were fascinated to see how this group of high schoolers had used the Jetson to make their robot smarter. Their feedback and advice was extremely insightful.
I would jump at the chance to present at NVIDIA any day. I got to build an autonomous, self-driving delivery bot, learning a lot in the process, and was able to bring all that back to our robotics team. They have even been kind enough to provide us with a Titan X GPU which will help our team do more Deep Learning in the future to make our robots even smarter. Plus, there was really delicious food and swag which made everything even better.
Mokshith Voodarla, VP of Cupertino Robotics