nVidia makes a 'Jetson' line of embedded computers with advanced graphical processing units (GPUs) designed for vision processing and artificial intelligence.   The capabilities of the CUDA GPU are quite impressive and when you run the demo software that's included it's clear that they are far beyond the capability of an !RPi.  The computers run the linux operating system (like the !RoboRio and Raspberry Pi) and are compact enough to mount on a robot and give it serious vision processing capability while not bogging down the !RoboRio.  

[[Image(Jetson_TK1.jpg, 240px, right, margin=10)]]
Team 2537 has a number of older Jetson TK1 computers running Ubuntu linux as trainers and recommends using the Jetson Nano (or newer) for use on robots.  For development, you'll want to use a USB keyboard/Mouse and HDMI monitor.  On your robot, the Jetson should be connected to the !RoboRio using Ethernet.  2537 Jetsons also make a remote graphical desktop available using RDP so you can connect to the computer remotely (over Ethernet).  Jetsons are conveniently powered from 12Vdc.

Jetsons' CUDA GPUs make them great tools to learn the premier graphical processing software: OpenCV (open computer vision) which is used to pre-process (segment) images prior to intelligent processing.  It also enables use of fast-powerful artificial intelligence tools such as convolutional neural networks (CNN) to interpret images.

Unlike the Raspberry Pi which stores its operating system on a removable micro-SD card, the Jetsons include built-in flash memory to store their operating system and your software.  You can re-flash (load) new software into the Jetson if you need to start with a fresh copy of the OS, CUDA, cuBLAS, OpenCV, and a host of other tools and demo software; for re-flashing, you will want to have a USB-to-Serial adapter such as [https://www.amazon.com/Sabrent-Converter-Prolific-Chipset-CB-DB9P/dp/B00IDSM6BW this one].  2537 Jetson TK1s are in 3D-printed cases; the case has room for an internal 2.5" SSD which is well worth adding since even a small SSD offers a lot more storage.

For detailed documentation see: [https://www.dalbert.net/?p=763 here]