Some Projects on Nvidia Jetson TK1 Development Board: Nintendo Emulator, USB3 Webcam, and Robotics
Nvidia Jetson TK1 is a development board powered by the company’s Tegra K1 quado core Cortex A15 processor, and especially a Kepler GPU that allows for OpenGL 4.4. It has shipped to developers around April/May, and some of them have showcased their projects, or tested some hardware.
Dolphin is an emulator for Nintendo GameCube and Wii console that supports full HD (1080p) rendering, and run on Android, Linux and Mac OS, and there’s also an Alpha version for Android. Ryan Houdek (Sonicadvance1), one of Dolphin’s developers, has leveraged Kepler’s OpenGL support via Nvidia’s GPU drivers, to port the emulator to the platform running on Ubuntu, but it should work as well on Tegra K1 hardware running Android such as XiaoMi MiiPad tablet. You can watch Mario Kart: Double Dash demo running at full speed on the Nvidia board below. According to the developer, such framerate would be not achievable on Qualcomm 800 because “Adreno Graphics Drivers are grossly inefficient compared to the TK1”.
The latest version of Dolphin for Android (Beta) dates December 7, 2013, so I’d assume the optimizations shown above are not available right now. You can find more demos on Ryan Houdek’s YouTube Channel.
Another developer, Korneliusz Jarzębski, has tested e-con Systems USB3 See3CAM_80 HD camera connected to the board’s USB 3.0 port, and using the camera’s “See3CAM” application. I understand that all that needed to be done was to enable hidraw for USB devices in the Linux kernel, and it just worked out of the box. The application can perform real-time video processing, applying videos filters (invert, particles, etc..), as well as changing image characteristics such as brightness, contrast and so on.
You can find a little more on his blog (Polish).
The last demo I’ll show today is a robot powered by Jetson TK1 board that can walk to the nearest person it can see. The robot detects person via a camera and GPU accelerated face detection (about 5 times faster than CPU-only face detection). Beside better performance, the robot is pretty power-efficient as it only draws about 7 watts, and last about 45 minutes powered by a small LiPo battery. The robot was showcased at the Robotics Science and Systems Conference last month, and while attendees were impressed by the performance and power consumption, they still noticed the board was a bit too big for most robots, especially quad copters. But the platform clearly has potential, and Shervin Emami, the person behind the project who happens to work for… Nvidia, mentioned work is being done on smaller Tegra K1 computer on modules that be installed in a custom motherboard of a robot without unnecessary ports.
If you are interested in seeing more projects running on Jetson TK1 development board, you can consider following “Embedded Tegra & Jetson TK1 Blog” on Google+.