This is the Super-Computer-On-Legs, or "SCOL" robot by NVIDIA. It uses a Jetson TK1 linux dev board featuring a Tegra K1 with 192 CUDA cores that give upto 300 GFLOPS (giga-flops per second) for under $200 in just a few Watts of power! It's currently booting up into Linux for Tegra that is basically Ubuntu 14.04. Then the robot will perform video-stabilization & face detection to walk towards the nearest person that it sees.
Both green LED's turn on if the person is straight ahead or just 1 LED turns on if the person is to the side of the robot so it will start turning soon. You'll notice that the reaction times are very slow, because its currently just using simple Arduino walking scripts that only reading the LED data at the end of each step, so there can be upto 1 or 2 seconds of lag.
A robot that can walk on 2 feet towards a person that it sees would have cost atleast $100,000 just 5 years ago and would probably take several man-years for a team of robotics engineers to get it working, since the onboard computer & battery system would require a walking robot that's roughly the height of a human, and thus was very expensive & complicated. But now that the computer can be so small, light & low-power, even a tiny & cheap & simple humanoid robot can do this for under $1000! In fact this is a completely open-source project, the full instructions to make this robot will be posted as a tutorial on the Jetson TK1 wiki, where you will be shown how to build this entire robot for about $1000!
What is the purpose of this robot? To show a small proof of concept for what is possible very easily & cheaply with a small Jetson TK1 based robot. The general rule of thumb I've always believed for robotics is that if you want to make a robot that is 2x larger then it will be about 5x more expensive, about 5x more complex, and take about 5x longer to build. If you believe this then you'll understand why it is so much easier to make a small robot compared to one that is 10x bigger! Because a bigger robot will be heavier and thus requires bigger motors and bigger mechanical frame and those will add weight too requiring even larger motors, and all this requires bigger batteries to power the bigger motors, and bigger motor controllers & safety systems, and then you'll have more problems of electrical interference, feedback loops, component tolerances, wear & tear, heat issues, friction & mechanical issues, difficulty with reliable fasteners, etc, etc.
That's why small robots advance at such a rapid rate and are so cheap & easy to build compared to human-sized robots trying to perform the same tasks. But small robots can only carry small lightweight low-power processors, so just 1 or 2 years ago they were limited to under 50MHz microcontrollers or occasionally FPGAs that are very tedious to develop. Now with cheap low-power ARM CPUs around the 0.5 to 10 GFLOPs range (such as Rasbperry Pi, BeagleBone Black, O-DROID, etc), it opens a whole lot of new possibilities in small robots. But with NVIDIA's new Tegra K1 processor in the 300 GFLOPs range while still being fairly easy to program and still similar dimensions & power draw, it will open a whole new era of small advanced robots!
