This is a fascinating piece of kit a quad copter with full autonomous flight modes and advanced machine learning camera based obstacle avoidance.

A lot of the AP type drones have these sorts of features but they are somewhat simplistic and more like flight assist than full autonomous.

This system has 13 cameras facing in ever direction with stereo pairs on every side top and bottom and one main camera it uses for imagery. It then uses a nvidia tegra processor and some custom machine learning to map it's environment and track it's intended target and plot its flight path.

As I said it's not a new idea but most of the examples I've seen were only capable of avoiding some large static object like buildings or the like and on top of that anything more than a simple follow and the tracking broke down also any fast movement or confusing backgrounds and it would lose it's target.

Check out this video showing this thing flying tracking a moving target through a fairly tightly packed forest and keeping the target in frame avoiding all the trees.

It's an impressive demonstration of the technology and it makes it look like you could have one of these following and filming you running or cycling or what ever with no need for any pilot input.

It's not a cheap thing weighing in at $2.5k which isn't exactly pocket change and I know that people that are into drones here such as myself and rob are more into the less automated racing drone end of the technology but still it's an interesting bit of kit and a great achievement to have all that running in a relatively small form factor. A great use of some neural network stuff too :D