What you're looking at above appears to be a pretty big breakthrough in artificial intelligence -- an avatar "learning" how to dribble in real time through AI. DeepDribble is a project from researchers at DeepMotion and Carnegie Mellon, and the idea is to demonstrate how an AI-controlled avatar can acquire the complex and subtle set of integrated physical skills and coordination that humans take for granted:
Basketball at its peak is known for intricate ball handling, quick pivoting, and fake outs. A good player will fluidly transition between have a variety of dribbling skills. In order to recreate this, the trained characters need to be able to transition between separately trained behaviors... Libin Liu and Jessica Hodgins use the same methods for training single skills to create a multi-skill control graph (which allows transition between motion fragments of two different skills). The training is done incrementally, behavior by behavior, to ensure integration occurs with proper transitions for each skill. The result is a player that can perform multiple ball handling skills in a variety of orders, changing course with seamless blending.
"First we build a bio-mechanical model of the digital character, using physics to simulate bones and virtual muscles," Deep Motion's Libin Liu tells me. "We put the digital character in the physically simulated virtual world and let it mimic how real players perform the motions using their muscle forces."
The training process, as you might have imagine, requires quite a lot of power -- but Liu says the learning arc took about as long as a single televised NBA game -- or even half of one: