Researchers Teach AI How To Dribble A Basketball Like An All-Star Point Guard
Do you need any more proof that your ball handling skills suck? Well, look no further than Carnegie Mellon University and DeepMotion, which together trained AI how to dribble a basketball and pull off increasingly advanced moves as it learned.
It can be hard to often rationalize in our minds that professional basketball players who make dribbling, crossovers, and pump fakes seem so effortless have honed those skills over many years of practice. However, researchers at Carnegie Mellon and DeepMotion were able to teach AI how to pull off similar feats in a matter of hours (via training). We're not talking about canned animations here that you might see in NBA 2K19, but actual "skills" that have been learned by the AI.
The physics-based system takes lessons from real-life players and uses a reinforcement deep learning model. The team stated off by developing hand and character models complete with joints with torque values. Next, without even using a ball, the models were given a chance to run around an environment and get a feel for interacting with obstacles on a basketball court (and how to avoid them).
Then, hand/arm movements were accounted for along with speed, velocity, and trajectory. Finally, with the actual basketball thrown in to the mix, the push towards dribbling and ball handling mastery could be initiated.
"The training is done incrementally, behavior by behavior, to ensure integration occurs with proper transitions for each skill," writes the DeepMotion team. "The result is a player that can perform multiple ball handling skills in a variety of orders, changing course with seamless blending."
The researchers, after initial training, were even able to pull of moves such as crossovers and dribbling between the AI player's legs. There's no mention of the AI being able to perform monster dunks or taking multiple steps without getting called for traveling like LeBron James, but those skills will likely be learned in time.
"Once the skills are learned, new motions can be simulated much faster than real-time," said Jessica Hodgins, CMU professor of computer science and robotics.
"This research opens the door to simulating sports with skilled virtual avatars," added Libin Liu, chief scientist at DeepMotion. "The technology can be applied beyond sport simulation to create more interactive characters for gaming, animation, motion analysis, and in the future, robotics."