Did you miss a session from the Future of Work Summit? Head over to our Future of Work Summit on-demand library to stream.
Dribbling a basketball looks deceptively easy. In truth, it takes years of hard work to achieve effortless motion across the court, and, as it turns out, at least as long for developers to model those skills in computer simulations. But a new approach involving artificial intelligence (AI) has the potential to speed things up a bit — at least in the case of developers.
Researchers at Carnegie Mellon University and DeepMotion, a California-based “motion intelligence” startup founded in 2014, have developed a physics-based system that learns dribbling skills from basketball players’ real-life movements.
“This research opens the door to simulating sports with skilled virtual avatars,” Libin Liu, the report’s lead author, told EurekaAlert. “The technology can be applied beyond sport simulation to create more interactive characters for gaming, animation, motion analysis, and in the future, robotics.”
The two teams employed a deep reinforcement learning model — an AI system that mirrors the ways humans respond to environments — in training a basketball-dribbling avatar, setting it loose on a virtual court for trials that numbered in the millions.
Three top investment pros open up about what it takes to get your video game funded.
It learned in two stages. First, it mastered the art of moving around the court without toppling over or running into obstacles. Then it learned how to control its arms and hands and, by extension, the speed, velocity, and direction of the digital basketball.
Physics-based dribbling is notoriously difficult to reproduce digitally, the researchers noted, because human basketball players make contact with the ball only briefly. Exact details — like how a ball spins after making contact with a player’s hand — are particularly tough to capture. And while skilled players can anticipate the timing and positioning of a ball, computer models lack their learned hand-eye coordination.
For those reasons, instead of capturing ball movement, the teams opted to use trajectory optimization to calculate the ball’s most likely paths for a given hand motion. They fed the aforementioned motion capture data — consisting of players rotating the ball around the waist, switching hands, and other dribbling tricks — into the deep learning model, and commenced training.
The result was arm and leg movements “closely coordinated” with movements in the real world. It was a smashing success, all-in-all — the model learned not only how to dribble between its legs, behind its back, and in crossover moves, but how to smoothly transition between those skills without losing the ball.
The teams believe the method can be generalized to other sports.
“Although our framework is designed for basketball skills, we believe that it can be extended to other motions, such as juggling, where the interaction between a simulated character and the manipulated object does not significantly affect the balance of the character,” they wrote. “In future work, we are also interested in investigating other sports, like soccer, where the balance control is tightly coupled to the sports maneuvers.”
It’s not the first time researchers have successfully applied AI to the world of basketball. In July, Mark Cuban-backed startup HomeCourt launched a smartphone app that uses machine learning and computer vision to tally made and missed shots with 99 percent accuracy.
GamesBeat's creed when covering the game industry is "where passion meets business." What does this mean? We want to tell you how the news matters to you -- not just as a decision-maker at a game studio, but also as a fan of games. Whether you read our articles, listen to our podcasts, or watch our videos, GamesBeat will help you learn about the industry and enjoy engaging with it. Learn More