NVIDIA's new AI model trains robots to move like LeBron and Ronaldo

NVIDIA and Carnegie Mellon researchers have introduced a new AI model called ASAP (Aligning Simulation and Real-World Physics for Learning Agile Humanoid Whole-Body Skill). Designed to teach humanoid robots complex movements through simulations, the model was recently showcased in a video demonstrating its ability to replicate the motions of iconic athletes, including Kobe Bryant, Cristiano Ronaldo, viral K-Pop star Rose, and NBA superstar LeBron James.

In the footage, you can spot the humanoid robot dressed in its favorite athlete's uniforms, performing actions ranging from Kobe Bryant's iconic fadeaway jump-shot, to a famous celebration coined by Lakers forward LeBron James. The way it works: researchers pre-train the robot in simulation, eventually teaching it to replay the motion based on collections of video footage. The advancement is a step towards making humanoid robots move like real people in the future.

"ASAP enables highly agile motions that were previously difficult to achieve, demonstrating the potential of delta action learning in bridging simulation and real-world dynamics," the researchers wrote.

2

Jim Fan, lead of NVIDIA's GR00T project, took to X to highlight the new model as a "real2sim2real" system, capable of mastering "extremely smooth and dynamic motions" for full-body control. Achieving this level of fluidity wasn't easy - researchers reported that tests struggled with translating simulated movements into real-world movements due to differences in physics. During testing, two robots overheated and sustained damage from the intense energy demands of the movements due to the stress it placed on their motors and hardware.

Fortunately, the researchers have triumphed. Although the movements still display some level of jank, the available footage gives us a glimpse into what robot basketball, or robot soccer, could look like in the future.

To find out more, check out ASAP's homepage.