Meta’s AI division has been busy in recent months finding ways to make concrete production more sustainable and machine translation better. Now one of the company’s ML teams has created a tool that builds realistic musculoskeletal simulations that run up to 4,000 times faster than state-of-the-art prosthetics. According to Meta CEO Mark Zuckerberg, the company can train the models to do things like twirl pens and rotate objects.
Mark Zuckerberg just announced MyoSuite, a new AI platform we developed to build realistic musculoskeletal simulations to help accelerate development of prosthetics. It could also help us build avatars that move more realistically in the metaverse. https://t.co/Q4gVboQGSapic.twitter.com/T27CAbXBg5
— Meta Newsroom (@MetaNewsroom) May 23, 2022
At the moment, Meta sees the platform’s usefulness in two different ways. To start, there’s the obvious metaverse angle. Zuckerberg suggests MyoSuite could help the company develop more realistic avatars for applications like Horizon Worlds. Another more interesting use case could see researchers tapping the platform to develop new prosthetics, as well as novel surgery and rehabilitation techniques. To that end, Meta says it plans to open-source the model.
Meta isn’t the first company to think to use AI to improve prosthetics. Back in 2019, an independent team of researchers created a machine learning system that allowed them to quickly tune a robotic knee to an individual patient. That same year, Intel unveiled a “neuromorphic” deep-learning chip that the company said would make prosthetic limbs more efficient.