In a leap forward for biomechanical research, Meta AI, in conjunction with McGill University, Northeastern University, and the University of Twente, presented the MyoSuite 2.0 platform. This pioneering release exhibited simulated skeletal body parts, driven by artificial intelligence, emulating nuanced human movements.
The AI-propelled skeletal arm, endowed with 39 muscles interacting through 29 joints, displayed advanced dexterity, handling a diverse range of objects in a manner reminiscent of a curious toddler. Similarly, simulated legs, with 80 muscles functioning across 16 joints, mirrored the developmental motions toddlers undergo in their journey to walking.
Vikash Kumar, a principal investigator of the project, elaborates on the intricate nature of human movements. Highlighting the complexity, he pointed out how each joint in the human system is facilitated by several muscles, and conversely, each muscle navigates through multiple joints. This stands in stark contrast to conventional robots with their one-motor-per-joint design. The continuous and dynamic activation patterns required to mobilize an arm or leg far surpass the simplicity of initiating a robot. Kumar believes that while recreating these nuanced motor strategies in MyoSuite poses substantial challenges, roboticists stand to gain immensely from understanding the human body’s control mechanisms. He posits that nature’s choice of this intricate system underscores its efficacy. “It suggests that if a simpler mechanism were feasible, evolution wouldn’t have opted for this elaborate design,” Kumar observes. Having recently transitioned from a dual role as a Meta research scientist and an adjunct professor at Carnegie Mellon University, Kumar now dedicates his expertise full-time to CMU’s Robotics Institute.
Concluding, MyoSuite 2.0’s aspiration revolves around harnessing machine learning to solve biomechanical control dilemmas, targeting the emulation of human-level agility and dexterity. The platform further provides a repository of benchmark musculoskeletal models and open-source tasks, propelling research and innovation in the domain.
