BioCoach, an innovative AI system developed by researchers from Drexel University and Michigan State University, is set to revolutionize the way we receive exercise feedback at home. This cutting-edge technology goes beyond simple rep counting or generic encouragement, offering real-time analysis of exercise videos with a focus on form, timing, and biomechanics. The system's unique approach lies in its ability to interpret not just the visual appearance of a movement but also the underlying 3D skeletal movement and body shape, providing detailed and timely feedback that is both interpretable and action-oriented.
One of the key strengths of BioCoach is its exercise-specific attention mechanism. Unlike generic systems that treat every joint equally, BioCoach understands the importance of different joints in various exercises. For instance, a squat requires scrutiny of hips, knees, and ankles, while a push-up focuses on shoulders, elbows, and wrists. This targeted approach allows BioCoach to provide concrete and anatomically specific feedback, such as 'Lumbar spine variance 12° (target: <5°). Maintain neutral spine alignment; engage the core to stabilize the lumbar region.'
The system's training involved a meticulous process of annotating exercise videos with detailed biomechanical targets and rationales. This effort resulted in a more than 2,400 annotations added to over 200 videos, creating a benchmark dataset that enabled the evaluation of BioCoach's performance against other video-language AI systems. The system's effectiveness was evident in its ability to provide more detailed and accurate feedback compared to existing models, particularly when the target feedback became more anatomically specific.
BioCoach's success against top AI researchers and companies, including MIT and NVIDIA, highlights its potential to enhance the quality and interpretability of live exercise feedback without compromising response speed. However, the system is still a prototype, and the team aims to extend its capabilities to estimate joint reaction forces and muscle activation patterns, further refining its ability to detect subtle compensatory movements.
The practical implications of BioCoach are far-reaching. It has the potential to transform at-home fitness and physical therapy, providing users with detailed corrections for common movements like squats, lunges, push-ups, and planks. By extending the expertise of human coaches and trainers between in-person sessions, BioCoach could revolutionize the way we receive exercise feedback, making it more accessible, detailed, and interpretable.
