BioCoach prototype – A new AI fitness prototype called BioCoach watches your exercise through a phone camera, reconstructs your skeleton in 3D, and delivers biomechanics-based corrections in real time. The approach targets the spike in at-home injuries during the pandemic—when peo
During the pandemic, the US Consumer Product Safety Commission recorded a 48% spike in at-home exercise injuries. People often blamed equipment. The numbers painted a different story: bad form, and the absence of a coach to correct it.
Now, researchers at Drexel University and Michigan State University say they’ve built a real-time system that tries to fill that gap—using only a phone camera. The prototype, called BioCoach, was presented at the Conference on Computer Vision and Pattern Recognition in June 2026.
BioCoach uses AI and live video to watch you exercise. analyze your body mechanics. and deliver specific corrections based on biomechanics. It processes the feed through two parallel streams. One uses a 3D convolutional neural network to learn your visual appearance and body movement patterns. The other reconstructs your skeleton in three dimensions. then examines joint angles. range of motion. and the phase of the movement.
Before it gives feedback, BioCoach identifies which joints matter most for the specific exercise you’re doing. If you perform push ups, it focuses on shoulders, elbows, and wrists—so the corrections are tied directly to the part of the body doing the work.
The guidance is also far more detailed than the usual app instruction to “keep your back straight.” The prototype can deliver anatomically precise feedback like “increase elbow flexion to 90 degrees at the bottom.”
In building BioCoach’s ability to explain not just what to change but why. the researchers trained it on Qualcomm’s Exercise Video Dataset. That training drew on over 200 re-annotated videos and over 2,400 new notes. The goal was to teach the system to provide corrections with context grounded in mechanics.
The team also tested BioCoach against similar programs from Nvidia, ByteDance, Alibaba, Salesforce, OpenAI, and MIT. In those comparisons, BioCoach outperformed Stream-VLM—an MIT and Nvidia program—on text quality and judged correctness. It also showed improvements in anatomy-specific feedback accuracy.
For now, BioCoach remains a prototype. The researchers say they’re working on adding the ability to estimate joint reaction forces and muscle activation patterns from a video feed.
The work received support from the National Science Foundation, and the researchers’ ambition is clear: if this approach can be translated into a consumer smartphone app, it could make expert-level corrective coaching accessible to people exercising indoors and outdoors.
Compared with today’s fitness tech, BioCoach’s selling point is its combination of vision and explanation. Apple Fitness+ and Mirror provide video-based workout programs, but their feedback is pre-recorded rather than dynamic. Peloton’s Movement-Tracking Camera can count reps and flag issues. but it requires dedicated hardware like Bike+. Tread+. or Row+—and it doesn’t explain the reasoning behind form corrections. Google’s Health Coach and Samsung Health use biometric signals such as heart rate and activity cadence. yet they can’t see you moving. meaning they don’t provide guidance for your form.
BioCoach. by contrast. is designed to combine 3D skeletal reconstruction with a language model that explains the mechanical consequence of each correction. If it reaches a phone as a real app. the promise is straightforward: fewer injuries driven by bad form. and coaching that helps people stick with safer. more sustainable workouts.
BioCoach Drexel University Michigan State University AI fitness coach smartphone camera computer vision 3D skeletal reconstruction biomechanics corrections CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION 2026 Consumer Product Safety Commission at-home exercise injuries Qualcomm Exercise Video Dataset Stream-VLM Nvidia ByteDance Alibaba OpenAI MIT