An AI fitness coach detects muscle movements while you work out and helps avoid beginner injuries.

      During the pandemic, the US Consumer Product Safety Commission noted a 48% increase in injuries related to at-home exercise. While one might attribute this to poor equipment, the true issue was improper form, as individuals lacked a coach to provide corrections.

      Researchers at Drexel University and Michigan State University have developed a prototype to address this very issue in real time through your phone camera, and it shows promise for future development as a legitimate fitness app (via Tech Xplore).

      So, what is the name of this system and how does it function?

      The prototype, named BioCoach, was showcased at the Conference on Computer Vision and Pattern Recognition in June 2026. It utilizes AI and live video from a camera to monitor your workout, evaluate your body mechanics, and provide specific corrections based on biomechanics.

      To achieve this, the system processes video through two parallel streams: the first employs a 3D convolutional neural network to recognize your visual appearance and movement patterns, while the second reconstructs your skeleton in three dimensions, examining your joint angles, range of motion, and the current stage of your exercise.

      Before giving you feedback, BioCoach determines which joints are primarily involved in the exercise you are doing. For example, if you are doing push-ups, it specifically tracks your shoulders, elbows, and wrists to provide tailored corrections.

      And it goes beyond the usual, vague advice like "keep your back straight." The prototype provides detailed anatomical guidance such as "increase elbow flexion to 90 degrees at the bottom."

      How does it compare to its competitors?

      The research team trained BioCoach using Qualcomm’s Exercise Video Dataset, which comprises over 200 re-annotated videos and more than 2,400 new annotations, allowing BioCoach to not only identify what needs correction but also explain why it is important.

      BioCoach has been tested against similar programs from Nvidia, ByteDance, Alibaba, Salesforce, OpenAI, and MIT, among others. It surpassed Stream-VLM, a program developed by MIT and Nvidia, in terms of text quality and assessed accuracy. The prototype also demonstrated greater precision in anatomy-specific feedback.

      Currently, the system remains a prototype, but the team is working on incorporating the capability to estimate joint reaction forces and muscle activation patterns all from video input.

      This research was funded by the National Science Foundation, which leads me to believe that BioCoach could evolve into a groundbreaking smartphone app that provides personalized corrective feedback, promoting proper form and posture, thus preventing injuries and fostering sustainable workout routines for individuals both indoors and outdoors.

      BioCoach is more sophisticated than most AI-based fitness coaches currently available.

      For comparison, both Apple Fitness+ and Mirror offer video workout programs, but their feedback is pre-recorded and lacks the dynamic quality that BioCoach provides.

      Peloton’s hardware features a Movement-Tracking Camera that counts repetitions and identifies issues; however, it necessitates specific equipment like Bike+, Tread+, or Row+, and does not clarify the rationale behind form corrections or their benefits.

      In the same vein, Google’s Health Coach and Samsung Health analyze biometric data such as heart rate and activity rhythm to recommend improvements, but since they cannot observe your movements, they do not provide any feedback on your form.

      In contrast, BioCoach is the first system to integrate 3D skeletal reconstruction with a language model that elucidates the mechanical implications of each correction. If it eventually becomes available as a consumer app, which I sincerely hope, it could provide expert coaching to anyone with a camera.