Yoga is widely practiced for improving physical and mental health, yet the biomechanical demands of individual yoga poses remain poorly characterized. Existing experimental studies typically examine only a small number of poses or muscles, limiting their ability to guide pose selection for rehabilitation or exercise prescription. Here we present a computational framework for predicting whole-body joint torques and contact forces from two-dimensional images of yoga poses. The framework combines manual pose digitization, a sagittal-plane MuJoCo human model, and constrained metabolic energy minimization subject to static equilibrium. We applied the framework to 55 sagittally symmetric yoga poses from Light on Yoga, generating a dataset of joint kinematics, contact locations, and predicted joint torques and contact forces. These poses collectively span nearly the full physiological range of motion of the major joints while exhibiting a wide range of predicted joint torques, with the greatest average demands on the lower extremities and lumbar region. Principal component analysis revealed a striking contrast between kinematics and kinetics: nearly all ten dimensions were required to represent the joint-angle data, whereas the first three principal components captured over 90% of the variance in joint torques. Thus, yoga poses exhibit substantial diversity in body configurations but comparatively low-dimensional patterns of predicted joint loading. Predicted joint torque rankings were also largely insensitive to uncertainty in the relative weighting of joint metabolic costs. This framework enables systematic biomechanical analysis of yoga poses and provides a foundation for quantitatively evaluating their mechanical demands and informing future rehabilitation strategies.
🌊
