Optimization-Based Analysis of a Cartwheel

The estimation of acting joint torques and ground reaction forces is of particular interest for the analysis and training of athletic human motions. Modern IMU-based motion capture systems can record the kinematics of motions without the constraint of a fixed capture volume but with a lower accuracy when compared to marker-based systems, however they can give no information about the underlying dynamics. We propose a general approach that allows us to analyze such motion recordings for which external ground reaction force measurements are not possible. To achieve this we use dynamically-consistent optimization to generate a physically valid least-squares fit of a dynamic model of the subject to the recorded motion. We demonstrate the method by fitting and analyzing a cartwheel motion. The resulting joint torques allow us to understand how the different joints are actuated throughout the motion and reconstructed contact forces give insight about the interaction with the environment. Calculation of the forces and torques are of great value during training of athletic motions both to improve technique and to prevent injury. Additionally we can estimate the deviation between the measurement and the physically valid fit.