Camera-based Online Torque Estimation of a Multi-link Pendulum via Extended Kalman Filtering

This paper presents a vision-based method for estimating the joint torques and motion of a 10 link passive pendulum. Torque estimation in these systems is challenging owing to the lack of direct torque sensors and the complex dynamics of multi-degree-of-freedom systems. To address this, a high-speed camera was employed to capture the joint angles, and an Extended Kalman Filter was used to estimate the angular velocities and torques online. Unlike traditional methods, this approach non-intrusively extracts dynamic information from high-degree-of-freedom systems. The proposed approach was validated through simulations and physical experiments, achieving a maximum normalized root mean square error of 0.84% across all joints. These results show that accurate torque estimation is possible with minimal sensor data, which is beneficial when sensor placement is restricted.