Torque Estimation for Robotic Joint With Harmonic Reducer Based on Deformation Calibration

Joint torque sensing is an important technique for high-performance control of modern robotic systems, especially in an environment that requires man-machine collaboration. However, in many cases, traditional torque sensors are not suitable for robots because of the inevitable increase of joint flexibility and joint size. To address this problem, two novel methods are proposed to estimate torque of robotic joint with harmonic reducer via calibration of its existing flexibility without the need for any additional elastic elements. The first approach utilizes a new harmonic drive compliance model, which is more convenient for calibration and less dependent on the manufacturer's parameters to estimate the output torque. The second method relies on a system based on a back-propagation (BP) neural network to fit the non-linear relationship among the output torque, motor current, and other information that can be obtained from double encoders mounted on motor-side and load-side. The two proposed methods were experimentally investigated and the results show that the estimated torque values were in good agreement with the measurements obtained using a commercial torque sensor. Finally, different suitable application scenarios are presented according to the specific performance of each technique.