Implementing Torque Control-Based Biped Walking of Humanoid Robots with High Reduction Gear and No Joint Torque Feedback

Most adult-sized bipedal and humanoid robots primarily use high-reduction gears to achieve high torque output of their joints. However, these high-reduction gears have problems such as high joint friction and low backdrivability, which can make accurate torque control difficult and potentially adversely impact system stability. Hence, most robots attempt to overcome these problems related to reduction gears by using torque sensors. In this study, we demonstrated the feasibility of torque-controlled walking without joint torque sensors with the bipedal robot RoK-3, which is equipped with high-reduction gears known as Harmonic Drive. The torque command generation adopts an integrated approach [1], [2] of model predictive control (MPC) and whole-body control (WBC), and a friction compensator is introduced to compensate for high joint friction. The experiments include balancing on a moving slope, up and down motions during a single foot support, and walking in place scenarios. Finally, we have demonstrated the implementation of torque control for robots using high-reduction gears and the feasibility of walking based on torque control, even without the use of torque sensors.