Joint Torque and Ground Reaction Force Estimation for a One-Legged Hopping Robot.

Robot force control performs better than position control in terms of dynamic and compliant control when interacting with complex environments. However, adding torque sensors to the robot’s joints or feet for precise torque control can significantly increase volume and weight. Furthermore, torque sensors are prone to be damaged during long-term robot walking, potentially making it impossible to control the robot. This paper proposes a method to estimate joint torque and ground reaction force for a one-legged hopping robot. The method involves analyzing the robot joint dynamics model, compensating for joint friction torque, and estimating joint torque. Nonlinear factors such as joint motor model errors and gearbox backlash can lead to errors in joint torque estimation. To address this issue, nonlinear errors are compensated through BP neural network training, resulting in more accurate estimation of joint torque. Based on the estimated joint torque, the ground reaction force is calculated. The effectiveness of the proposed method was verified by comparing external force measurement data during a one-legged hopping robot jumping to the ground with the estimated ground reaction force.