Dynamic parameters identification of a humanoid robot using joint torque sensors and/or contact forces

This paper describes the identification of the humanoid robot TORO using two methods: the classical joint torque based identification, and the more recent base-link approach. In our past research we have developed an identification method for dynamics identification of humanoids using only the external forces because most humanoid robots are not equipped with torque sensors, and using the motor current requires a precise model of the transmission. With TORO it is now possible to measure the pure joint torque and thus to compare the efficiency of both methods for humanoid robots. This paper compares the joint torque measured by the torque sensor and the joints torque predicted by the inverse dynamics, using the identified dynamic parameters as well as the contact forces measured with the ankle force torque sensors and the predicted one using the dynamics model and the identified parameters. We discuss how the identified parameters differs using each method both with direct validation and cross validation.