Compliant Robot Behavior Using Servo Actuator Models Identified by Iterative Learning Control

System parameter identification is a necessary prerequisite for model-based control. In this paper, we propose an approach to estimate model parameters of robot servo actuators that does not require special testing equipment. We use Iterative Learning Control to determine the motor commands needed to follow a reference trajectory. To identify parameters, we fit a model for DC motors and friction in geared transmissions to this data using a least-squares method. We adapt the learning method for existing position-controlled servos with proportional controllers via a simple substitution. To achieve compliant position control, we apply the learned actuator models to our humanoid soccer robot NimbRo-OP. The experimental evaluation shows benefits of the proposed approach in terms of accuracy, energy efficiency, and even gait stability.