System stiffness estimation with the candidate observers algorithm

Environment parameter estimation is of critical importance to enable control adaptation in robotic tasks involving contact with unknown and unstructured environments. Such adaptation is required for achieving dynamic consistency of the system response. We present an algorithm that complies with requirements for online operation involving minimal resources, not relying on velocity or position data. The algorithm is theoretically supported and offers high quality estimations of environment stiffness. It uses only force measurements and the output of two observers of the system plant with different nominal setups. The results are highly satisfactory and significant improvements of the dynamic response of the manipulator can be seen in scenarios of unknown and variable environment stiffness.