Force-Based Stiffness Estimation for Robotic Tasks

The performance of direct force control approaches for robotic tasks depends on the accuracy of environment model parameters. A mismatch between real and nominal environment parameters degrades the system response, being a critical problem for precise contact tasks. Adaptive control can tackle this problem based on online contact parameters estimation. In this paper, we develop an algorithm for online stiffness estimation relying on force data, only. No position information is necessary. Our algorithm - the Candidate OBserver Algorithm (COBA) - uses two observers with different candidate stiffnesses to infer the correct system stiffness. Theoretical analysis and experimental results are presented.