A. Radulescu, J. Nakanishi, D.J. Braun and S. Vijayakumar, Optimal Control of Variable Stiffness Policies: Dealing with Switching Dynamics and Model Mismatch, J. P. Laumond, N. Mansard, J.B. Lasserre, Geometric and Numerical Foundations of Movements, Springer Tracts in Advanced Robotics, vol. 117, pp. 393-419, 2017.
This chapter develops an optimal control framework for robotic systems with variable stiffness actuators (VSAs) operating under switching dynamics and discontinuous state transitions. The method jointly optimizes control commands, stiffness profiles, switching instances, and movement duration using a hybrid dynamics formulation. To address discrepancies between model predictions and real-world behavior, the framework is extended with an adaptive learning algorithm that continuously updates the system model and re-plans control policies. The approach is validated through simulations and hardware experiments on a brachiating robot with variable stiffness actuation.
Why it matters: Robots with variable stiffness actuators can achieve safer, more efficient movement, but their complexity makes control difficult, especially during contact-rich tasks. This work shows how optimal control combined with adaptive learning can overcome switching dynamics and model mismatch, enabling robust and efficient use of VSAs in realistic environments.