• iterative learning control
• soft robotics
• variable stiffness actuators

Despite the classic nature of the problem, trajectory tracking for Soft Robots, i.e. robots with elastic elements deliberately introduced in their design, still presents several challenges. One of these is related to the complexity of Soft Robots dynamics, which presents hard-to-model nonlinearities, especially if compared with rigid robotic systems. Another challenge is to preserve the mechanical behavior of the Soft Robot. Indeed, in this thesis, I show how a feedback action proportional to the tracking error changes the natural compliance of the Soft Robot, de facto fighting against their main design purpose. Hence the key idea is to exploit a mostly feedforward component. In order to cope with the dynamics complexity, a decentralized, iteratively learned feedforward action has been adopted, combined with a locally optimal feedback control. Stability and convergence properties of the proposed algorithm are analyzed together with the stiffness preservation. The effectiveness of the method is experimentally verified on several robotic structures and working conditions, e.g. unexpected interactions where Soft Robots design is critical for safety and robustness.