Decentralized Trajectory Tracking Control for Soft Robots Interacting with the Environment

Despite the classic nature of the problem, trajectory

tracking for soft robots, i.e. robots with compliant elements

deliberately introduced in their design, still presents several

challenges. One of these is to design controllers which can

obtain sufficiently high performance while preserving the physical

characteristics intrinsic to soft robots. Indeed, classic control

schemes using high gain feedback actions fundamentally alter the

natural compliance of soft robots effectively stiffening them, thus

de facto defeating their main design purpose. As an alternative

approach, we consider here to use a low-gain feedback, while

exploiting feedforward components. In order to cope with the

complexity and uncertainty of the dynamics, we adopt a decentralized,

iteratively learned feedforward action, combined with

a locally optimal feedback control. The relative authority of the

feedback and feedforward control actions adapts with the degree

of uncertainty of the learned component. The effectiveness of the

method is experimentally verified on several robotic structures

and working conditions, including unexpected interactions with

the environment, where preservation of softness is critical for

safety and robustness.