Robotic skins incorporate sensors and actuators into stretchable and flexible planar substrates. Wrapping a robotic skin around a passive, deformable structure imparts controllable motion onto that structure, rendering it an active robotic system. Robotic skins can be applied to the surface of a structure, then removed and re-applied to the surface of another structure. This reconfigurability enables use of the same robotic skin to achieve multiple motions, which depend on the interaction between the skin and its host structure. Considering the broad range of use cases for robotic skins in resource-limited environments, it may not be possible to pre-characterize this skin–structure interaction for all potential systems. Therefore, it is advantageous to have systems that can learn their models in situ, which saves considerable time in realizing a functional system. Previously, we have shown that robotic skins can be used to estimate state and stiffness of the underlying passive structure they are attached to. In this letter, we demonstrate how this ability to measure state and stiffness can be used to learn model parameters in situ for feedforward control, and show how feedback control can be implemented simultaneously with the same system. We further show how this learning is expandable to multi-segment systems and will compensate for gravitational effects by adjusting model parameters.