Smart hydrogels possess exceptional biocompatibility and stimuli-responsiveness, offering significant potential for applications in biomedicine and flexible electronics. However, the design of hydrogel-based composite systems is complicated by nonlinear mechanical behaviours including large swelling deformation and complex contact conditions. These factors often render traditional simulation-based design methods inefficient due to high computational costs and a restricted exploration of the design space. To facilitate application-oriented design, this study develops structural optimization methods for smart hydrogel actuators and hydrogel-based metamaterials. First, an explicit optimization method is proposed for hydrogel actuators undergoing finite deformation to satisfy specific motion requirements. The structural topology and multi-material distribution are explicitly characterized using moving morphable voids and components. To ensure numerical convergence during intermediate design stages, a material model featuring penalization is integrated with degree-of-freedom removal technology. This framework enables the precise synthesis of actuating behaviours by targeting specific nodal displacements in the post-deformation state. Furthermore, to design hydrogel-based functional metamaterials with unconventional mechanical properties, a machine learning-driven optimization framework is established. A back-propagation neural network is trained to map geometric features to mechanical responses, which, when combined with a multiple-group genetic algorithm, allows for the efficient discovery of optimal metamaterial configurations. Numerical and experimental results demonstrate that the proposed methods successfully achieve the design of bionic hydrogel actuators with prescribed movements, as well as metamaterials exhibiting programmed negative hydration expansion and tissue-like mechanical responses.