Toward generic control for soft robotic systems

Abstract Conventional robotics equates motor intelligence with the generation and execution of precise actions derived from system models. In high-dimensional and strongly uncertain systems such as soft robots, this precision-centric paradigm breaks down because it tightly binds intelligence to system-specific dynamics and control structures, preventing generalization across morphologies and tasks. Here we propose an alternative view in which intelligent behavior arises from regulating the boundaries of behavioral feasibility, allowing stable and adaptive behavior to emerge through environment interaction. Guided by this perspective, we introduce a general control framework that specifies high-level intentions to constrain unsafe or infeasible behaviors rather than prescribing detailed action commands, thereby shifting control from action generation to behavior regulation. Experiments across diverse soft robotic systems demonstrate robust and transferable motor behavior without reliance on specific models, pointing to a new paradigm for motor intelligence in complex systems.