Abstract
Soft robotics has become a rapidly expanding field that redefines how intelligence can emerge from the interplay of body, material, and control. Unlike conventional rigid robots, soft robots leverage compliance, deformation, and morphological computation to enable safe, adaptive interaction with complex environments. This unique embodiment underpins the concept of embodied intelligence (EI), in which sensing, actuation, and reasoning are physically coupled within the robot’s morphology rather than isolated in separate modules. Achieving such embodied intelligence requires the deep integration of four interrelated technologies: modeling, simulation, perception, and control. Modeling provides the physical understanding that enables robots to interpret and predict their own body dynamics. At the same time, simulation extends this capacity into virtual space, allowing embodied behaviors to be designed, tested, and optimized efficiently. Perception transforms the soft body into a distributed sensory organ, enabling robots to sense through deformation and to interact with their surroundings in an inherently adaptive manner. Control, finally, closes the loop of embodiment by exploiting mechanical compliance and learning-based strategies to generate coordinated and context-aware behaviors that arise naturally from morphology-environment coupling. This review offers a comprehensive synthesis of the four pillars of embodied intelligence in soft robotics, emphasizing how they coevolve to realize physically grounded intelligence, and discusses the remaining challenges and future directions toward scalable, robust, and autonomous soft robotic systems.