Despite the widespread development of highly intelligent robotic systems exhibiting great precision, reliability, and dexterity, robots remain incapable of performing basic manipulation tasks that humans take for granted. Manipulation in unstructured environments continues to be acknowledged as a significant challenge. Soft robotics, the use of less rigid materials in robots, has been proposed as one means of addressing these limitations. The technique enables more compliant interactions with the environment, allowing for increasingly adaptive behaviours better suited to more human-centric applications.

Embodied intelligence is a biologically inspired concept in which intelligence is a function of the entire system, not only the controller or `brain'. This thesis focuses on the use of embodied intelligence for the development of soft robots, with a particular focus on how it can aid both perception and adaptability. Two main hypotheses are raised: first, that the mechanical design and fabrication of soft-rigid hybrid robots can enable increasingly environmentally adaptive behaviours, and second, that sensing materials and morphology can provide intelligence that assists perception through embodiment. A number of approaches and frameworks for the design and development of embodied systems are presented that address these hypotheses.

It is shown how embodiment in soft sensor morphology can be used to perform localised processing and thereby distribute the intelligence over the body of a system. Specifically in soft robots, sensor morphology utilises the directional deformations created by interactions with the environment to aid in perception. Building on and formalising these ideas, a number of morphology-based frameworks are proposed for detecting different stimuli.

The multifaceted role of materials in soft robots is demonstrated through the development of materials capable of both sensing and changes in material property. Such materials provide additional functionality beyond their integral scaffolding and static mechanical characteristics. In particular, an integrated material has been created exhibiting both sensing capabilities and also variable stiffness and `tack’ force, thereby enabling complex single-point grasping.

To maximise the intelligence that can be gained through embodiment, a design approach to soft robots, `soft-rigid hybrid' design is introduced. This approach exploits passive behaviours and body dynamics to provide environmentally adaptive behaviours and sensing. It is leveraged by multi-material 3D printing techniques and novel approaches and frameworks for designing mechanical structures.

The findings in this thesis demonstrate that an embodied approach to soft robotics provides capabilities and behaviours that are not currently otherwise achievable. Utilising the concept of `embodiment' results in softer robots with an embodied intelligence that aids perception and adaptive behaviours, and has the potential to bring the physical abilities of robots one step closer to those of animals and humans.