Design of robotic physical intelligence

Bridging the performance gap between robotics and biological systems demands significant advancements across multiple domains, including mechanical design, control theory, modeling, and artificial intelligence. In this thesis, I will introduce innovative methodologies for multi-domain robotic design and novel modeling and control techniques to create robots that leverage their physical structure to become adaptive, efficient, and capable of complex tasks. The research is demonstrated through three main case studies: the design and control of soft manipulators, the development of quadrupedal robots with natural emergent locomotion patterns, and the exploration of Large Language Model (LLM)-driven co-design approaches. By advancing these robotic fields, the work seeks to contribute to developing robots able to leverage their physical properties toward intelligent behaviors.