Cognitive developmental robotics aims to develop robots capable of human-like
learning, interaction, and behavior by grounding concrete and abstract concepts
in sensorimotor experiences and social interactions. This talk introduced
examples on language grounding in cognitive developmental robotics, and
explores how principles like "starting small", "embodied intelligence" and
"super-embodiment" can address the limitations of AI tools, such as large
language models (LLMs), which rely heavily on large datasets and lack
sensorimotor grounding. By integrating incremental, multimodal learning and
redefining embodiment to encompass physical, mental, and social processes, we
can enable robots to better understand and utilize abstract concepts. The talk will
also reflect on the pros and cons of using foundation models in cognitive robotics.