Talk-to-Robot Demo

Key Insight

Integrating large language models (LLMs) with robotic control enables natural language task planning by decomposing high-level user commands into a sequence of executable primitives. In frameworks like SayCan, the LLM proposes actions ("say") while a learned policy evaluates the feasibility of each action in the current environment ("can"). This combined approach ensures the robot plans actions that are both semantically correct and physically executable in simulation or real-world kitchens.