Natural language guided goals for robotic manipulation

Goal specification for robotic manipulation is a critical area of research, defining how we interact with robots and how we communicate with them. We propose using universally guided diffusion to generate natural language goal aligned rigid body transforms, from segmented point clouds. We explore various methods for guidance, including their challenges and strengths, and discuss which future directions would be fruitful. This thesis explores endowing robots with the ability to understand and fulfill human goals, potentially enhancing their utility and versatility in various domains.