Foundational models for robotics will be developed

Foundational models in language, such as Chat-GPT, have been all anyone following AI development can talk about these days. These models can handle almost any topic, any context, and give a believable answer. That level of flexibility is greatly needed in robotics. Right now while you can train a robot to “see” or to “move,” it is often very context specific. Much progress is now being made in “few-shot” or “zero-shot learning,” where robots are taught to recognize objects or perform actions that they have not explicitly learned before (i.e. wasn’t in the training data). These methods are rapidly improving and building blocks to finally creating foundational models for robotic vision or motion.