RoboCat is a new AI agent that can perform a variety of tasks across different robotic arms and incorporates a “self-improvement” training cycle. Developed by researchers at OpenAI, RoboCat can learn a new task with as few as 100 demonstrations, drawing from a vast and diverse dataset to reduce the need for human-supervised training. By generating its own training data, RoboCat can improve its technique and rapidly adapt to different robotic devices. Its ability to independently learn skills and match them with different robots could pave the way for more helpful, general-purpose robotic agents.
https://www.deepmind.com/blog/robocat-a-self-improving-robotic-agent