Google’s DeepMind has developed a self-improving robotic agent, RoboCat, that can learn new tasks without human oversight. This technological advancement represents substantial progress towards creating versatile robots for everyday tasks.

Introducing RoboCat: DeepMind’s newly developed robot, named RoboCat, is a groundbreaking step in artificial intelligence (AI) and robotics. This robot is capable of teaching itself new tasks without human supervision.

  • RoboCat is termed as a “self-improving robotic agent.”
  • It can learn and solve various problems using different real-world robots like robotic arms.

How RoboCat Works: RoboCat learns by using data from its actions, which subsequently improves its techniques. This advancement can then be transferred to other robotic systems.

  • DeepMind claims RoboCat is the first of its kind in the world.
  • The London-based company, acquired by Google in 2014, says this innovation marks significant progress towards building versatile robots.

Learning Process of RoboCat: RoboCat learns much faster than other state-of-the-art models, picking up new tasks with as few as 100 demonstrations because it uses a large and diverse dataset.

  • It can help accelerate robotics research, reducing the need for human-supervised training.
  • The capability to learn so quickly is a crucial step towards creating a general-purpose robot.

Inspiration and Training: RoboCat’s design was inspired by another of DeepMind’s AI models, Gato. It was trained using demonstrations of a human-controlled robot arm performing various tasks.

  • Researchers showed RoboCat how to complete tasks, such as fitting shapes through holes and picking up pieces of fruit.
  • After these demonstrations, RoboCat trained itself, improving its performance after an average of 10,000 unsupervised repetitions.

Capability and Potential of RoboCat: During DeepMind’s experiments, RoboCat taught itself to perform 253 tasks across four different types of robots. It could adapt its self-improvement training to transition from a two-fingered to a three-fingered robot arm.

  • RoboCat is part of a virtuous training cycle, getting better at learning additional new tasks the more it learns.
  • Future development could see the AI learn previously unseen tasks.
  • This self-teaching robotic system is part of a growing trend that could lead to domestic robots.

Source (The Independant)

PS: I run a ML-powered news aggregator that summarizes with an AI the best tech news from 50+ media (TheVerge, TechCrunch…). If you liked this analysis, you’ll love the content you’ll receive from this tool!