DeepMind introduces self-improving AI model called RoboCat

The development could result in many more general-purpose robots being developed at a faster rate.
Loukia Papadopoulos
RoboCat can control a variety of robotic arms.jpg
RoboCat can control a variety of robotic arms.


Robots today can do a variety of tasks as long as they are trained on real-world data. But what if they could bypass this step? It would result in many more general-purpose robots being developed at a faster rate.

Google’s DeepMind has introduced a self-improving AI model called RoboCat that may just be the key to machines that can self-generate new training data to improve their technique without too much human interference. 

This is according to a blog by the firm published on Tuesday.

“RoboCat learns much faster than other state-of-the-art models. It can pick up a new task with as few as 100 demonstrations because it draws from a large and diverse dataset. This capability will help accelerate robotics research, as it reduces the need for human-supervised training, and is an important step towards creating a general-purpose robot,” state the researchers in their post.

They go on to explain that the new model is based on DeepMind’s multimodal model Gato (Spanish for “cat” which explains the name). This model can process language, images, and actions in both simulated and physical environments. 

The researchers made use of Gato’s architecture that comes with a large training dataset of sequences of images and actions of various robot arms solving hundreds of different tasks.

They then trained RobotCat to learn new tasks by following five steps: 

1. Collect 100-1000 demonstrations of a new task or robot, using a robotic arm controlled by a human.

2. Fine-tune RoboCat on this new task/arm, creating a specialized spin-off agent.

3. The spin-off agent practices on this new task/arm an average of 10,000 times, generating more training data.

4. Incorporate the demonstration data and self-generated data into RoboCat’s existing training dataset.

5. Train a new version of RoboCat on the new training dataset.

This diverse training taught the AI model to operate different robotic arms within a few hours. And RobotCat was quick to adapt. Even though it had not been trained on arms with two-pronged grippers, it was able to adapt to a more complex arm with a three-fingered gripper and twice as many controllable inputs.

The more new tasks it learned, the better it got at learning additional new tasks. Early  versions of RoboCat were successful just 36 percent of the time on previously unseen tasks. However, the latest and most advanced RoboCat, which had trained on a greater diversity of tasks, more than doubled this success rate on the same tasks.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board