DeepMind’s RoboCat Learns to Perform Various Robotics Tasks

DeepMind, the artificial intelligence research organization under Alphabet, announced the development of an AI model called RoboCat that can perform multiple tasks across various robotics arm models. RoboCat has pioneered the ability to solve and adapt to tasks and use several real-world robots simultaneously. The results were based on data from a combination of simulations and real-world applications from previous models inside virtual environments along with human-generated robotic control. The model was then refined on specific tasks, creating a specialized spin-off, that practiced on the tasks nearly 10,000 times. The final version of the RoboCat model had trained on 253 tasks and was benchmarked for 141 variations of these tasks both in simulation and in real life. DeepMind claims that RoboCat could herald a new era in solving new tasks in robotics using fewer demonstrations. Going forward, the research team hopes to reduce the number of demonstrations needed to teach RoboCat to complete a new task to fewer than 10.