Google DeepMind has unveiled an advanced system for teaching robots new tasks

Google's DeepMind team has unveiled a new artificial intelligence algorithm, RT-2, that allows robots to efficiently transfer concepts learnt on relatively small datasets to different scenarios.

Here's What We Know

According to the developers, RT-2 demonstrates improved generalisation abilities, as well as semantic and visual understanding beyond what the robot has learned from the data. This includes interpreting new commands and responding to user commands by performing elementary reasoning.

The system effectively demonstrates the ability to determine the best approach for a particular novel task based on available contextual information.

As an example, the developers gave a scenario in which the robot is asked to throw away rubbish. In many models, the user must teach the machine to determine what is considered waste, and then train it to pick up the item and throw it in a bin. This level of minutiae is not very appropriate for systems that must perform many different tasks.

Instead, RT-2 uses the Internet to gain knowledge. The large array of web data allows the algorithm to learn on the fly how to perform tasks it didn't know how to do before.

According to the team, moving from RT-1 to RT-2 improved the efficiency of new tasks from 32% to 62%.

Source: DeepMind