Google has written a "robot constitution". It promises that AI droids will not kill humans

By: Bohdan Kaminskyi | 08.01.2024, 15:49

Google DeepMind

The robotics team at Google-owned DeepMind has unveiled three new technologies that should help robots make faster, more efficient and safer decisions about tasks. One of them is the AutoRT system, which uses a "Robot Constitution" based on Isaac Asimov's Three Laws of Robotics.

Here's What We Know

The Constitution is a set of rules that prohibit robots from choosing potentially dangerous tasks involving humans, animals, sharp objects, and electrical appliances. This allows, for example, sending a robot assistant to get paper to the printer, but prevents it from accidentally hitting a person on the way.

For added safety, AutoRT also has built-in features to stop the robots automatically if the force on the joints is exceeded and the ability to be manually switched off by the operator. In seven months, the system has completed more than 77,000 tests with 53 robots in Google's offices, some in fully autonomous mode.


AutoRT performs four steps for each task

In addition, DeepMind has introduced the SARA-RT architecture to improve the robots' accuracy and speed, as well as RT-Trajectory technology to improve physical tasks like wiping a desk. The developers hope their advances will help bring robots into offices and other public spaces.

Source: DeepMind