Google's DeepMind robotics team has unveiled three new developments that it says will help robots make faster, better and safer decisions in real life.
One of these systems includes a training data collection system with “robotic architecture” to ensure that your robotic office assistant can fetch printer paper for you, but without any human jostling at their desk.
Robot constitution
Google's data collection system, AutoRT, uses language models that work together to understand known or unknown environments and select appropriate tasks.
The robot's constitution is inspired by Three laws of robotics By Isaac Asimov, it is described as a set of safety-focused rules that tell the model to avoid selecting tasks that involve humans, animals, sharp objects, and even electrical devices.
Robots with on/off button
For added safety, DeepMind has programmed the robots to automatically stop if the force on their joints exceeds a certain threshold, and given them a physical switch that human operators can use to deactivate them.
Over the course of seven months, Google deployed a fleet of 53 AutoRT robots in four different office buildings and conducted more than 77,000 tests. For testing purposes, these rudimentary robots were controlled remotely by human operators, while others worked either on a script basis or completely autonomously using Google's Robotic Transformer (RT-2) AI learning model.
According to Google's blog, the robot had to understand the environment to perform tasks such as putting a snack on the table.
Google mentions other DeepMind technologies like SARA-RT and RT-Trajectory to help robots better perform specific physical tasks, like clearing a table.
We're still a long way from autonomous robots that can serve drinks or iron clothes, but when they become available, they may have learned from a system like AutoRT.
