DeepMind, Google's AI company, has unveiled advances that will help robots make faster, better and safer decisions.
One of them is a training data acquisition system with a “Constitution of the Robot” which uses a visual language model (VLM) and a broad language model (LLM) to understand its environment, adapt to unknown environments, and decide appropriate tasks.
The robot constitution, inspired by Isaac Asimov's “Three Laws of Robotics”, is described as a series of “Safety-oriented instruction” that tells the LLM that Avoid tasks that involve people, animals, sharp objects and even electrical devices.
For added safety, DeepMind programmed the robots to automatically stop when the force applied to their joints exceeded a certain threshold and included a physical switch that human operators could use to deactivate them.
For seven months Google deployed a fleet of 53 AutoRT robots in four different office buildings and conducted more than 77,000 tests. Some robots were controlled remotely by human operators, while others acted according to a script or completely autonomously.
DeepMind's other technological innovations include SARA-RT, a neural network architecture designed to make the RT-2 transformation robot more precise and faster. It also announced RT-Trajectory, which adds 2D contours to help robots better perform certain physical tasks, such as cleaning a table.