Google DeepMind's 'Robot Constitution' revealed: What does it entail
Google's DeepMind robotics team has revealed three exciting advancements to enhance speed, precision, and safety of robots in uncontrolled environments. One of them is a data-gathering system called AutoRT, featuring a "Robot Constitution," inspired by American professor Isaac Asimov's "Three Laws of Robotics." The constitution focuses on safety, guiding the large language model (LLM) to steer clear of tasks involving humans, animals, sharp objects, and electrical appliances. This way, robots can work efficiently without causing harm to people or surroundings.
AutoRT system and its deployment
The AutoRT system combines a visual language model (VLM) and LLM to comprehend its environment, adapt to new settings, and choose suitable tasks. Google tested 53 AutoRT robots in four office buildings over seven months, conducting more than 77,000 trials. Some robots were remotely controlled by humans, while others followed a script or operated autonomously using Google's Robotic Transformer (RT-2) AI learning model. Each robot is equipped with a camera, a robot arm, and a mobile base.
Automatic halt and manual kill switch ensure additional safety
To enhance safety measures, DeepMind incorporated programming in the robots that trigger an automatic halt if the force on their joints exceeds a specified threshold. Additionally, a physical kill switch is provided, allowing human operators to deactivate the robots as an additional safety precaution.
What about SARA-RT and RT-Trajectory?
Besides AutoRT, DeepMind also introduced two other technologies: SARA-RT and RT-Trajectory. SARA-RT is a neural network architecture that enhances the accuracy and speed of the existing Robotic Transformer RT-2. On the other hand, RT-Trajectory adds 2D outlines to help robots better perform specific physical tasks, like wiping down a table. These AI breakthroughs bring us closer to creating robots that can autonomously carry out various tasks while ensuring safety and efficiency.
