Articles

Feature stories, news review, opinion & commentary on Artificial Intelligence

Google DeepMind's Advances in Robotics: AutoRT, SARA-RT, and RT-Trajectory

Robotics AI Robots


What’s new: Google DeepMind introduced three major advances in robotics: AutoRT, SARA-RT, and RT-Trajectory, enhancing the speed, decision-making, and environment understanding of robots.

How it works:

  • AutoRT: Utilizes large foundation models for better training of robots. It directs robots to collect diverse training data in novel environments, helping them understand practical human goals.
  • SARA-RT: Converts Robotics Transformer models into more efficient versions. It employs a method called “up-training,” reducing computational requirements and increasing speed without sacrificing quality.
  • RT-Trajectory: Enhances robots' ability to generalize tasks. It adds visual trajectories to training videos, doubling the task success rate compared to existing models.

Real-world evaluations: Over seven months, AutoRT successfully orchestrated up to 52 unique robots in various settings, collecting a diverse dataset of 77,000 robotic trials.

Safety considerations: AutoRT incorporates robust safety protocols, including a Robot Constitution inspired by Asimov’s laws, to ensure real-world safety. The system also includes practical safety measures from classical robotics.

Implications: These developments represent significant steps toward creating more capable and efficient robots. They build on the foundation of RT-1 and RT-2 models, envisioning a future with integrated systems for advanced robotics.

Why it matters: Google DeepMind's innovations in robotics show promising advances in robot autonomy, efficiency, and safety. These systems could revolutionize how robots interact in real-world environments, bringing us closer to a future with versatile, intelligent helper robots.

We're thinking: These advancements underscore the rapid evolution of robotics, blending AI with traditional robotics to create more adaptable and efficient machines. As these technologies continue to develop, they hold the potential to transform various industries and everyday life.