Google creates new 'robot specifications' to avoid sci-fi movie plot becoming reality

Google formulates new "robot code" to ensure that science fiction movies will not come true

The DeepMind robotics team recently announced three new developments aimed at helping robots make faster, better, and safer decisions outside the laboratory. For one, they developed a system that collected training data through "robot codes" to ensure that a robotic office assistant could fetch more printer paper for users but avoid attacking human colleagues who happened to be in the way. This development will further improve the efficiency and safety of robots.

AutoRT, Google's data collection system, utilizes visual language models (VLM) and large language models (LLM) working together to understand the environment, adapt to unfamiliar situations, and decide on appropriate tasks. The design of this system is inspired by Asimov's "Three Laws of Robotics", which is described as a set of safety-centered guiding principles to ensure that AI avoids making choices involving humans, animals, sharp objects, and even electrical appliances when operating robots. task.

In order to improve safety, DeepMind has programmed the robot so that once the robot's joints receive excessive force, it will automatically stop running. In addition, the robot is equipped with a physical kill switch that allows a human operator to shut down the robot immediately. Google deployed a fleet of 53 AutoRT robots in four different office buildings over seven months and conducted more than 77,000 trials. Some of the robots are remotely controlled by human operators, while others operate completely autonomously based on scripts or Google’s Robot Transformer (RT-2) AI learning model. Such safety measures and experimental design ensure the safety and controllability of the robot.

The robot used in the experiment was mainly aimed at practicality and was equipped with a camera, a robotic arm and a mobile base. Each robot uses VLM (AI Visual Language Module) to perceive the environment and recognize objects. Next, LLM (AI Large Language Module) will provide suggestions for a series of creative tasks, such as "put snacks on the countertop", and act as a decision-maker to select suitable tasks for the robot to perform. This system design can enhance the intelligence level of the robot and enable it to better adapt to different task requirements.

The above is the detailed content of Google creates new 'robot specifications' to avoid sci-fi movie plot becoming reality. For more information, please follow other related articles on the PHP Chinese website!