Google has highlighted current studies aimed at empowering robots to understand our requests with great precision, exactly as humans desire.
The massive search company announced three recent innovations that help robots make decisions quickly and securely.
One of the improvements is the AutoRT system, which gathers training information through “robot programming rules” to achieve responsible use of artificial intelligence.
The “Robot Constitution” is known as a set of requirements that focus on safety. These requirements govern the advanced linguistic model to avoid selecting tasks that may involve interacting with humans, animals, sharp tools, or even electrical devices. Google drew inspiration for the “Robot Charter” from Isaac Asimov’s “Three Laws of Robotics.”
Google has developed the AutoRT system to leverage large foundational models to perform a variety of functions.
The AutoRT system can operate the visual language model and the large linguistic model together to analyze the surrounding environment, adapt to new conditions, and make decisions related to suitable tasks.
Google has prepared the robots to autonomously stop if the force on the joints exceeds a specified limit. They have also equipped them with a physical button that allows human operators to disable them.
Google distributed fifty-five AutoRT robots over seven months to four diverse work locations, where they conducted over seventy thousand experiments.
Some robots are controlled by remote human operators, while others either operate according to a pre-programmed scenario or autonomously using the artificial intelligence learning system developed by Google, known as RT-2.
The robots used in the tests demonstrate functionality, as they are equipped with a camera, a mechanical arm, and a movable base.
Each robot in these tests used a visual linguistic pattern to recognize its surroundings and elements within its visual field. Then, the expanded linguistic pattern presented a set of innovative tasks that the robot could perform. The robot, in turn, acted as a leader making decisions in selecting the most suitable task to execute.
One of Google’s new technologies includes SARA-RT, which is a neural network structure designed to enhance the accuracy and speed of the existing RT-2 machine learning model.
Additionally, the company unveiled RT-Trajectory, which provides two-dimensional schematic graphics that enhance the robots’ ability to efficiently perform assigned physical tasks, such as cleaning tables.
