Google’s DeepMind team has made significant strides in robotics by integrating Gemini 1.5 Pro AI into their RT-2 robots. This advancement allows the robots to better understand and navigate their surroundings through natural language interactions.
Source: https://www.instagram.com/p/C9SN9uToGHR
The process involves recording a video tour of a specific area, which the robot then “watches” using Gemini 1.5 Pro to learn about the environment. This enables the robot to respond to various commands based on its observations, such as locating charging points or items within the space.
In tests conducted in a large operating area, the Gemini-powered robots achieved a 90% success rate across over 50 user instructions. The AI also showed potential for more complex task planning beyond simple navigation.
While the technology is impressive, there’s still room for improvement. The robots currently require 10-30 seconds to process instructions, a delay not shown in Google’s demonstration videos. Nevertheless, this development marks a significant step towards more advanced and helpful household robots in the future.
