Home Business Robotics Google robots complete complex...

Google robots complete complex tasks using new Gemini AI models

Robotics

BusinessHonor - Google Gemini AI Tackles Complex Tasks

Business Honor
01 October, 2025

Google’s robots use Gemini AI to perform multi-step tasks with safety and accuracy.

Google Deep Mind has developed new AI technology that helps robots perform complex tasks. The updated robots use the Gemini Robotics 1.5 system, which allows them to see, understand, and complete multi-step actions without human help.

In one test, a robot passed the “banana test.” It sorted three types of fruit into different plates based on color. Older versions could only follow one instruction at a time, such as placing a banana in a bowl. With the new system, the robot understood all steps and finished the task correctly. Another robot, Apollo, was tested by sorting laundry into black and white baskets. After the baskets were moved during the task, the robot noticed the change and continued sorting correctly. This showed that the robot could adjust to new situations without being reprogrammed.

The robots learn by interacting with their surroundings by a method called embodiment learning. This helps them to understand what to do with different situations. For example, a robot could be asked to sort waste into trash, recycling, and compost bins. It would look up local rules online, check the items, and then place them in the correct bin. The Gemini Robotics 1.5 system includes two parts. One part helps the robot turn images and instructions into actions. The other part helps it plan and think through each step needed to complete a task.

Google also focused on safety. The robots are trained to avoid dangerous actions, follow rules, and think before they act. The robots passed with the high safety criteria that were set according to a new safety test called ASIMOV. These new abilities prove advancements in robotics and allow machines to operate more independently and securely in everyday situations.