DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"
Summary
Google DeepMind introduced Gemini Robotics 2.5, an AI system designed to help robots understand instructions and perform physical tasks. The company reported improved lab results and a 96% success rate in factory trials, and said it is preparing an open-source safety benchmark for robots working alongside people.Key Facts
- Gemini Robotics combines a model that interprets instructions and video with a model that generates robot actions.
- DeepMind tested the system on more than 200 activities, including serving food with tongs and using a pipette.
- DeepMind said Gemini Robotics 2.5 had a 30% to 100% higher task success rate and ran 20% to 30% faster than the prior version in laboratory tests.
- In trials with Mercedes-Benz and Apptronik, Gemini-powered Apollo 2 robots kitted automotive parts in European factories with a reported 96% success rate.
- DeepMind said reaching near-perfect reliability on specific factory tasks remains difficult for a general-purpose robot.
- The team used Gemini 4 Argon to generate simulated training examples based on real-world camera data and to analyze failures.
- DeepMind said it combined simulated examples with real-world training demonstrations to train Gemini Robotics 2.5.
- The article says DeepMind is preparing an open-source safety benchmark called ASIMOV-jailbreak, but provides no further details in the available text.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
