Google DeepMind said its Gemini Robotics 2.5 system helped Apptronik's Apollo 2 humanoid robots sort car parts with a 96% success rate in tests on Mercedes-Benz production lines.

Key Takeaways

  • Google DeepMind introduced Gemini Robotics 2.5 as a vision-language-action model focused on dexterity, reliability, and safety
  • DeepMind said Apptronik Apollo 2 robots using the system reached a 96% success rate in a proof-of-concept on Mercedes-Benz production lines
  • DeepMind said it evaluated Gemini Robotics 2.5 across more than 200 tasks and reported higher success rates and faster completion times than previous baselines
  • DeepMind said Gemini 4 Argon helped accelerate development by acting as a simulation engine for robotics research

DeepMind announced Gemini Robotics 2.5 on Tuesday, calling it its most capable model yet for turning what a robot sees and is told into physical action. It said the model is rolling out to its robotics partners.

Google DeepMind said Gemini Robotics 2.5 increases both the breadth of skills a robot can learn and the speed at which it can acquire them, with the goal of helping robots work in less structured real-world environments.

Gemini Robotics 2.5 Moves Beyond Lab Tricks

DeepMind said it tested Gemini Robotics 2.5 across more than 200 tasks, including serving food with tongs, tying a bow on a gift box, and pipetting liquid.

The company said the model showed significant gains in precision, robustness, and speed. Compared with previous baselines, DeepMind said Gemini Robotics 2.5 achieved success rates that were 30% to 100% higher depending on the task type, while completing tasks 20% to 30% faster.

DeepMind also said the model works with different robot types, including Apptronik's Apollo 2 humanoid robot and the Franka Duo platform.

Also Read: Meta Teams Up With Rival Google To Put Its 3 Muse AI Models In The Cloud

From General AI To Physical AI

DeepMind said it partnered with Apptronik and Mercedes-Benz to test Apollo 2 humanoid robots on active automotive production lines in Kecskemét, Hungary, and Sindelfingen, Germany.

In that proof-of-concept, DeepMind said Apollo 2 autonomously performed automotive kitting by sorting individual parts from bulk bins into organized trays in the order needed for each vehicle on the line.

Reliability, Safety And Deployment Questions

DeepMind said Gemini Robotics 2.5 uses inference-time reasoning guidance to steer around physical hazards in real time.

In its evaluations, the company said this improved safe task completion around obstacles by 33% and around nearby people by 43%.

How Gemini Robotics 2.5 Turns Language Into Motion

DeepMind said it used Gemini Robotics ER to keep track of workflow and Gemini Robotics 2.5 as the actions model controlling Apollo 2's hands and arms during the Mercedes-Benz pilot.

The company said Gemini 4 Argon helped accelerate development by acting as a simulation engine, generating varied simulated practice runs from real-world camera scenes and helping researchers analyze task failures from video footage.

Apptronik CEO Jeff Cardenas said the models improved dramatically in just a few weeks of extra training and carried over to Mercedes-Benz production with very little drop-off.

Read Next: Google Launches 2 New AI Chips For Bots, Four In A Single Year

Originally published on Fathom News.