Apollo 2 Reaches 96% Success in Mercedes-Benz Pilot Using Two Gemini Models
Testing covered more than 200 tasks, with success gains of 30% to 100% over earlier baselines and completion times 20% to 30% shorter.

Apptronik’s Apollo 2 humanoid robot achieved a 96% success rate in a Mercedes-Benz parts-kitting pilot that paired Google DeepMind’s Gemini Robotics 2.5 with Gemini Robotics ER.
The pilot ran on Mercedes-Benz production lines in Germany and Hungary. Gemini Robotics ER tracked the workflow, while Gemini Robotics 2.5 controlled Apollo 2’s hands and arms.
The model was evaluated on more than 200 tasks. Gains in task success ranged from 30% to 100% over earlier baselines, depending on the task, while completion times fell by 20% to 30%.
The pilot adds a factory use case to publicly disclosed humanoid work, which also includes demonstrations and data collection. Humanoid deployments remain limited across the field.
Gemini Robotics 2.5 is a vision-language-action model. Google DeepMind is gradually making it available to robotics partners.