DeepMind’s AI humanoids are already doing factory work at Mercedes

Google DeepMind says robot maker Apptronik's Apollo 2 humanoids achieved a 96% success rate sorting automotive parts on active Mercedes-Benz production lines, using AI to handle preparation work that feeds vehicle assembly.

The company disclosed the proof of concept on Nov. 10, identifying sites in Kecskemét, Hungary, and Sindelfingen, Germany. The work provides a specific test for its Gemini Robotics 2.5 model: arranging the components needed by the people building cars.

The task was automotive kitting, which involves taking parts from bulk bins and organizing them in the order needed for each vehicle. DeepMind's example involved picking up metal brackets and placing them into sequencing containers. That ties the demonstration to a defined step in preparing an assembly line's supplies. The trays are meant to match the parts needed for each vehicle, so getting the sequence right is part of the task. The robot has to organize the components as well as move them.

A defined preparation task is working in a pilot; the sustained operating burden remains unquantified.

Company-reported proof of concept. Evaluation duration and uptime were not disclosed. The diagram is schematic and does not depict Apollo 2's actual hardware.

Sources and evidence

Sources

1. Google DeepMind pilot disclosure

Supporting facts

• Reported statement Google DeepMind said on November 10, 2026 that Apptronik Apollo 2 humanoids achieved a 96% success rate in an automotive-kitting proof of concept on active Mercedes-Benz production lines. [1]
• Reported statement DeepMind described automotive kitting as moving individual parts from bulk bins into trays in the sequence needed for each vehicle; its concrete robot example involved metal automotive brackets and sequencing containers. [1]
• Reported statement DeepMind described the kitting task as autonomous but did not disclose the success definition, trial size, evaluation duration, reset or human-intervention rates, cycle times, operating costs or uptime for the Mercedes-Benz proof of concept. [1]

Method

Qualitative process schematic based on DeepMind's description, disclosed November 10, 2026. Bracket shapes and their count are illustrative, not trial observations. The 96% is the reported figure, without inferred denominator or derived arithmetic.

Scope

Company-reported proof of concept. Evaluation duration and uptime were not disclosed. The diagram is schematic and does not depict Apollo 2's actual hardware.

Infographic by CryptoSlate

Two models divided the work. Gemini Robotics ER tracked the workflow, while Gemini Robotics 2.5 controlled the robot's hands and arms, according to DeepMind.

Apptronik CEO Jeff Cardenas said in the announcement that the models improved after a few weeks of post-training and transferred to Mercedes production with little performance drop-off.

The relationship predates this result. In March 2024, Apptronik announced an agreement to explore uses for its original Apollo, including delivering assembly kits, inspecting components and transporting containers of kitted parts. It argued that humanoids could handle repetitive, physically demanding work in spaces designed for people, potentially avoiding a full factory redesign.

What 96% leaves unanswered

DeepMind calls the kitting autonomous, but does not define whether success means an individual pick, a completed tray or another unit. Its disclosure provides no trial count and does not report reset frequency, human-intervention rates, cycle times, uptime or operating costs.

For factory operators, the unanswered questions are how much work the robots finish in a shift and how often a person must step in to keep production moving.