Google Robots Sort Car Parts With 96% Success

Google DeepMind says humanoid robots from Apptronik completed automotive parts sorting at Mercedes-Benz sites in a proof-of-concept demonstration.

Updated on Nov. 11, 2026 in Robotics

Humanoid robot reaching into a metal parts bin in a bright industrial assembly hall beside steel racks of sequencing containers.
Google DeepMind said Apptronik's Apollo 2 humanoid robots sorted automotive parts at Mercedes-Benz plants in Hungary and Germany with a 96% success rate in a proof-of-concept demonstration. AI Illustration. Upload story photo >

Live Poll

Do you think factory robots will make your job better or worse?

Google DeepMind disclosed on November 10 that Apptronik's Apollo 2 humanoid robots achieved a 96% success rate sorting automotive parts at Mercedes-Benz sites in Kecskemét, Hungary, and Sindelfingen, Germany. The demonstration built on a March 2024 agreement between Apptronik and Mercedes-Benz to explore uses for the original Apollo robot.

Why it matters

The demonstration tests whether humanoid robots can handle repetitive, physically demanding factory work in spaces designed for people, potentially without a full factory redesign. Apptronik argues this could reduce the need for costly infrastructure changes on production lines.

The 96% figure comes from a proof-of-concept disclosure that does not define success as an individual pick, completed tray, or other unit. The disclosure reports no trial count, reset frequency, human-intervention rate, cycle time, uptime, or operating cost.

The players

Google DeepMind

Google's artificial intelligence research laboratory, which developed the Gemini Robotics models used to control the robots in the demonstration.

Apptronik

A Texas-based robotics company that builds the Apollo humanoid robot and announced its initial partnership with Mercedes-Benz in March 2024.

Mercedes-Benz

The German luxury automaker that hosted the demonstration at its Kecskemét and Sindelfingen production sites.

The details

Apollo 2 robots performed automotive kitting, picking metal brackets from bulk bins and placing them into sequencing containers in the order each vehicle requires. Two Gemini models split the work: Gemini Robotics ER tracked the workflow, while Gemini Robotics 2.5 controlled the robot's hands and arms.

Timeline

  1. March 2024: Apptronik and Mercedes-Benz agreed to explore uses for the original Apollo robot.

  2. November 10: Google DeepMind disclosed the Apollo 2 proof of concept.

The Tech Race

The demonstration extends Google's Gemini Robotics 2.5 control models from research settings into a live automotive kitting task, placing DeepMind in direct competition with other AI labs building general-purpose robot software. It follows a broader push by Apptronik and similar firms to position humanoids as a flexible alternative to fixed industrial automation.

For workers and consumers, the near-term effect is limited to pilot programs rather than changes on existing lines. The demonstration shows whether humanoids can handle sorting tasks without redesigning factory spaces, which would affect future job roles on production floors.

The takeaway

Watch whether future reports define success per pick and disclose intervention rates, since a 96% figure without those details cannot yet show whether humanoid sorting is ready for sustained production use.

Further reading

Explore more in Robotics coverage of humanoid machines moving from labs onto factory floors.

Live Poll

Do you think factory robots will make your job better or worse?