DeepMind’s Apollo 2 Humanoids Reach 96% Success in Mercedes Parts‑Sorting Pilot
DeepMind reported that its Apollo 2 humanoid robots achieved a 96% success rate sorting automotive parts on Mercedes‑Benz lines in Germany and Hungary.
DeepMind reported that its Apollo 2 humanoid robots achieved a 96% success rate sorting automotive parts on Mercedes‑Benz lines in Germany and Hungary.
DeepMind announced on November 10 that its Apollo 2 humanoid robots, built by Apptronik, completed a pilot on Mercedes‑Benz production lines in Germany and Hungary with a reported 96% success rate in sorting automotive parts. The test focused on the kitting process—organising components from bulk bins into trays that match the exact sequence needed for each vehicle—marking one of the first real‑world deployments of AI‑driven humanoids for repetitive factory work.
According to CryptoSlate, the pilot was run at the Sindelfingen plant in Germany and the Kecskemét facility in Hungary. The robots used DeepMind’s Gemini Robotics 2.5 model to control the arms and hands, while a separate Gemini Robotics ER system tracked workflow. The task involved picking metal brackets and placing them into sequencing containers, a step that directly feeds the vehicle assembly line.
DeepMind highlighted the 96% success figure but did not define the metric—whether it refers to individual picks, completed trays, or another unit. The company also omitted data on human‑intervention rates, cycle times, shift‑level throughput, or operating costs, leaving manufacturers without a full picture of the robots’ practical utility.
Apptronik’s CEO Jeff Cardenas said the robots showed improvement after a few weeks of post‑training and transferred to the Mercedes lines with little performance drop‑off. The partnership between DeepMind and Apptronik dates back to March 2024, when the firms first explored using the original Apollo platform for tasks such as component inspection and container transport. The current pilot suggests humanoids could handle physically demanding, repetitive work in existing factory layouts, potentially avoiding costly plant redesigns.
Observers will be looking for additional data from DeepMind on shift‑level productivity, human‑intervention frequency, and cost analysis. Further pilots with other OEMs or expanded tasks could clarify whether humanoid AI can scale beyond proof‑of‑concept to become a standard element of automotive manufacturing.
They sorted automotive components from bulk bins into vehicle‑specific kitting trays, a step that prepares parts for assembly.
The pilot took place at the Sindelfingen plant in Germany and the Kecskemét plant in Hungary.
DeepMind has not announced a timeline; further information is expected as the pilot progresses or additional trials are conducted.
Discussion 0