Robotics · 10 November 2026 · 4 min read
DeepMind reports a 96 percent success rate for Apollo 2 on Mercedes-Benz lines
Google DeepMind said on 10 November that Apptronik’s Apollo 2, using Gemini Robotics 2.5, sorted car parts into trays on production lines in Kecskemét and Sindelfingen at a 96 percent success rate. The post does not say how many robots ran, or for how long. The model is going to robotics partners, not to a public download.

Google DeepMind said on 10 November that Gemini Robotics 2.5 was tested with Apptronik and Mercedes-Benz on active production lines in Kecskemét, Hungary, and Sindelfingen, Germany. The task was automotive kitting: taking parts from bulk bins and placing them in trays in the order a vehicle on the line requires. DeepMind says Apollo 2 did this at a 96 percent success rate. The post does not state how many robots were used, how many attempts were counted, or over how many hours or days the rate was measured.
Jeff Cardenas, Apptronik’s co-founder and chief executive, is quoted saying that Apollo 2 with Gemini Robotics improved over a few weeks of post-training and carried over to Mercedes-Benz production with very little drop-off. DeepMind says Gemini 4 Argon was used as a simulation engine, generating practice runs from a camera view of a real scene and using video of failures to test changes. DeepMind also says safety steering raised safe completion around obstacles by 33 percent and around nearby people by 43 percent. Those percentages are the company’s.
DeepMind says Gemini Robotics 2.5 is rolling out to its robotics partners. The post does not announce a public download or a public API.