# DeepMind tests Gemini Robotics 2.5 on a Mercedes-Benz production line

> Gemini Robotics 2.5 is being tested on a Mercedes-Benz production line, where it reached a 96% success rate sorting metal parts in a pilot.

- Status: reported (One outlet, not independently verified yet.)
- Topic: AI in Practice
- Published: 2026-11-15T20:29:01+00:00
- Updated: 2026-11-15T21:10:00.773694+00:00
- Web page: https://ai-news.ikangai.com/story/deepmind-tests-gemini-robotics-2-5-on-a-mercedes-benz-production-line/
- Seen in: THE DECODER
- Change: capability (New capability)
- Applies to: Gemini Robotics 2.5, Gemini Robotics ER

## What changed

Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model that helps a humanoid robot sort metal parts on Mercedes-Benz production lines in a pilot.

Google DeepMind has introduced Gemini Robotics 2.5, a vision-language-action model that it says can help a humanoid robot sort metal parts on active Mercedes-Benz production lines with a 96% success rate in a pilot. The robot, Apptronik's Apollo 2, handled kitting. It reached into bulk bins, picked up automotive brackets and placed them into trays in the order each vehicle on the line needed them. The tests ran at Mercedes-Benz plants in Kecskemét, Hungary, and Sindelfingen, Germany.

The model is one of two cooperating systems. According to Google, Gemini Robotics ER keeps track of the workflow, while Gemini Robotics 2.5 controls the hands and arms, reaching into bins and correcting for gaps as they appear. A vision-language-action model takes camera input and language instructions and turns them into robot movements, so the same system that reads a task description also produces the motor commands. Google said the pilot shows promising progress toward industrial use. However, the 96% figure comes from a single proof-of-concept, and the announcement does not describe how many attempts were counted or how the success threshold was defined.

Google said it evaluated the model on more than 200 tasks, including serving food with tongs, tying a bow on a gift box and pipetting liquids. Against its own previous baselines, Google reports a 30% to 100% higher success rate, depending on task type, and tasks completed 20% to 30% faster. Those comparisons are internal. The post does not name the baselines beyond its earlier Gemini Robotics models.

Apptronik CEO Jeff Cardenas said the models improved dramatically within a few weeks of post-training and carried over to Mercedes-Benz production "with very little drop-off." The pilot is only described as a proof of concept. Google's announcement says the model is rolling out to robotics partners Agile, Apptronik and Mercedes-Benz.

On safety, Google describes inference-time reasoning guidance that steers the robot around physical hazards in real time. In Google's evaluations, that raised safe task completion by 33% near obstacles and by 43% near people. Google also released ASIMOV-Jailbreak, an open-source benchmark of hundreds of prompts that disguise unsafe requests across dozens of scenes. Its own model, Gemini 4 Argon, resists jailbreaks 99.5% of the time, against 82% for the previous baseline, Gemini Robotics ER 2. Because Google wrote both the model and the benchmark, these are Google's own measurements.

Google also used Gemini 4 Argon during development as a simulation engine. According to the announcement, Argon can generate thousands of simulated practice runs from a real camera scene. This builds a digital twin, which is combined with real demonstrations so a robot generalizes to new scenes and objects. Argon can also suggest why a task failed from video footage and test fixes both in simulation and on physical robots. The post does not say how much real-world data was replaced by simulated data, so that method's contribution to the results remains unquantified.

## Why it matters

Manufacturers can see robot AI models moving from demos into pilots on real production lines.

## What you can do

Identify repetitive sorting or handling tasks in your plants that could be piloted with a robotics model and measured against current automation.

## Original source

- [Gemini Robotics 2.5: Accelerating Physical Intelligence — Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-5-accelerates-real-world-physical-intelligence/)

## Further reporting

- [Google DeepMind's Gemini Robotics 2.5 scores 96 percent sorting metal parts in a Mercedes-Benz assembly pilot](https://the-decoder.com/google-deepminds-gemini-robotics-2-5-scores-96-percent-sorting-metal-parts-in-a-mercedes-benz-assembly-pilot/)

---

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it.

Cite as: IKANGAI AI News, "DeepMind tests Gemini Robotics 2.5 on a Mercedes-Benz production line", 2026-11-15, https://ai-news.ikangai.com/story/deepmind-tests-gemini-robotics-2-5-on-a-mercedes-benz-production-line/
