Gemini Robotics 2.5 handles manipulation of deformable objects, robots working at Mercedes-Benz
Google DeepMind has expanded the Gemini Robotics 2.5 model with improved manipulation of deformable objects, deployed robots at the Mercedes-Benz manufacturing plant, and announced the open source safety benchmark ASIMOV-jailbreak. The model was evaluated on more than 200 tasks.
Google DeepMind announced that the Gemini Robotics 2.5 model has improved dexterity when manipulating deformable objects such as fabric or ribbon. According to the company, the model was evaluated on more than 200 tasks, including serving food with tongs or working with a pipette. According to the company, the Apollo 2 humanoid robot from Apptronik managed to tie a bow using this model, a task that has long been difficult for robotic manipulators. According to the company, robots with this version of the model are already working at the Mercedes-Benz manufacturing plant. The company also announced an open source safety benchmark called ASIMOV-jailbreak, intended for testing robotic systems.
The Gemini Robotics 2.5 model was previously released with integration of the Gemini 4 Argon model as a simulation engine. According to the company, Argon generates thousands of simulated training runs from real video recordings of scenes and creates a digital twin of the environment, which, combined with real training data, speeds up the robot's ability to generalize to new scenes and objects. According to the company, Argon also helps diagnose the causes of task failures from video recordings and test improvements both in simulation and on real robots.
Sources agree that this is a step on the path toward what the company refers to as "physical AGI" – increasing the dexterity, real-world reliability, and safety of robotic systems. Details of the ASIMOV-jailbreak benchmark itself are not elaborated on in the sources.
Why it matters
Manipulating deformable materials (textiles, cables, packaging) has long been a weakness of robotic systems, so improvement in this area is relevant for real-world deployment in logistics and manufacturing. Deployment at the Mercedes-Benz manufacturing plant is concrete evidence that the technology is moving from demo environments into operation, which is important for companies considering investment in robotic automation. The announcement of an open safety benchmark also signals that the question of safety for robots operating near humans is being addressed.
What was added since the original report
Verified updates
-
Gemini Robotics 2.5 increased manipulation capabilities with a specific focus on deformable objects; The model was evaluated on more than 200 tasks; The Apollo 2 humanoid robot managed to tie a bow; Robots with the new version are already working at the Mercedes-Benz manufacturing plant; Google DeepMind announced the open source safety benchmark ASIMOV-jailbreak
- Gemini Robotics 2.5 increased manipulation capabilities with a specific focus on deformable objects
- The model was evaluated on more than 200 tasks
- The Apollo 2 humanoid robot managed to tie a bow
- Robots with the new version are already working at the Mercedes-Benz manufacturing plant
- Google DeepMind announced the open source safety benchmark ASIMOV-jailbreak
Relevant practical impact
What this means
For a business
Companies in manufacturing and logistics have a new signal about the maturity of robots' manipulation dexterity with deformable materials (textiles, cables, packaging film) and a concrete deployment reference at the Mercedes-Benz manufacturing plant, which they can use to assess the technology's readiness for their own automation projects.
DevelopmentCheck the original
Event sources
independently confirmed · 2 publishers, 1 independent. We count feeds from the same owner only once.