Google DeepMind has released Gemini Robotics 2.5, a vision-language-action model that it says gives robots greater dexterity, reliability and safety in unstructured settings. The company tested the model on more than 200 distinct tasks, including serving food with tongs, tying a bow on a gift box and pipetting liquids. Compared with previous baselines, DeepMind reports a 30% to 100% higher success rate depending on the task type, with tasks completed 20% to 30% faster. The model works across several robot types, including Apptronik's Apollo 2 humanoid with a five-finger hand and the Franka Duo platform with two-fingered grippers. The model is now rolling out to robotics partners.
The most concrete test was a proof-of-concept with Apptronik and Mercedes-Benz on active automotive production lines in Kecskemét, Hungary, and Sindelfingen, Germany. There, Apollo 2 performed automotive kitting, picking metal brackets from bulk bins and sorting them into trays in the order each vehicle needed. A separate embodied reasoning model tracked the workflow while Gemini Robotics 2.5 controlled the arms and hands. The pilot reported a 96% success rate, and Apptronik said the models carried over to the Mercedes-Benz line with very little drop-off after a few weeks of post-training. DeepMind also said it has improved steerability so robots can avoid physical hazards in real time, which it says raised safe task completion around obstacles by 33% and around nearby people by 43%. It has also introduced a benchmark for testing robot-specific adversarial jailbreaks.
For Australian businesses and professionals, this changes little in the short term. Gemini Robotics 2.5 is available only to selected robotics partners, not as an off-the-shelf product, so there is no direct cost or purchasing decision to make yet. Manufacturers and logistics operators considering humanoid or industrial robot pilots may want to track the results, and any deployment would bring workplace health and safety duties and privacy questions if robots capture images of workers. Companies that are not evaluating robotics can safely ignore it for now.
Key Points
Gemini Robotics 2.5 is a vision-language-action model aimed at improving robot dexterity, reliability and safety.
DeepMind reports 30% to 100% higher success rates and 20% to 30% faster task completion across more than 200 tasks.
A Mercedes-Benz and Apptronik pilot achieved a 96% success rate on automotive parts kitting.
Safety features cut the rate of collisions with obstacles and people, measured as safe task completion, by 33% and 43% respectively.
The model is rolling out only to robotics partners for now, with no general public availability.