Google DeepMind unveils AI for robot control and predicting the effects of DNA variants

Executive Briefing
“Google DeepMind has unveiled Gemini Robotics 2.5 for robot control and AlphaGenome 2 for...”
Key Takeaways
- 1.DeepMind introduced Gemini Robotics 2.5 as a new vision-language-action model. The company says it is intended to help robots learn skills and operate in real-world settings that are less structured than a laboratory.
- 2.DeepMind reported a 96% success rate for packing automotive parts in a real-world proof-of-concept test at Mercedes-Benz. In other evaluations, success ranged from 30% to 100% by task, with completion times 20% to 30% shorter than the previous version’s baseline.
- 3.Gemini Robotics 2.5 is being rolled out through DeepMind’s robotics partners, including Agile Robots, Apptronik and Boston Dynamics.
- 4.AlphaGenome 2 adds predictions for specific cell types and states, potentially exposing effects of DNA variants that tissue-level models miss. Its predictions are research leads, not substitutes for experiments, and the model is not approved for clinical or diagnostic use.
- 5.According to DeepMind, AlphaGenome 2 predicts DNA methylation, a process involved in regulating gene activity, alongside outputs available in the previous generation such as gene expression and chromatin accessibility.
Google DeepMind has unveiled Gemini Robotics 2.5 for robot control and AlphaGenome 2 for predicting how DNA variants affect cells. The robotics model is reaching partners, while the biological predictions still require experimental validation.
Key Developments and Technical Context
DeepMind introduced Gemini Robotics 2.5 as a new vision-language-action model. The company says it is intended to help robots learn skills and operate in real-world settings that are less structured than a laboratory.
DeepMind reported a 96% success rate for packing automotive parts in a real-world proof-of-concept test at Mercedes-Benz. In other evaluations, success ranged from 30% to 100% by task, with completion times 20% to 30% shorter than the previous version’s baseline.
Gemini Robotics 2.5 is being rolled out through DeepMind’s robotics partners, including Agile Robots, Apptronik and Boston Dynamics.
AlphaGenome 2 adds predictions for specific cell types and states, potentially exposing effects of DNA variants that tissue-level models miss. Its predictions are research leads, not substitutes for experiments, and the model is not approved for clinical or diagnostic use.
According to DeepMind, AlphaGenome 2 predicts DNA methylation, a process involved in regulating gene activity, alongside outputs available in the previous generation such as gene expression and chromatin accessibility.
Strategic Analysis and Industry Implications
The Mercedes-Benz result offers a concrete industrial use case, but success on parts packing cannot establish how well the model performs across every robot task.
If experiments validate its cell-state predictions, AlphaGenome 2 could help researchers prioritize DNA variants and potential drug targets for investigation, rather than serve as a medical conclusion.
Outlook and Key Factors to Watch
The source does not detail the task set, test conditions or methods used to calculate Gemini Robotics 2.5’s success rates.
The source provides no independent validation results for AlphaGenome 2’s predictions in newly profiled cell types and states.







