Quick take
Google DeepMind evaluated Gemini Robotics 2.5 on more than 200 dexterity tasks, reporting success rates 30% to 100% higher and completion times 20% to 30% faster.
On Apptronik’s Apollo 2 humanoid robots working on Mercedes-Benz production lines, DeepMind reported a 96% success rate for automotive kitting.
Ars Technica’s Ryan Whitwam said the factory result was likely below repetitive-production reliability, while DeepMind’s Kanishka Rao called the final 4% a “steep road.
DeepMind reported a 99.5% jailbreak-resistance result for Gemini Robotics 2.5 using its ASIMOV-Jailbreak physical-safety benchmark.
Why it matters
For manufacturers, the shift is from evaluating robots mainly through controlled demonstrations to considering them as flexible factory assistants, though human correction may still be necessary. The next questions are whether DeepMind can reduce the frequency and cost of failures, and whether cloud-based operation and its safety benchmark hold up und
Discussion
Start the conversation
Your voice can be the first to spark an engaging conversation.