This might be one of the clearest examples of why robot training data matters. @axisrobotics recently showed that carefully selected simulation data could take average real robot success from 16.7% to 85.8%, using just 10 real world demonstrations. And the interesting part? It wasn't simply about throwing more simulation data at the problem. The right data made the difference. That finding makes the latest Axis Robotics partnership even more interesting. With @lotuscars and @GeelyAutoGlobal, Axis Suite is taking its Sim2Real approach into automotive manufacturing, where one simulation trained policy can handle 500–800 distinct automotive parts for production-line sorting. So what we're seeing is a progression from: Research → simulation → training → real-world deployment. The research shows why carefully selected simulation data matters. The partnership shows what that approach can look like when it reaches an actual production environment. This is the part of Physical AI I'm watching closely. How far can Sim2Real scale beyond automotive manufacturing? you can follow the Axis campaign too, here s.kaito.ai/wES8vK7 #AxisRobotics #PhysicalAI #Sim2Real