Post by Centific

The Centific Robotics & Physical AI team set out to measure a problem robotics researchers have struggled to quantify: how much robot behavior changes when moving from simulation to the real world. By analyzing more than 1,400 teleoperated robot episodes, the team put hard numbers on a challenge that had largely been discussed anecdotally. Working with N1 Robotics, we then evaluated how neural retargeting preserves human grasp intent, making simulated training data behave more like real-world manipulation and improving the quality of data used to train robotics foundation models. Our research suggests that better robot learning depends not just on larger models, but on training data that faithfully captures how people manipulate objects in the physical world. Read more: https://lnkd.in/gVh2wqYc #Centific #Robotics #PhysicalAI