Post by SiliconANGLE & theCUBE

Skild AI solved robotics' data problem by embracing simulation at scale 🚀 In this #MACHINA Summit 2026 exclusive (sponsored by Path Robotics), #theCUBE’s John Furrier speaks with Abhinav Gupta, Co-Founder & President of Skild AI, about how their solution combines pre-training on videos and simulation to effectively create the massive training datasets robotics need without waiting for real-world deployment. “The biggest problem of robotics is the data problem: how do you get enough data to build robot models essentially? The problem is there are a few forms of data like videos and simulation where they scale, but these are not good enough forms of data. Then there are a few forms of data like teleoperation and robot data from deployment. These don't scale, but it's the best form of data,” Gupta shares. “At Skild AI, the way we are doing it is we first do pre-training with the poor form of data, which is videos and simulation. We first learn what the tasks are through lots of videos. The only problem is you cannot learn tasks from videos because if that was true, all of us will become Roger Federer. We'll just watch his videos and learn how to play like he can. Yeah, you can execute them, and that's where simulation plays a big role for us. We learn what the task is and then you go and practice in simulation at scale by changing conditions,” he adds. 💡 Get more insights! https://lnkd.in/gAqes6Uu #Robotics #PhysicalAI #SimulationData #DigitalTwins #EnterpriseAI

Video Content