Skild AI

Pittsburgh International Airport is excited to announce two new projects now being tested through xBridge, our innovation program that connects airport teams, airline partners and technology companies to evaluate emerging solutions in a real-world airport environment. As Pittsburgh continues to lead in robotics and technology innovation, PIT remains a place where new ideas can be tested, refined and applied to real-world aviation challenges. Skild AI is exploring how robotics can support airport operations by helping teams monitor air quality, identify service needs and gain additional visibility in busy terminal areas. Arrow Analytics, Inc. is testing technology designed to give airlines better insight into carry-on bag volume and overhead bin capacity, helping support smoother boarding and more informed operational decisions. Together, these projects demonstrate how xBridge moves innovation from concept to practical application, helping team members work more efficiently while enhancing the passenger experience. Discover more in Blue Sky News: https://brnw.ch/21x63Nj.

Huge congratulations to the team, this is very cool work, Deepak Pathak and Skild AI! What people may not realize: in Go, the simulator is the game, so self-play is exact. In robotics, the simulator is only an approximation of physics, and every one of those 140 simulated years has to survive the sim-to-real gap. Getting that transfer to work is a great achievement here. Sim-trained policies are behind most of the impressive robot results we've seen, and this is a great example. I especially love the framing of self-play as post-training- one of the few ways to break past human capabilities.

140 years of soccer practice in simulation. ⚽ Skild AI’s robot policy learned to dribble, shield, and tackle through self-play in NVIDIA Isaac Sim before being transferred to a real humanoid robot.

Another big step toward general-purpose robotics. Self-play helped drive major breakthroughs in AI, including AlphaGo and AlphaStar. Skild AI is now bringing that learning loop into the physical world. S1 spent 140+ years playing soccer against itself in simulation and went from falling over to dribbling, tackling, recovering and playing against humans in the real world. And through self-play, S1 can keep improving without new human demonstrations. Incredible work from Deepak Pathak, Abhinav Gupta, and the Skild AI team! Make way for the #Messinator!

This is the AlphaGo moment for physical AI. Skild's S1 model taught itself to play soccer through pure self-play from over 140 simulated years of competing against its own past versions; going from falling over to dribbling, shielding the ball, tackling, and even showing early signs of teamwork. No human demonstrated any of it. Self-play unlocked superhuman performance in Go, StarCraft, and Dota. Deepak Pathak, Abhinav Gupta, and the Skild team are bringing that same recursive-improvement engine into the physical world — soccer is just the first testbed. We at Felicis backed Skild AI Series A because we believed robotics foundation models could learn the way language models do, from experience, not just human demonstration. Today's announcement proves it. Congrats to Deepak Pathak, Abhinav Gupta, and the whole Skild AI team. 🤖⚽