Found the next G.O.A.T. It just needs charging. 🐐🔋

Robots have entered the game. What’s next? ⚽🤖 S1 developed its football skills through 140 years of simulated play. The trained policy was then transferred onto a real robot, bringing those skills and strategies onto the pitch. The most surprising part? Its best opponent is itself. Through self-play against recent versions of its own policy, the competition gets stronger as it improves, pushing it to adapt, discover new strategies, and keep getting better.

Deployment is where every AI model meets its toughest critic: REALITY. It's the hardest and most consequential part of the journey. Unseen conditions expose weakness across the system. That's where AI faces it's final boss: proving whether polished lab demos translate into useful work. We're meeting that challenge by tracking failures, fixing the causes and building the reliability that everyday operations demands. Tackling deployment challenges early in my career has been deeply rewarding and I couldn't have asked for a better learning experience. Proud to see the @SkildAI team turning ambitious ideas into dependable systems that can stand up to the real world.

What if teaching a robot a completely new task was as simple as showing it a video? One of the biggest challenges in AI is DATA. Robots need massive amounts of diverse training data to handle variations in environments, objects, physical constraints, and unexpected situations. The ability to overcome much of this complexity through a single video demonstration is truly mind-blowing and could be another major breakthrough for AI and robotics. S1 demonstrates this in practice, learning unseen, long-horizon tasks from a single video without task-specific post-training. It can compose skills, adapt to changes, and recover from mistakes. On unseen tasks, S1 achieved 66% success vs. 9% with language prompting. One video. A new task. No task-specific post-training. An exciting step toward truly general-purpose robots. 🤖 Read the full S1 blog: skild.ai/blogs/s1?v=1 Brilliant work by the entire @SkildAI team! 🎉 @deepakpathak @gupta_abhinav_