A robot that keeps crawling after losing all four legs to a chainsaw sounds like sci-fi horror, but it's actually the latest breakthrough in AI robotics. Startup Skild AI just demonstrated their "omni-bodied brain" - a single AI model that can control any robot hardware and adapt to catastrophic damage in real-time, marking a potential ChatGPT moment for physical AI.
Skild AI just shattered the robotics playbook with a demonstration that's equal parts impressive and unsettling. The startup's AI-powered quadruped robot continues moving even after researchers hack off all four legs with a chainsaw - a feat that CEO Deepak Pathak calls proof of "a new, more general kind of robotic intelligence."
The breakthrough centers on what Skild calls an "omni-bodied brain" - a single AI model trained to control dozens of different robot bodies across countless tasks. Unlike traditional robotics AI that's custom-built for specific hardware, Skild's approach creates one brain that adapts to any physical form it encounters, even ones it's never seen before.
"Any robot, any task, one brain. It is absurdly general," Pathak told Wired's Will Knight in an exclusive demonstration. The company's LocoFormer model learns continuously from experience, rebuilding its understanding of physics when faced with dramatic changes like missing limbs or motor failures.
The financial stakes are enormous. Skild AI raised $300 million in 2024 at a $1.5 billion valuation, positioning itself against deep-pocketed competitors like the Toyota Research Institute and rival startup Physical Intelligence. The race mirrors the early days of large language models - whoever cracks general robotic intelligence first could dominate an industry worth hundreds of billions.
What makes Skild's approach revolutionary is its cross-platform adaptability. During testing, researchers placed a four-legged robot on its hind legs, and the AI immediately recognized the new configuration, operating the robot dog like a humanoid. The same model controlled two-legged robots, four-legged systems, and even wheeled platforms it had never encountered during training.
The damage adaptation goes beyond missing limbs. Skild's robots continue functioning when legs are tied together, motors fail, or weight distribution changes dramatically. In one test, a quadruped with both legs and wheels adapted to motor failures by balancing on two wheels like an unsteady bicycle. The AI rebuilds its internal model of physics in real-time, learning from falls and failures to improve performance.
This breakthrough addresses robotics' biggest challenge: data scarcity. Traditional methods like teleoperation and simulation don't generate enough training examples for AI models to achieve human-like adaptability. By training one model across dozens of robot types and thousands of scenarios, Skild creates the massive datasets that powered the AI revolution in text and images.
The implications extend far beyond walking robots. Skild is already testing the same approach for robotic manipulation, training their AI brain on simulated robot arms that can adapt to new hardware and changing environments like reduced lighting. The company is working with unnamed industrial clients, suggesting commercial applications are already underway.
"It is so exciting to me personally, dude," Pathak said about the results, though he acknowledges they might seem "creepy" to others. The chainsaw demonstration certainly evokes Terminator-style scenarios, but the real breakthrough is creating AI that understands physics and adaptation at a fundamental level.
Competitors aren't standing still. Toyota's Research Institute is developing similar general-purpose robot AI, while Physical Intelligence recently emerged from stealth with its own universal robotics approach. The winner of this race could define the next decade of automation across manufacturing, logistics, and consumer applications.
The technology works by applying transformer architectures - the same AI foundation behind ChatGPT - to robotic control. Just as language models learn patterns across text to generate coherent responses, Skild's model learns motion patterns across robot types to generate adaptive behaviors. The result is AI that reasons about physics rather than just following programmed instructions.
Skild AI's chainsaw-surviving robot marks a pivotal moment in robotics - the transition from specialized machines to truly adaptable artificial intelligence. While the demonstration might look like something from a horror movie, it represents the same fundamental breakthrough that transformed text AI with ChatGPT. As this technology matures and scales, we're looking at a future where robots aren't just tools programmed for specific tasks, but intelligent agents that can adapt to any physical challenge. The race to commercialize this capability is just beginning, and the winners will reshape industries from manufacturing to home assistance.
