Robots Are Finally Starting to Work
TL;DR
Physical Intelligence co-founder Quan Vang explains why robotics is approaching its 'GPT-1 moment,' where cross-embodiment AI models trained on diverse hardware are beginning to exhibit emergent zero-shot capabilities and scaling laws previously unseen in the field.
🤖 The Cross-Embodiment Breakthrough 3 insights
Generalist models outperform specialists
The Open X-Embodiment paper demonstrated that a single policy trained across 10 different robot platforms performed 50% better than individual models optimized for specific hardware, proving scaling laws apply to robotics.
Cross-embodiment solves hardware drift
Unlike single-robot approaches that suffer from hardware variations and software changes every few months, training on diverse platforms forces models to learn abstract 'robot control' rather than specific motor mappings.
Emergent zero-shot capabilities appearing
Current models can now perform complex precision tasks and multi-object reasoning zero-shot—capabilities that required hundreds of hours of data collection just last year.
🧠 From Language Models to Physical Control 3 insights
Vision-language models enable semantic control
Research like RT2 and PaLM-E showed that adapting powerful vision-language models to output robot actions transfers internet-scale knowledge to low-level control, allowing commands like 'move the can to Taylor Swift' without specific training.
SayCan bridged planning and semantics
SayCan was the first demonstration of using language models for robotic planning, leveraging common sense knowledge to reduce the need for task-specific robot data.
The three-pillar hierarchy
Robotics requires solving semantics (now unlocked via LLMs), planning, and real-time control—the latter being the final frontier where vision-language models are now making breakthroughs.
📊 Deployment Strategy & Data Reality 3 insights
The data capture problem vs. generation problem
While robotic data is constantly generated in labs and industry, there has been no incentive to capture it in standardized formats; Open X-Embodiment was the first effort to aggregate multi-robot data but remains 'a drop in the bucket' compared to what's needed.
Mixed autonomy enables immediate scaling
Current systems are deployable today using a 'peeling an onion' approach: deploy a strong base model with human oversight for edge cases, then improve through real-world exposure rather than waiting for full autonomy.
Real-world validation with YC companies
Physical Intelligence partnered with Weave to deploy laundry-folding robots in commercial laundromats, demonstrating that current models can handle deformable objects and unstructured environments with human assistance.
Bottom Line
Founders should start building vertical robotics applications now using mixed autonomy systems, as cross-embodiment AI models are rapidly generalizing across hardware and approaching the capability to perform complex physical tasks without task-specific training.
More from Y Combinator
View all
How A Prototype Built During A Missed Flight Became A New Gusto Product
Gusto co-founder Eddie Kim explains how a prototype built during a 5-hour airport delay evolved into Gusto Co-founder, an AI agent that automates repetitive small business tasks by leveraging existing customer data and simple chat interfaces rather than requiring technical expertise.
India Can Create The Largest AI Companies
India is positioned to create the world's largest AI companies because the technology rewards deep technical expertise over local market knowledge, leveling the global playing field and allowing Indian founders to win enterprise customers through cold outreach and superior product merit rather than geographic proximity or networks.
Zynga Founder: Consumer Is Not Investible Right Now - Thats Why You Should Build It
Zynga founder Mark Pincus argues that while consumer startups are currently out of favor with investors, AI agents create unprecedented opportunities to reinvent everyday services. He shares his "Proven Better New" product framework and explains why founders must kill their ego to survive the inevitable failure of novel features.
Why Domain Experts Are Winning Right Now
Bryant Chou, co-founder of Webflow, demonstrates how his new startup Ploy enables domain experts to autonomously execute world-class marketing and web design, arguing that deep industry experience is becoming the ultimate competitive advantage for leveraging AI effectively.