Boom

The bottleneck for mass-producing humanoids is not just AI. It's the actuator and the supply base that sits underneath it. Study these companies for each of the components. - Japan dominates precision mechanical (reducers, bearings). - Germany and Switzerland own high-end motion. - China owns volume motors, reducers, and the upstream magnet supply. Who am I missing?

Eternal Sunshine of the Middle Managerless Mind

Our mission is to make it easy for anyone to deploy a robot to help them in the real world We wrote an intuitive guide to understanding modern robotics, catered toward an audience that understands technology but not AI robotics We hope that this short blog post embeds in you the core principles that will bring further curiosity.

How do drones swarm without a leader? How does infrastructure heal itself? How do autonomous systems fail gracefully? @AmarSBedi of @tashiprotocol and @mikeahorton of @GEODNET join @RichardRobinson to chat orchestration in the physical world. Join: x.com/i/spaces/1jGXggydjgBKZ

Networking, inspiration, and tandoori pizza await at industry group @Intercognitive's foundation event in SF next Thursday. If you're working on physical AI, don't miss this one!

Your phone's 3-10m accuracy is fine for Google Maps. But for a drone landing in a backyard, a car navigating, or an autonomous tractor running between crop rows? 3 meters of drift is a crash waiting to happen. Physical AI needs @GEODNET 🧵 👇

The bottleneck of frontier robotics isn’t compute, labeling, or the models themselves. It’s data collection. While language models scaled effortlessly on open internet text, robotics requires physical trajectories, motor torques, and tactile forces that cannot simply be scraped from a webpage. Every token has to be fought for. Here is a breakdown of the 7 data types shaping the industry today, each representing a trade-off between collection cost and action-label purity: 1. Real Teleoperation (AgiBot World, DROID). Collected by humans guiding hardware, it scales linearly with human hours. 2. Low-cost Capture (Mobile ALOHA, UMI handheld). It drives collection cost down while keeping real physics, though it introduces an embodiment mapping problem when transferring human hand actions to robotic joints. 3. Fleet / Deployment Data (Tesla Optimus, Figure). These are trajectories from robots already working in the field. Tesla is betting its automotive fleet infrastructure transfers to Optimus. It generates powerful, real edge cases, but requires scaled deployment. 4. Simulation (NVIDIA Isaac Sim, Genesis). While offering near-infinite scale, the sim-to-real gap still struggles to model contact-rich dynamics like slipping, twisting, friction. 5. World-Model Synthetic (NVIDIA Cosmos 3). NVIDIA just shipped Cosmos 3, which natively outputs action trajectories, not only video pixels. If a world model can accurately simulate the laws of physics natively, it reduces the need for manual teleop data drastically. 6. Egocentric video (Ego4D, Meta’s Project Aria). First-person human video captured with head-mounted rigs. Far more scalable than teleop and closer to a robot’s own viewpoint. Still carries no robot action signals on its own. 7. Internet video (Youtube, TikTok). Maximum scale, lowest cost, effectively free. It captures the widest range of objects, tasks and physical situations, but with zero action labels and (mostly) a third-person viewpoint. Collecting data is only the step one. The next great execution challenge is engineering a coherent training recipe that can blend these heterogeneous data sources into a single model.

Talking about 2030 should be fun.

From Shibuya to the world.

Demo: spatial compute node calculating a path for a robot. In a future where we send robots to do our errands, venues may not want to give those visiting robots their whole map. Instead, robots will just communicate with the local spatial compute node and get a path.