Jiabin
36 posts
Introducing Subtask Annotations in LeRobot Datasets. LeRobot now supports subtasks as a native dataset field, so you can label task progression frame-by-frame, train hierarchical policies, and build stronger reward models with stage-aware supervision. Available today, with an annotation space + tooling to easily add subtasks and high-level dialogue to your dataset. huggingface.co/spaces/lerobot…
Introducing Helix 02 It's our most powerful model to date - it's using the whole body to do dishes end-to-end and it's fully autonomous
🦾🔥 2026 will be the year of robotics. And you should start a robotic company right now! Let me explain you why and show you the opportunities in the video – but here is an outline: We're in an Will Smith spaghetti moment. Remember how AI-generated video looked horrific two years ago? That's where robotics is right now. Computer vision is solved. VLAs (vision language action models) are starting to work. The reliability problem is being cracked as we speak. And unlike software, where you're competing against 15,000 marketing AI startups, humanoids has maybe 200 companies worldwide. Warehousing, the most crowded robotics vertical, has 700. Plus what are you going to do? Build a SaaS that claude can one-shot? The macro tailwinds are also obvious: Dark factories. Self-driving everything. Drones dominating warfare. China pushing automation hard. The West needing to reindustrialize with an aging workforce. But the real unlock is that small teams can now move incredibly fast. In the video we show robots built by one person, that is a year later already shown at CES, and raised couple million euros. Components costs are also dropping. Plus production suppliers actually want to work with startups now. In the video we are also going into opportunities. One mental model is simple: robotics is the next SaaS. Look at any industry, find one specific task, and build a robot that can do it better, faster, or around the clock. But we go through multiple mental models more in the video I uploaded the full video right here on X. But if you got a second, i'd appreciate a share, like, subscribe on youtube (link below!) ⬇️
Generalist robots don’t fail due to a lack of generality. They fail due to a lack of proficiency where it matters. We introduce SOP, enabling generalist policies to improve from real-world experience across distributed robot fleets, without sacrificing generality. 🧵 agibot.com/research/sop
Atlas tirelessly grasped the car's roof components. What were the human workers standing by and watching thinking?
