Rex Zhang

@snarkolepsy Thank you so much! Really appreciate the support. 🙌

@RexDQZhang Wishing you a lot of success.

Welcome to the official PathOn Robotics 'Testing Facility' (aka the robot gym). Also known as the corner of my basement right next to my gym. We've been running our custom sensor/software on the Go2 for our initial GTM. Bypassing the factory brain and injecting an enterprise-grade OS requires zero fancy office space—just a clear floor, a bright light, and absolute focus. Hyper capital-efficient build mode. Let's keep shipping.🤖🛠️

@KZLi10 It is A1. Behind it is P1S.

@RexDQZhang Is that a Bambu A1, but the print base is quite large.

We don't spend money on an office. R&D lab = dining room. Workshop = basement. I'm a solo founder building robots. 🧵

@web3MIO We raised a Pre-Seed SAFE last year. But my salary is exactly $0 and our office rent is $0. 100% of the capital goes directly into hardware, compute, and R&D. Extreme capital efficiency is how we survive and win.

@RexDQZhang How do you stay funded?

@yangWao Good eye! 📷 That’s a Livox MID-360 on a Lekiwi base. We use it to stress-test our 3D navigation stack.

@RexDQZhang lekiwi hi5

Six months in: tight team, design partners, a working product, live demos — all from a dining table and a basement. Every dollar goes into the build. Sharing it all openly from here. 🚀 #buildinpublic #solofounder #robotics

13 years to get here — Peking U → Berkeley PhD → autonomous vehicles → Amazon → LLM papers → web agents. None of it lit me up. May 2025: picked up LeRobot, won the SF hackathon a month later 🏆 July: founded the company July–Oct: every rookie mistake — sunk Nov: burned the boats 🔥, went solo

Excited to share the first version of our in-house dexterous hand design at @pathon_robotics ! 🦾 We're building a unified dexterous manipulation pipeline that works across multiple hand platforms — both off-the-shelf hands and our own design, which we're now prototyping. Next week, we'll integrate eFlesh, a low-cost tactile sensor, to give the system rich contact feedback for fine manipulation. @Raunaqmb The goal: a hardware-agnostic pipeline where you can plug in your preferred hand + sensor stack and get dexterous manipulation out of the box. More updates coming soon 👇

@LeRobotHF @therobotstudio @IlirAliu_

Stock SO-101: 5 DoF + non-symmetric gripper. SOTA grasp models: 6D poses + symmetric grippers. So we upgraded the SO-101 to 6DoF and designed a symmetric parallel-jaw gripper — so any SOTA grasp model drops in. 🦾 Here's the full arc 🧵👇

Fully open-sourced — reproduce the whole upgrade: 🧩 STL + STEP for every part 📋 Bill of Materials 🔧 Step-by-step assembly guide 📸 Print orientation diagrams github.com/PathOn-AI/path… What should we grasp next? 🤖

Step 3 — Real-robot vision-language grasping: 4 unrelated objects. 4 grasps. 1 pipeline. 🧸 toy bear 🟢 foam cylinder 🟥 red cube 🖊️ whiteboard eraser Tell it what to pick up. It figures out the rest. ⚡

Step 2 — Real robot integration with ROS2: 🦾 MoveIt for motion planning ☁️ Point cloud scene understanding 📐 Full URDF + TF tree Reliable motion planning is the foundation no CV accuracy can replace.

Step 1 — Vision-language grasping in simulation: 🔍 SAM3 for segmentation 🦾 Grasp gen model → 6D pose 💬 "pick up the banana" End-to-end, no hardcoded logic.

Introducing Hi Robot – Hierarchical Interactive Robot Our first step at @physical_int towards teaching robots to listen and think harder. A 🧵 on how we make robots more steerable 👇

My Chinese name is Danqing. Uber/Lyft always pronounced it as either "DanKing" or "DanQueen." I started telling people yeah I'm DanKing When I started my company I needed an English name people would remember. Tried "Danni" first. Too soft. Not me. I always loved the name Rex 🦖 But it's a guy's name. Spent time looking for a female version. Then I thought — who says I can't be Rex? I'm already DanKing 👑 So Rex it is. So Rex it is.

The power of the Claw, in the palm of a robot hand. Agentic robotics is here! Today, we open-source CaP-X: vibe agents, alive in the physical world. They incarnate as robot arms and humanoids with a rich set of perception APIs, actuation APIs, and auto synthesize skill libraries as they go. CaP-X is a strict superset of our old stack, because policies like VLAs are “just” API calls as well. It solves many tasks zero-shot that a learned policy would struggle with. And we are doing much more than vibing. CaP-X is our most systematic, scientific study on agentic robotics so far: - We build a comprehensive agentic toolkit: perception (SAM3 segmentation, Molmo pointing, depth, point cloud), control (IK solvers, grasp planner, navigation), and visualization (EEF, mask overlays) that work across different robots. - CaP-Gym: LLM’s first Physical Exam! 187 manipulation tasks across RoboSuite, LIBERO-PRO, and BEHAVIOR. Tabletop, bimanual, mobile manipulation. Sim and real. Can’t wait to see the gradients flow from CaP-Gym to the next wave of frontier LLM releases. - CaP-Bench: we benchmark 12 frontier LLMs/VLMs (Gemini, GPT, Opus, Qwen, DeepSeek, Kimi, and more) across 8 evaluation tiers. We systematically vary API abstraction level, agentic harness, and visual grounding methods. Lots of insights in our paper. - CaP-Agent0: a training-free agentic harness that matches or exceeds human expert code on 4 out of 7 tasks without task-specific tuning. - CaP-RL: if you get a gym, you get RL ;). A 7B OSS model jumps from 20% to 72% success after only 50 training iterations. The synthesized programs transfer to real robots with minimal sim-to-real gap. 3 years ago, our team created Voyager, one of the earliest agentic AI that plays and learns in Minecraft continuously. Its key ideas — skill libraries, self-reflection loops, and in-context planning — have since influenced many modern agentic designs. Today, the agent graduates from Minecraft and gets a real job. It’s April Fool’s, but this Claw is getting its hands dirty for real! Link in thread:

不要随便和强者对话 否则你真的会抑郁 张雪这股强烈的明确自己人生意义并一定要去做的信念 到底是怎么来的？只能说是天赋了 和他交流多了连采访人都貌似有点抑郁和自我怀疑 “我自己想要的又是什么呢？我自己怎么没有他这么强大的信念呢？是我自己有什么问题么？”

🤲Tactile sensing is powerful for robot manipulation, but hardware is still difficult to access, reproduce, and scale. 🎯That’s why we built FlexiTac: an open-source, low-cost, and scalable tactile sensing solution designed for real robotic systems. • Project page: flexitac.github.io We hope FlexiTac can help democratize tactile sensing for robotics research. (1/n)