谁是藏镜人

Howdy, i'm Alec 👋 I'm a researcher focused on humanoid agility and control. I use reinforcement learning to make physics-based characters run, climb, vault, slide, recover, handle objects, and more. It is my goal to make embodied intelligence easier to direct, whether that be for humanoid robots or digital characters. I'm also a co-author of Metatron, a research collaboration with NVIDIA exploring how motion goals can guide whole-body movement.
research.nvidia.com/labs/par… This reel brings together some of what I've been working on over the last year. I post here daily, sharing my progress, my failures, and what I learn along the way. If you're working on humanoid robots, animation, or games, feel free to say hi!

Mini 6-Axis Robot Arm: Building a robotic arm from scratch is a pretty impressive engineering challenge! 🤖 This low-cost 6-DOF robotic arm was developed to explore what it actually takes to build and control a robot in the real world. And there's a lot more involved than simply making six joints move. The arm features modular joints, custom-designed gearboxes, and a spherical wrist that allows the end effector to approach objects from different orientations. What's particularly interesting is the combination of mechanical design and software. The project covers everything from low-level firmware and motor control to motion planning with ROS 2. And that's where things get challenging. In simulation, you can test movements without worrying too much about mechanical imperfections. On real hardware, you suddenly have to deal with friction, backlash, motor limitations, and the physical behavior of every joint. Even a relatively simple movement requires all these components to work together. I love projects like this because they make robotics engineering more accessible while showing just how much work goes into building a functioning robot. Sometimes the best way to understand robotics is to build one yourself! ⚙️ jcgullberg.github.io/project… Credit: @jamesgullberg

We're building a multi-layered robotics architecture, where the LLM drives high-level reasoning and skill harness. Here's a demo. The task: use the right arm to arrange three objects from the tray in a row along the X axis of the table, ordered by height from tallest to shortest. You can see the agent turn the task into continuous motion through skill harness, linking scene understanding, height-based ordering, and placement planning to physical execution. This is an exciting start.

Built an interactive 3D Task & Motion Planning (TAMP) + Fleet Scheduling simulator scaling from a 6-slot Franka FR3 workcell up to a 23-robot Smart Factory (19 FR3 arms + 4 mobile manipulators). 👉tamp.mincasurong.com Scaling the Temporal Execution Hypergraph (TEHG) up to a 6-cell factory floor (19 FR3 arms + 6 mobile manipulators). Task planners shouldn't dispatch trajectories that violate mounting geometry. ​Testing a Plan 👉 Gate 👉 Write-Back loop on a reconfigurable 3-arm Franka FR3 workcell (20 mount layouts). #Robotics #PhysicalAI #FrankaEmika #Automation #TAMP #IsaacSim

Got a bit carried away with Solstice again haha. Added proper volumetric clouds and changing weather. The clouds move with the wind, cast shadows and actually affect the GI, skylight and reflections. All running in Edge using Rust + WebGPU. Demo: samg-coder.github.io/solstic… Code: github.com/SamG-Coder/solsti…

this is the most underrated robot demo you'll see this week @VerneRobotics used under 2 hours of post-training data to teach its open-weight WAM to load and move a snack cart using both arms and its mobile base. makes you wonder how much data your own workflow would need :)

Introducing our Open Modular Humanoid Arms🦾. Use them for building your humanoid on a wheeled base or disassemble them and rebuild them into humanoid legs and run RL locomotion. Better capability with @RobStride_com actuators, 5kg max load. Making something people want🤗

Robots learn by imitating human teleop demos. But some tasks are too hard to demonstrate Raw VR teleop success for hard tasks: 0% 🍅, 10% 🖍️, 0% 🍷 Our CoRL work, GLIDE, can bump this to 90%! How? By smartly editing teleop code. A refreshingly robotics-first way to use LLMs!

🎉Sharing our SIGGRAPH 2026 paper Divide and Truncate(DAT): one rule that guarantees penetration-free contact of robots, soft bodies, cloth, and multi-physics. Robust, blazing fast and no deadlocks.🧵 with @JerryHsu32 @milesmacklin Youssef Ayman Website: ankachan.github.io/Projects/…

Introducing AIRoA-0 (A-0) 🤖 A VLA that learns long-horizon, precise bimanual manipulation from 16,151 hours of robot-free data. Co-led with the amazing @yh_kris 🙌 Weights, dataset, code & recipe are open. 🔗 a-0.airoa.io 🧵👇