Thinh Le retweeted
Every great machine begins on a workbench. Now you can build a humanoid robot on yours. docs.menlo.ai/
Thinh Le retweeted
Asimov is now zero-shot sim2real for locomotion.
You train the policy in simulation. It runs on the robot without additional tuning.
Blog: menlo.ai/research/zero-shot-…
Order yours: menlo.ai/asimov-1
Thinh Le retweeted
Built a robot configurator using open-source CAD files from Asimov 1
Opus 5 picked the materials it thought fit best and generated a bunch of poses to test
One thing missing from the CAD files: hands
So, That’s the next challenge for the models
Thinh Le retweeted
🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!
🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇
🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex!
Check out the configuration details in our official API docs: api-docs.deepseek.com/quick_…
Thinh Le retweeted
DiffusionGemma is an open, experimental model that brings our text diffusion research to Gemma 4. It’s a racehorse 🏇achieving up to 4x faster inference by generating entire blocks of text simultaneously vs predicting token-by-token (word-by-word) output!
Thinh Le retweeted
tried video spatial reframing, captured by iphone 14pro
#spatialreframing #4DGS
Thinh Le retweeted
This is what the future of robotics looks like in 2026
No expensive lab
No waiting weeks for parts to arrive
No crashing your drone and destroying hours of work
Just open your browser drag and drop components and test real flight physics instantly
Check this out
FPV hexacopter fully assembled in the simulator
Live flight controls
Real time telemetry showing altitude speed and attitude
You can iterate and test immediately
This kind of browser based simulation is going to change everything for students and builders
The people who learn to design and test in software first will move way faster than everyone else stuck waiting on hardware
What do you think
Is real time simulation the biggest change coming to mechanical engineering and robotics?
Thinh Le retweeted
You can train drones in 30 seconds. This was a year ago. PufferLib is 5x faster now!
Thinh Le retweeted
We built a humanoid robot from scratch in 3 months, got 100+ pre-orders in weeks, and tomorrow I’m talking about how we see robotics at @aiDotEngineer. Come through. Let’s talk agent-based robotics.
Thinh Le retweeted
We're open-sourcing Asimov v1, a humanoid robot.
With Asimov v1, you can build, train on, and make it your own humanoid robot. It's the first step of building a humanoid labor force for the rest of us.
Asimov v1 is 1.2 m tall, 35 kg, with 25 actuated degrees of freedom. Structural parts machined in 7075 aluminium and 3D-printed in MJF PA12 nylon.
We're releasing the mechanical design and simulation files. Ready for locomotion policy training out of the box.
The BOM is open too. Source everything yourself, or order the DIY Kit. All components, ready to assemble. $499 deposit, $15,000 target price. Ships end of summer 2026.
GitHub: github.com/asimovinc/asimov-…
Manual: manual.asimov.inc
DIY Kit: asimov.inc/diy-kit
Most humanoid robots are controlled by the companies that build them. Asimov v1 is built for the rest of us. Build it, test it, and share your feedback with the community.
Thinh Le retweeted
New ComfyUI Extension Alert: Mesh to Motion Explorer. 🧩🚀
This interactive 3D scene editor lets you:
✅ Choose from 124 preset motions (Human, Fox, Bird, etc.)
✅ Align reference images perfectly using Z-Image.
✅ Drive LTX 2.3 via Unicontrol / IC LoRA.
Turning the "uncertainty" of AI into "certainty." Workflow breakdown: 👇
🔗 youtu.be/w9EslLID3-o
#ComfyUITutorial #MachineLearning #OpenSourceAI #AITools
Thinh Le retweeted
Sam3 + Body is freaking amazing. I'm in the process of building an open-source @rerundotio and @Gradio demo that is similar to what Meta provided. Got the basic functionality up and running, now I need to hook it up to a Gradio interface. It's a really good model
Egocentric slam collection on my own hardware achieved with @rerundotio! Previously, I'd all my own ego collections used a static camera.
This is obviously not a good long-term solution and only really made sense for initial development. Now I can easily collect samples that have super-accurate slam from a quest device, and the giant field of view from a stereo + RGB camera rig.
As you can see in the video, I'm still working through some bugs, but I wanted to share some progress anyway. Hoping to have all kinks worked out by Friday.
The main thing I need to fix is a weird offset happening between each of the rigs. I need to stop representing cameras as individuals, but rather as rigs that have fixed relative positions to ensure rigid motion.