正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Replicated the laundry task in mujoco web, and also self-trained the RL WBC. By Opus 5.5 in 3 hrs. Agentic real2sim is nearly solved.

Vector Wang

2,492 subscribers

48,049 次观看 • 2 天前 •via X (Twitter)

艺术教育科学技术

Anya Rossi• Live Now

Private livecam show

35 条评论

Lindon Gao 的头像

Lindon Gao1 天前

Wow nice work!

Vector Wang 的头像

Vector Wang1 天前

Thanks Lindon! Thrilled to get acknowledged by you. I am always fascinated by what Dyna has done.

link 的头像

link1 天前

@ZeYanjie Cool! How did you trained a urr based rl policy?

Yannis He 的头像

Yannis He1 天前

forwarded to our eng 🫡

Saket Tiwari 的头像

Saket Tiwari2 天前

awesome! does mujoco web support cloth folding in sim?

Vector Wang 的头像

Vector Wang2 天前

Mujoco supports cloth simulation, but the performance is not good and can't really do real-time in web. So in the video the cloth is only presented as rigid body.

Leo 的头像

Leo2 天前

Did the three-hour build match how the towels behave, or mainly the robot and laundry room?

Vector Wang 的头像

Vector Wang2 天前

Mujoco supports cloth simulation, but the performance is not good and can't really do real-time in web. So in the video the cloth is only presented as rigid body.

Paolo AI 的头像

Paolo AI2 天前

@ttktjmt let's add this to MjSwan ?

Tim 的头像

Tim2 天前

Wow very impressive, I will look into that 🔥

Tony Rick 的头像

Tony Rick1 天前

laundry task in 3 hrs with opus is kinda nuts

Sergey 的头像

Sergey1 天前

Very cool, mindblowing what coding agents are capable of.

Ignacio López-Francos 的头像

Ignacio López-Francos1 天前

This is very cool

Feng Gao 的头像

Feng Gao2 天前

Was the whole simulation also replicated by Opus 5.5?

Vector Wang 的头像

Vector Wang2 天前

The robot trajectory is retrieved from official website (only kinematic) and Opus did the scene and robot dynamics to match this trajectory.

Haitong Wang 的头像

Haitong Wang1 天前

This is so cool. Are the towels modeled as well in sim?

Vector Wang 的头像

Vector Wang1 天前

Currently all rigid bodies so it's easier to run on web.

Chad L. 的头像

Chad L.2 天前

super cool!

Jakie PLA 的头像

Jakie PLA1 天前

3 hours? I could not help myself. This wants to live on a factory cart. Print brackets, bolt on wheels, and let Opus do the boring RL tuning. OH YES.

Max Riazantsev 的头像

Max Riazantsev1 天前

That looks great! Are you going to try the same with XLeRobot? It is cheap enough to actually test sim-to-real on the real hardware

Brian Zheng 的头像

Brian Zheng1 天前

Hi will you release the codebase

Alexander 的头像

Alexander1 天前

is this mujoco? the rendering quality look really high?

Vector Wang 的头像

Vector Wang1 天前

Mujoco web, so I can have additional renders customized (while still runs in real time in the browser)

Alexander 的头像

Alexander1 天前

what additional renders? how are you doing that

Vector Wang 的头像

Vector Wang1 天前

Basically I pre-baked the lighting using blender first.

Greyyy 的头像

Greyyy2 天前

The rigid-cloth clarification is useful. Which physical properties were estimated from the source assets, and which used defaults? A sensitivity sweep on friction and inertia would be interesting before treating this as a faithful dynamics model.

Jerry Chéng 的头像

Jerry Chéng2 天前

Wow this is insane! Did the real2sim modeled the robot as well? Would be great if you can share the harness - I have been trying lately as well

Vector Wang 的头像

Vector Wang2 天前

Robot file is extracted from Dyna's website and added physical properties, and also generated collision mesh. Didn't use special designed harness, just prompted my requirements in claude code.

Jerry Chéng 的头像

Jerry Chéng2 天前

That's sooo brilliant! Love the idea - and still amazed by the real2sim abilities, agentic robotics is here

刘海若（Hairuo Liu）的头像

刘海若（Hairuo Liu）2 天前

will it be open-sourced?

Vector Wang 的头像

Vector Wang2 天前

After I confirm with Dyna, it’s their IP after all

xard 的头像

xard2 天前

again it needs to be proved effective when deployed in real world!

newuhe 的头像

newuhe1 天前

Nice work！How do you generate URR data？

Vector Wang 的头像

Vector Wang1 天前

~98% synthetic (random keyframes + min-jerk + step targets → FK), seeded by ~97 s of Dyna's public ego-replay clips

rozgo 的头像

rozgo1 天前

This looks really good. Nice work.

相关视频

Is Opus 5.5 better than GPT-6 Astra at Physical Intelligence "literally" ? put GPT-6 Astra and Claude Opus 5.5 in two Unitree G1s in MuJoCo Ring Fence. Every 2 sim seconds: state → LLM → joint targets → act. Goal: force the other out of the ring, and Astra won after 31 second.

Is Opus 5.5 better than GPT-6 Astra at Physical Intelligence "literally" ? put GPT-6 Astra and Claude Opus 5.5 in two Unitree G1s in MuJoCo Ring Fence. Every 2 sim seconds: state → LLM → joint targets → act. Goal: force the other out of the ring, and Astra won after 31 second.

Jiafei Duan

54,246 次观看 • 21 天前

We gave Sonnet 5.5, Haiku 5.5 and Opus 5.5 the same task in Motion: “Make a motion design video about Claude.” Here’s what each one made 👇

We gave Sonnet 5.5, Haiku 5.5 and Opus 5.5 the same task in Motion: “Make a motion design video about Claude.” Here’s what each one made 👇

Motion

71,305 次观看 • 6 天前

🚀 Anthropic's Claude Opus 5.5 is now generally available in GitHub Copilot. Our early testing shows it stands out in efficiency, delivering task resolution comparable to Claude Opus 5 while using significantly fewer steps and tokens, with fast error recovery on multistep tasks. Try it out for agentic coding, long-running agentic tasks, and knowledge work. Find it in the GitHub Copilot app, CLI, and Visual Studio Code.

🚀 Anthropic's Claude Opus 5.5 is now generally available in GitHub Copilot. Our early testing shows it stands out in efficiency, delivering task resolution comparable to Claude Opus 5 while using significantly fewer steps and tokens, with fast error recovery on multistep tasks. Try it out for agentic coding, long-running agentic tasks, and knowledge work. Find it in the GitHub Copilot app, CLI, and Visual Studio Code.

GitHub

123,922 次观看 • 22 天前

I asked Opus 5.5 to make a game inspired by the discovery of the first computer, the Antikythera mechanism. Everything you see and hear is procedurally generated in real time, from code by Opus 5.5. It's a single 3 MB HTML file. Playable Claude artifact below.

I asked Opus 5.5 to make a game inspired by the discovery of the first computer, the Antikythera mechanism. Everything you see and hear is procedurally generated in real time, from code by Opus 5.5. It's a single 3 MB HTML file. Playable Claude artifact below.

edwin

394,747 次观看 • 22 天前

Jev + Opus 5.5, side by side. same agent. same task. same 60 decisions. Opus-only: $2.24 - 3m 37s Jev + Opus: $0.43 - 1m 15s the difference is who makes the small calls: which tool runs next - Jev is this output good enough - Jev is the goal met - Jev rewrite the refund handler - Opus 57 of 60 decisions never touched Opus. what you're watching: left routes every decision through Opus 5.5. right asks Jev first, Opus only fires when the task needs real reasoning. per decision: $0.032 -> $0.00025. watch the cost counters, not the output. the output is identical. full setup in the post below ↓

Jev + Opus 5.5, side by side. same agent. same task. same 60 decisions. Opus-only: $2.24 - 3m 37s Jev + Opus: $0.43 - 1m 15s the difference is who makes the small calls: which tool runs next - Jev is this output good enough - Jev is the goal met - Jev rewrite the refund handler - Opus 57 of 60 decisions never touched Opus. what you're watching: left routes every decision through Opus 5.5. right asks Jev first, Opus only fires when the task needs real reasoning. per decision: $0.032 -> $0.00025. watch the cost counters, not the output. the output is identical. full setup in the post below ↓

Mahax

19,565 次观看 • 16 天前

Trained in dingsim. And blue is the same policy but piloting a robot using HF's shipped mujoco simulator. Sim2sim transfer, trained in my Sota rigid body simulator. Trained in under 2 minutes on one 4090 Kneel

Trained in dingsim. And blue is the same policy but piloting a robot using HF's shipped mujoco simulator. Sim2sim transfer, trained in my Sota rigid body simulator. Trained in under 2 minutes on one 4090 Kneel

kache

23,485 次观看 • 1 个月前

Everything in this city is procedural and endless. Running in a browser. WebGPU. 100% code (by Opus 5.5)

Everything in this city is procedural and endless. Running in a browser. WebGPU. 100% code (by Opus 5.5)

Ryan Sael

700,596 次观看 • 6 天前

Really excited to release mjviser, a web-based MuJoCo viewer, powered by Viser. It has almost all the features of the native MuJoCo viewer, but runs in your browser. Load and simulate any MuJoCo model with a single uv command 👇 uvx mjviser

Really excited to release mjviser, a web-based MuJoCo viewer, powered by Viser. It has almost all the features of the native MuJoCo viewer, but runs in your browser. Load and simulate any MuJoCo model with a single uv command 👇 uvx mjviser

Kevin Zakka

22,902 次观看 • 6 个月前

Claude Opus 5.5 completely MOGS GPT 6 Sol. Same prompt. Same task. Look at the results. Opus 5.5 has way better taste and design. It is not subtle. You can see it immediately. GPT 6 Sol is a step up from GPT 5.6 Sol. But next to Opus 5.5 it is not even close. Anthropic is back.

Claude Opus 5.5 completely MOGS GPT 6 Sol. Same prompt. Same task. Look at the results. Opus 5.5 has way better taste and design. It is not subtle. You can see it immediately. GPT 6 Sol is a step up from GPT 5.6 Sol. But next to Opus 5.5 it is not even close. Anthropic is back.

Bridgebench

55,327 次观看 • 22 天前

My vid on Opus 5.5 and GPT 6.1 Sol: - Sol 6.1 is miles better than 6.0, it's worth trying out & the price is nuts - Opus 5.5 is the best Claude model in a very long time

My vid on Opus 5.5 and GPT 6.1 Sol: - Sol 6.1 is miles better than 6.0, it's worth trying out & the price is nuts - Opus 5.5 is the best Claude model in a very long time

Ben Davis

20,050 次观看 • 12 天前

Sonnet 5.5 basically Opus 5.5 but 50% cheaper and much faster. I've been early testing it and it's incredible. If this is pacing the frontier, sign me up. Demos down below 👇 3D simulation is solved:

Sonnet 5.5 basically Opus 5.5 but 50% cheaper and much faster. I've been early testing it and it's incredible. If this is pacing the frontier, sign me up. Demos down below 👇 3D simulation is solved:

Matthew Berman

281,566 次观看 • 16 天前

🚀 Claude Opus 5.5 is now available in VS Code with GitHub Copilot. Built for agentic coding, long-running tasks, and knowledge work. Give it a try today!

🚀 Claude Opus 5.5 is now available in VS Code with GitHub Copilot. Built for agentic coding, long-running tasks, and knowledge work. Give it a try today!

Visual Studio Code

76,120 次观看 • 22 天前

Opus 5.5 is the best model in the world. Full breakdown with special guest Thariq

Opus 5.5 is the best model in the world. Full breakdown with special guest Thariq

Matthew Berman

24,934 次观看 • 22 天前

Opus 5.5 beats Jev and GPT-6 Sol in 3 moves: Rubik’s Cube race We made 4 models run with same limits for all: 20 moves or 5 minutes. Every move = 1 API call. Results: Opus 5.5: solved in 1m 12.87s with the perfect 3 moves GPT-6 Sol: wrong first move, out of time after 4 moves Jev: 20 moves in 20 seconds, ran out of moves Laya: 20 moves in 9.5s, ran out of moves Opus, GPT, and Jev ran via Laya ran locally.

Opus 5.5 beats Jev and GPT-6 Sol in 3 moves: Rubik’s Cube race We made 4 models run with same limits for all: 20 moves or 5 minutes. Every move = 1 API call. Results: Opus 5.5: solved in 1m 12.87s with the perfect 3 moves GPT-6 Sol: wrong first move, out of time after 4 moves Jev: 20 moves in 20 seconds, ran out of moves Laya: 20 moves in 9.5s, ran out of moves Opus, GPT, and Jev ran via Laya ran locally.

AI/ML API

385,681 次观看 • 19 天前

Using Fable 5 to add fur and a new voice to my RL creature. The voice is based on the neural network activation (pictured top left). Made with Three.js WebGL / WebGPU Runpod trained with JAX and MuJoCo from Google DeepMind

Using Fable 5 to add fur and a new voice to my RL creature. The voice is based on the neural network activation (pictured top left). Made with Three.js WebGL / WebGPU Runpod trained with JAX and MuJoCo from Google DeepMind

Aaron Lemke

187,855 次观看 • 3 个月前

Introducing Exa Agent: frontier web research at less than half the cost of GPT 5.5 and Opus. /agent orchestrates a mixture of cost-effective models to complete any web research task, from simple data enrichments to building gigantic lists.

Introducing Exa Agent: frontier web research at less than half the cost of GPT 5.5 and Opus. /agent orchestrates a mixture of cost-effective models to complete any web research task, from simple data enrichments to building gigantic lists.

Exa

1,074,567 次观看 • 4 个月前

I asked Claude Opus 5.5 to make me an aquarium in Three.js in the style of the legendary aquascaper Takashi Amano. 3d graphics are now a solved issue. It even added Amano Shrimp!

I asked Claude Opus 5.5 to make me an aquarium in Three.js in the style of the legendary aquascaper Takashi Amano. 3d graphics are now a solved issue. It even added Amano Shrimp!

Alix Ollivier

38,735 次观看 • 19 天前

Claude Opus 5.5 + Higgsfiled built this animated windmill in Blender in 15 minutes. The modeling, rigging, texturing and animation – all made by AI.

Claude Opus 5.5 + Higgsfiled built this animated windmill in Blender in 15 minutes. The modeling, rigging, texturing and animation – all made by AI.

Higgsfield AI 🧩

52,259 次观看 • 22 天前

JEV + Opus 5.5 is insane for live design... I built a live site redesigner with JEV + Opus 5.5 Paste any link → press Start → scroll, and Jev + Opus 5.5 rebuild every section of the site in front of you Full production ship in 20 seconds: 1. IntersectionObserver fires when a section is 30%+ in the viewport 2. Jev returns one typed decision in ~0.1s: { layout, copy, drop, type, palette, p } 3. Opus 5.5 writes the component (TSX) + a CSS patch for the chosen style 4. The new section wipes in with clip-path, the old one blurs out 5. Next section enters the queue, one at a time, no race conditions Output: 8 sections of a 2015 hosting site rebuilt in ~20s, streamed line by line in the terminal 3 styles, one renderer: orthographic globe + lambert shading → ASCII / 2-color halftone / ink stipple Scroll yourself and it redesigns whatever you land on Jev decides fast, Opus designs it

JEV + Opus 5.5 is insane for live design... I built a live site redesigner with JEV + Opus 5.5 Paste any link → press Start → scroll, and Jev + Opus 5.5 rebuild every section of the site in front of you Full production ship in 20 seconds: 1. IntersectionObserver fires when a section is 30%+ in the viewport 2. Jev returns one typed decision in ~0.1s: { layout, copy, drop, type, palette, p } 3. Opus 5.5 writes the component (TSX) + a CSS patch for the chosen style 4. The new section wipes in with clip-path, the old one blurs out 5. Next section enters the queue, one at a time, no race conditions Output: 8 sections of a 2015 hosting site rebuilt in ~20s, streamed line by line in the terminal 3 styles, one renderer: orthographic globe + lambert shading → ASCII / 2-color halftone / ink stipple Scroll yourself and it redesigns whatever you land on Jev decides fast, Opus designs it

Movez

107,322 次观看 • 20 天前