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。

I will defend my PhD thesis, “Physics-Constrained Generative Models for Computational Design,” on Tuesday, September 8, 2026, at PM ET. Location: MIT MIT CSAIL 32-G449 Zoom: Committee: Kaiming He, Bill Freeman, Wojciech Matusik Generative models can now propose candidate designs at high throughput, yet the physical world remains the... show more

Minghao Guo

1,369 subscribers

79,092 次观看 • 2 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

36 条评论

Robert Scoble 的头像

Robert Scoble2 个月前

@MIT_CSAIL Your work caught my eye. Thank you and knock it out of the park!

Minghao Guo 的头像

Minghao Guo2 个月前

Relying on downstream validation creates a fundamental trade-off between throughput and physical realizability: generating a candidate may take seconds, while validating it can take hours to months. Physical realizability is a scaling problem, not just a quality problem. (2/7)

Minghao Guo 的头像

Minghao Guo2 个月前

Inspired by Marr’s separation of levels of analysis, we organize generative computational design into three levels: physical, computational, and generative. Because each level abstracts the one below, physical requirements can be weakened or lost at the interfaces between them. (3/7)

Minghao Guo 的头像

Minghao Guo2 个月前

This thesis treats physical requirements as first-class citizens of the generative system. The central idea is to carry those requirements bottom-up through explicit interfaces, from the physical world through computation and into generation. We formulate computational design as constrained optimization. (4/7)

Minghao Guo 的头像

Minghao Guo2 个月前

We realize these interfaces through three core elements: representation, constraints, and objective evaluation. At the representation interface, procedural molecular grammars, medial skeletons, and tetrahedral sphere primitives define candidate spaces that are valid by construction. At the constraint interface, differentiable projection modules enforce physical feasibility during generation. (5/7)

Minghao Guo 的头像

Minghao Guo2 个月前

At the objective interface, unified simulators evaluate design behavior with scientific fidelity at generative throughput, spanning physical regimes from molecular dynamics to aerodynamics. Together, these contributions lay the foundation for universal precision design: general-purpose systems that tailor physically realizable designs to the specific requirements of a problem, individual, or environment, at scale. (6/7)

Minghao Guo 的头像

Minghao Guo2 个月前

A research map of selected publications across my PhD study: #GenerativeAI #AIforScience #ComputationalDesign #ScientificMachineLearning #PhDDefense #worldmodel

Minghao Guo 的头像

Minghao Guo2 个月前

Missing the time haha! It’s 2-3pm ET!

중심 的头像

중심2 个月前

@MIT_CSAIL at what time?

BrenJ 的头像

BrenJ2 个月前

@MIT_CSAIL @evelovesolive @beffjezos felt like this was up y'alls alley.

DrKnowItAll 的头像

DrKnowItAll2 个月前

@MIT_CSAIL Great work! This looks really impressive.

rob 的头像

rob2 个月前

@MIT_CSAIL Great stuff here.

samir sawarkar 的头像

samir sawarkar2 个月前

@MIT_CSAIL The next leap in generative AI isn’t generating more designs. It’s generating designs that cannot violate the physics. That’s the shift from “AI can imagine it” to “AI can engineer it.”

Aman Chourasia 的头像

Aman Chourasia2 个月前

@MIT_CSAIL This is great!

Daniel Samanez 的头像

Daniel Samanez2 个月前

@MIT_CSAIL 👍 at what time is it?

Ralf Sigmund🇺🇦🇪🇺 @sistar-hh.bsky.social 的头像

Ralf Sigmund🇺🇦🇪🇺 @sistar-hh.bsky.social1 个月前

@MIT_CSAIL Congrats. It was a honor to listen to You leading us through Your work. All the best!

Xenonostra' 的头像

Xenonostra'2 个月前

@MIT_CSAIL I wish you the best of luck I'm sure you will do well

GSC 的头像

GSC2 个月前

@MIT_CSAIL

Zhiyuan 的头像

Zhiyuan2 个月前

@MIT_CSAIL solid work! may you good luck!

Yuxi Xiao 的头像

Yuxi Xiao2 个月前

@MIT_CSAIL Wow nice video for the phd thesis!!! congrats

Big data annotation 的头像

Big data annotation2 个月前

@MIT_CSAIL 大佬，牛逼

Raj Kachhadiya (KARAM) 的头像

Raj Kachhadiya (KARAM)2 个月前

@MIT_CSAIL Interesting, congratulations!

Anand Kumar 的头像

Anand Kumar2 个月前

@MIT_CSAIL Interesting. Looking forward to it.

Neil 的头像

Neil1 个月前

@MIT_CSAIL Wow this is such an interesting research direction, do you think the future of 3D is headed this way?

Yossi Eliaz 的头像

Yossi Eliaz2 个月前

@MIT_CSAIL loved the video!

likhesh 的头像

likhesh2 个月前

@MIT_CSAIL Amazing, what time exactly?

Mandar Wagh 的头像

Mandar Wagh2 个月前

congrats, and that is a committee. the split i keep getting stuck on: physics constraints are mostly differentiable, so they go straight into a loss. manufacturing constraints mostly are not — tool reach, support-free overhang, demoldability, minimum wall are combinatorial, and you cannot gradient your way to a draft angle. does the thesis treat those as one kind of constraint, or as two different problems? genuinely curious which way you landed.

the guy 的头像

the guy2 个月前

@MIT_CSAIL 🧡

Jesse Jr Lim (林振燊) 的头像

Jesse Jr Lim (林振燊)2 个月前

@MIT_CSAIL wow very cool.. paused the vid alot to see whats going on

X.M. 的头像

X.M.2 个月前

@MIT_CSAIL all the best king

Volatile Markets 的头像

Volatile Markets2 个月前

@MIT_CSAIL Love this!

Verdi 的头像

Verdi2 个月前

@MIT_CSAIL congrats!! you didnt share a time tho? > "on Tuesday, September 8, 2026, at PM ET."

Jacques Mₜ=∏ₛ((1−λₛ)+λₛEₛ) 的头像

Jacques Mₜ=∏ₛ((1−λₛ)+λₛEₛ)2 个月前

@MIT_CSAIL PhaGGOT

Curtis 的头像

Curtis1 个月前

@MIT_CSAIL Grest work, great slides, great marketing video!

Black Canvas 的头像

Black Canvas2 个月前

@MIT_CSAIL Wish you the best and interesting work!

Supratik Bhattacharya 的头像

Supratik Bhattacharya2 个月前

@MIT_CSAIL Wow! this is super interesting. all the best!

相关视频

My lab, at Karolinska in Stockholm, is looking for a PhD student with a computational/quantitative background to work on probabilistic/generative models of protein (structure and sequence). The research will involve methods development and applications in vaccine design.

My lab, at Karolinska in Stockholm, is looking for a PhD student with a computational/quantitative background to work on probabilistic/generative models of protein (structure and sequence). The research will involve methods development and applications in vaccine design.

Ben Murrell

21,646 次观看 • 1 年前

Super excited to share the last paper of my PhD: "Hallucination in World Models is Predictable and Preventable"✨ We train a 350M-param generative world model on a large dataset w/ 210 tasks and show that we can predict *when* hallucination happens and use that to fix it! 🧵1/n

Super excited to share the last paper of my PhD: "Hallucination in World Models is Predictable and Preventable"✨ We train a 350M-param generative world model on a large dataset w/ 210 tasks and show that we can predict when hallucination happens and use that to fix it! 🧵1/n

Nicklas Hansen

59,676 次观看 • 4 个月前

Excited to show some surprising inventions on generative multiplayer games we made at Google with Stanford. We call the work MultiGen. I've always been inspired by early studios like id Software with Doom or Blizzard with Warcraft bringing networked video games to the next level. We are at the point in history where we can make strides like them, but for generative games. It's a strange feeling to be in the age of generative video games while still discovering how exactly to train the models and design the tools that make them useful. All of the tools that have been invented for classic game engines need to be redesigned for generative games. For example level and world design is not entirely possible with existing technology. We introduce editable memory to diffusion game engines that allow for design of new levels via a minimap. But we can easily imagine how this can be expanded with different creation tools. The end goal of this research direction is to allow game designers to be able to guide the generation process of their world, at the granularity that they prefer. Editable memory also allows us to add multiplayer to Generative Doom. We were amazed when we saw GameNGen some years ago, and now you can play it live with friends in real-time, on your couch or even online. Shared representations like our editable memory seem like the future for this type of experience. Models are, in some cases, expensive and approximate encoders but great interpolators and extrapolators. Leveraging their strengths lets you have completely new experiences that can be realized now and not in the distant future. This work was started at my previous team and continued in collaboration with Stanford. Congratulations to all for the discoveries.

Excited to show some surprising inventions on generative multiplayer games we made at Google with Stanford. We call the work MultiGen. I've always been inspired by early studios like id Software with Doom or Blizzard with Warcraft bringing networked video games to the next level. We are at the point in history where we can make strides like them, but for generative games. It's a strange feeling to be in the age of generative video games while still discovering how exactly to train the models and design the tools that make them useful. All of the tools that have been invented for classic game engines need to be redesigned for generative games. For example level and world design is not entirely possible with existing technology. We introduce editable memory to diffusion game engines that allow for design of new levels via a minimap. But we can easily imagine how this can be expanded with different creation tools. The end goal of this research direction is to allow game designers to be able to guide the generation process of their world, at the granularity that they prefer. Editable memory also allows us to add multiplayer to Generative Doom. We were amazed when we saw GameNGen some years ago, and now you can play it live with friends in real-time, on your couch or even online. Shared representations like our editable memory seem like the future for this type of experience. Models are, in some cases, expensive and approximate encoders but great interpolators and extrapolators. Leveraging their strengths lets you have completely new experiences that can be realized now and not in the distant future. This work was started at my previous team and continued in collaboration with Stanford. Congratulations to all for the discoveries.

Nataniel Ruiz

105,085 次观看 • 8 个月前

MIT CSAIL Director Daniela Rus TED Talks: How AI will step off the screen and into the real world.

MIT CSAIL Director Daniela Rus TED Talks: How AI will step off the screen and into the real world.

MIT CSAIL

13,733 次观看 • 2 年前

Reality capture and generative AI make a powerful combination that is yet to be fully explored. Once you can perceive the world, genAI gives you the ability to reskin & augment it -- whether it's for VFX or interior design. Explore the possibilities in this compilation of my recent AI experiments. What would you create with this tech?

Reality capture and generative AI make a powerful combination that is yet to be fully explored. Once you can perceive the world, genAI gives you the ability to reskin & augment it -- whether it's for VFX or interior design. Explore the possibilities in this compilation of my recent AI experiments. What would you create with this tech?

Bilawal Sidhu

186,827 次观看 • 3 年前

He was just a janitor at MIT. One day, while mopping the floor, he glanced at a math problem on the blackboard. He had no idea it was one of the most advanced, difficult equations in the world, yet he solved it effortlessly,this gained him a lot of attention.

He was just a janitor at MIT. One day, while mopping the floor, he glanced at a math problem on the blackboard. He had no idea it was one of the most advanced, difficult equations in the world, yet he solved it effortlessly,this gained him a lot of attention.

Ola

63,466 次观看 • 7 个月前

Generative World Models running in Real-Time are one of the many incredible research efforts at Runway that excite me every single day. It will be incredibly interesting to see what new genre of games this tech will spawn.

Generative World Models running in Real-Time are one of the many incredible research efforts at Runway that excite me every single day. It will be incredibly interesting to see what new genre of games this tech will spawn.

Nicolas Neubert

19,406 次观看 • 1 个月前

Opus 4.5 (new model from Anthropic) can do something no model could do until now: It can close the design loop. This is (in my testing) the first model that can reliably do the visual loop of Generate Code ⇛ Render ⇛ Look at it ⇛ Improve. Step 3 (Visual critique) has been a real problem for so long. Models will be myopic or defensive when reviewing designs - it's been hard to tell if this has been a training or a vision problem. Opus 4.5 seems to have fixed it - it's not amazing, but it finally *can* do it. Claude models have historically (once again, in my testing) lagged behind in terms of visual understanding. Gemini for the last year has been the best - but there might be a new leader.

Opus 4.5 (new model from Anthropic) can do something no model could do until now: It can close the design loop. This is (in my testing) the first model that can reliably do the visual loop of Generate Code ⇛ Render ⇛ Look at it ⇛ Improve. Step 3 (Visual critique) has been a real problem for so long. Models will be myopic or defensive when reviewing designs - it's been hard to tell if this has been a training or a vision problem. Opus 4.5 seems to have fixed it - it's not amazing, but it finally can do it. Claude models have historically (once again, in my testing) lagged behind in terms of visual understanding. Gemini for the last year has been the best - but there might be a new leader.

Hrishi

58,331 次观看 • 11 个月前

Meta's Chief AI Scientist Yann LeCun offers a critical take on the humanoid robot boom. Speaking at MIT, LeCun claimed the "big secret" of the industry is that current companies "have no idea" how to make their robots "smart enough to be generally useful." He argues that while humanoids can be trained for narrow manufacturing tasks, a truly autonomous domestic robot is impossible without fundamental AI breakthroughs. For LeCun, this means moving beyond current generative models and toward "world model planning-type architectures"—systems that can learn to understand and predict the physical world. The future of these billion-dollar startups, he says, depends entirely on this next wave of AI research.

Meta's Chief AI Scientist Yann LeCun offers a critical take on the humanoid robot boom. Speaking at MIT, LeCun claimed the "big secret" of the industry is that current companies "have no idea" how to make their robots "smart enough to be generally useful." He argues that while humanoids can be trained for narrow manufacturing tasks, a truly autonomous domestic robot is impossible without fundamental AI breakthroughs. For LeCun, this means moving beyond current generative models and toward "world model planning-type architectures"—systems that can learn to understand and predict the physical world. The future of these billion-dollar startups, he says, depends entirely on this next wave of AI research.

Humanoids daily

1,444,029 次观看 • 1 年前

we designed + built a new website for Odyssey insane tech, they're building general purpose world models to simulate interactive worlds in real time 🤯 one of the coolest projects my team and I at The Memetic Design Company have had the chance to work on!

we designed + built a new website for Odyssey insane tech, they're building general purpose world models to simulate interactive worlds in real time 🤯 one of the coolest projects my team and I at The Memetic Design Company have had the chance to work on!

abhinav

15,544 次观看 • 8 个月前

Everything you love about generative models — now powered by real physics! Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics simulation platform designed for general-purpose robotics and physical AI applications. Genesis's physics engine is developed in pure Python, while being 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090 (see tutorial: The Genesis physics engine and simulation platform is fully open source at We'll gradually roll out access to our generative framework in the near future. Genesis implements a unified simulation framework all from scratch, integrating a wide spectrum of state-of-the-art physics solvers, allowing simulation of the whole physical world in a virtual realm with the highest realism. We aim to build a universal data engine that leverages an upper-level generative framework to autonomously create physical worlds, together with various modes of data, including environments, camera motions, robotic task proposals, reward functions, robot policies, character motions, fully interactive 3D scenes, open-world articulated assets, and more, aiming towards fully automated data generation for robotics, physical AI and other applications. Open Source Code: Project webpage: Documentation: 1/n

Everything you love about generative models — now powered by real physics! Announcing the Genesis project — after a 24-month large-scale research collaboration involving over 20 research labs — a generative physics engine able to generate 4D dynamical worlds powered by a physics simulation platform designed for general-purpose robotics and physical AI applications. Genesis's physics engine is developed in pure Python, while being 10-80x faster than existing GPU-accelerated stacks like Isaac Gym and MJX. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090 (see tutorial: The Genesis physics engine and simulation platform is fully open source at We'll gradually roll out access to our generative framework in the near future. Genesis implements a unified simulation framework all from scratch, integrating a wide spectrum of state-of-the-art physics solvers, allowing simulation of the whole physical world in a virtual realm with the highest realism. We aim to build a universal data engine that leverages an upper-level generative framework to autonomously create physical worlds, together with various modes of data, including environments, camera motions, robotic task proposals, reward functions, robot policies, character motions, fully interactive 3D scenes, open-world articulated assets, and more, aiming towards fully automated data generation for robotics, physical AI and other applications. Open Source Code: Project webpage: Documentation: 1/n

Zhou Xian

3,825,040 次观看 • 1 年前

Before AI can understand the physical world, it needs data about the physical world. Not just images. Spatial environments. Real streets, buildings, parks, stores, museums, and public spaces captured in 3D. OVER has already reconstructed 290,000+ real-world 3D scenes, creating a growing dataset for the next generation of spatial and embodied AI. And this is exactly the kind of real-world spatial data AI companies are looking for. We're already in conversations with multiple AI and Robotics companies exploring how OVER's dataset can support the development and training of models that need to understand the physical world. The next AI race will be about real-world data.

Before AI can understand the physical world, it needs data about the physical world. Not just images. Spatial environments. Real streets, buildings, parks, stores, museums, and public spaces captured in 3D. OVER has already reconstructed 290,000+ real-world 3D scenes, creating a growing dataset for the next generation of spatial and embodied AI. And this is exactly the kind of real-world spatial data AI companies are looking for. We're already in conversations with multiple AI and Robotics companies exploring how OVER's dataset can support the development and training of models that need to understand the physical world. The next AI race will be about real-world data.

Over the Reality 🌐

24,343 次观看 • 1 个月前

The fact that you can create a video like this in just a few minutes is pretty insane and magical. Anyone can place themselves into any world they want, real or imagined, instantly and soon in real time. The Moore's law of generative AI was right: the quality, resolution, and control of models doubles about every year.

The fact that you can create a video like this in just a few minutes is pretty insane and magical. Anyone can place themselves into any world they want, real or imagined, instantly and soon in real time. The Moore's law of generative AI was right: the quality, resolution, and control of models doubles about every year.

Cristóbal Valenzuela

20,713 次观看 • 6 个月前

Plaqueboymax says he plans on retiring from Streaming at 32 yrs old to start a family so his kid can take over the world 💔 "I gotta raise a young offspring so he can take over the world.. Its not going to be me to conquer the world, it will be my son"

Plaqueboymax says he plans on retiring from Streaming at 32 yrs old to start a family so his kid can take over the world 💔 "I gotta raise a young offspring so he can take over the world.. Its not going to be me to conquer the world, it will be my son"

yoxic

51,953 次观看 • 11 个月前

"I'm a shooter, I can handle the ball, and I can make my floaters." Shams Charania will be suiting up for the 2026 NBA All-Star Celebrity Game 🏀 Friday, Feb. 13 at 7 PM ET on ESPN and the ESPN App ⭐

"I'm a shooter, I can handle the ball, and I can make my floaters." Shams Charania will be suiting up for the 2026 NBA All-Star Celebrity Game 🏀 Friday, Feb. 13 at 7 PM ET on ESPN and the ESPN App ⭐

NBA on ESPN

1,385,490 次观看 • 9 个月前

Can Claude Code design proteins? We're just about to release our wet lab API + SDK so we made a quick demo to test how agents can interact with protein design models and submit the proteins to our lab for testing. This closes the loop between the computational design and the experiment validation We queued up a bunch of agent-designed proteins for wet lab validation, will release the results in a couple weeks on Proteinbase

Can Claude Code design proteins? We're just about to release our wet lab API + SDK so we made a quick demo to test how agents can interact with protein design models and submit the proteins to our lab for testing. This closes the loop between the computational design and the experiment validation We queued up a bunch of agent-designed proteins for wet lab validation, will release the results in a couple weeks on Proteinbase

Julian Englert

81,792 次观看 • 9 个月前

Very proud to showcase our latest demo of how our Engineering World Model SKY-001 can design a complex Jaguar car with 1000s of complex parts. In order to get us to Autonomous Manufacturing, we need to design systems that can think like a ‘AI General Engineer’: planning complex physical structures, long term sequences, and feasibility directly in the physical domain rather than approximating it through next-token prediction. Engineering design operates in the spatial reasoning domain through hierarchical decomposition, constraint satisfaction, and the sequential construction of complex physical objects. World models are going to be the next frontier in AI after LLMs and SKY-001 is just the beginning.

Very proud to showcase our latest demo of how our Engineering World Model SKY-001 can design a complex Jaguar car with 1000s of complex parts. In order to get us to Autonomous Manufacturing, we need to design systems that can think like a ‘AI General Engineer’: planning complex physical structures, long term sequences, and feasibility directly in the physical domain rather than approximating it through next-token prediction. Engineering design operates in the spatial reasoning domain through hierarchical decomposition, constraint satisfaction, and the sequential construction of complex physical objects. World models are going to be the next frontier in AI after LLMs and SKY-001 is just the beginning.

Sam Pasupalak

80,523 次观看 • 4 天前

Figma's Dylan Field on why design is the new code, and where the value goes as models get better: "If you believe that coding models will continue to get better, then you also believe that it's more outer loop architecture, the decisions you make about how to go build a system that matter, not the code itself." "The code is material, and then the value rises to the top and that becomes brand, it becomes product design, your marketing, your point of view in the world, and I think overall your judgment calls and strategy." "Product design is really at the center of a lot of that." "For designers, they will have a bigger role to play if they step up. Not everyone will step up. Some people will be like, 'But I liked what I was doing.'" "Overall the way you'll shape products is through design." Dylan Field w/ sophia dew on MTS

a16z

36,784 次观看 • 26 天前

Researchers at MIT developed Draft, a tool that lets engineers design and customize robots by adjusting parameters such as limb length, body size, and motor power. It uses real-world hardware data and reinforcement learning to estimate robot feasibility and test performance in simulation. The tool helps researchers explore robot designs faster without rebuilding CAD models for every change.

Researchers at MIT developed Draft, a tool that lets engineers design and customize robots by adjusting parameters such as limb length, body size, and motor power. It uses real-world hardware data and reinforcement learning to estimate robot feasibility and test performance in simulation. The tool helps researchers explore robot designs faster without rebuilding CAD models for every change.

Space and Technology

29,566 次观看 • 26 天前