Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

🏗️ Policy Adaptation from Foundation Model Feedback #CVPR2023 Instead of using foundation model as a pre-trained encoder (generator), we use it as a Teacher (discriminator) to tell where our policy did wrong and helps it adapts to new envs and tasks.

Xiaolong Wang

19,926 subscribers

24,396 views • 3 years ago •via X (Twitter)

Science & Technology Education #CVPR2023

Anya Rossi• Live Now

Private livecam show

5 Comments

Xiaolong Wang3 years ago

Work was led by Yuying @tttoaster_ when she was interning in our lab at UCSD, collaborating with @anna_macalus on working with the robots. Please check arxiv: Full video introduction:

Ted Xiao3 years ago

Nice work! Using VLMs unlocks automated feedback, which is often quite expensive to produce by humans For another style of this same method (language feedback from CLIP) but in the offline setting, check out our work DIAL:

Xiaolong Wang3 years ago

Thanks for sharing. Yes, definitely quite relevant! Very interesting idea on applying to offline setting. I think the shared core idea is we are all using VLM to provide some sort of reward signals.

Rishabh Agarwal3 years ago

Nice, looks very relevant to Reincarnating RL:

Xiaolong Wang3 years ago

Thank you for making the connection. Yes, I think we have a bit of flavor on progressively learning. The focus has been on VLM for supervision but extending more on the progressive learning direction can actually be quite an interesting ... hmm

Related Videos

Introducing Empathic Voice Interface 2 (EVI 2), our new voice-to-voice foundation model. EVI 2 merges language and voice into a single model trained specifically for emotional intelligence. You can try it and start building today.

Introducing Empathic Voice Interface 2 (EVI 2), our new voice-to-voice foundation model. EVI 2 merges language and voice into a single model trained specifically for emotional intelligence. You can try it and start building today.

Hume AI

165,599 views • 1 year ago

Can we build a generalist robotic policy that doesn’t just memorize training data and regurgitate it during test time, but instead remembers past actions as memory and conditions its decisions on them?🤖💡 Introducing SAM2Act—a multi-view robotic transformer-based policy that integrates a visual foundation model with a memory architecture for robotic manipulation. Project page: 🧵👇

Can we build a generalist robotic policy that doesn’t just memorize training data and regurgitate it during test time, but instead remembers past actions as memory and conditions its decisions on them?🤖💡 Introducing SAM2Act—a multi-view robotic transformer-based policy that integrates a visual foundation model with a memory architecture for robotic manipulation. Project page: 🧵👇

Jiafei Duan

87,573 views • 1 year ago

Cosmos Policy turns a pretrained video diffusion model into a robot controller. Instead of redesigning the architecture, it injects robot state, actions, and values directly as latent frames inside the video model

Cosmos Policy turns a pretrained video diffusion model into a robot controller. Instead of redesigning the architecture, it injects robot state, actions, and values directly as latent frames inside the video model

Robots Digest 🤖

22,933 views • 4 months ago

We trained a foundation model on 18 million heart ultrasound videos to predict structure instead of pixels. Introducing EchoJEPA, the first foundation-scale JEPA for medical video. Paper: Code: 🧵 1/n

We trained a foundation model on 18 million heart ultrasound videos to predict structure instead of pixels. Introducing EchoJEPA, the first foundation-scale JEPA for medical video. Paper: Code: 🧵 1/n

Alif Munim (d/acc)

590,210 views • 4 months ago

We release Cosmos Policy 💫: a state-of-the-art robot policy built on a video diffusion model backbone. - policy + world model + value function — in 1 model - no architectural changes to the base video model - SOTA in LIBERO (98.5%), RoboCasa (67.1%), & ALOHA tasks (93.6%) 🧵👇

We release Cosmos Policy 💫: a state-of-the-art robot policy built on a video diffusion model backbone. - policy + world model + value function — in 1 model - no architectural changes to the base video model - SOTA in LIBERO (98.5%), RoboCasa (67.1%), & ALOHA tasks (93.6%) 🧵👇

Moo Jin Kim

148,765 views • 4 months ago

👇Introducing DPPO, Diffusion Policy Policy Optimization DPPO optimizes pre-trained Diffusion Policy using policy gradient from RL, showing 𝘀𝘂𝗿𝗽𝗿𝗶𝘀𝗶𝗻𝗴 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁𝘀 over a variety of baselines across benchmarks and sim2real transfer

👇Introducing DPPO, Diffusion Policy Policy Optimization DPPO optimizes pre-trained Diffusion Policy using policy gradient from RL, showing 𝘀𝘂𝗿𝗽𝗿𝗶𝘀𝗶𝗻𝗴 𝗶𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁𝘀 over a variety of baselines across benchmarks and sim2real transfer

Allen Z. Ren

77,612 views • 1 year ago

Today we're introducing TRIBE v2 (Trimodal Brain Encoder), a foundation model trained to predict how the human brain responds to almost any sight or sound. Building on our Algonauts 2025 award-winning architecture, TRIBE v2 draws on 500+ hours of fMRI recordings from 700+ people to create a digital twin of neural activity and enable zero-shot predictions for new subjects, languages, and tasks. Try the demo and learn more here:

Today we're introducing TRIBE v2 (Trimodal Brain Encoder), a foundation model trained to predict how the human brain responds to almost any sight or sound. Building on our Algonauts 2025 award-winning architecture, TRIBE v2 draws on 500+ hours of fMRI recordings from 700+ people to create a digital twin of neural activity and enable zero-shot predictions for new subjects, languages, and tasks. Try the demo and learn more here:

AI at Meta

6,922,166 views • 2 months ago

While uniform stairs seems to be a solved problem, AGIBOT's X2 shows off how its AGILE foundation model adapts to uneven obstacles. Watch it handle a mix of 15cm, 25cm, and 20cm steps.

While uniform stairs seems to be a solved problem, AGIBOT's X2 shows off how its AGILE foundation model adapts to uneven obstacles. Watch it handle a mix of 15cm, 25cm, and 20cm steps.

Humanoids daily

20,693 views • 16 days ago

World Model meets robot policy! Robbyant's LingBot-VA: unifies video world modeling and robotic policy learning. - A single model generates both future video and the actions to make it real. - Long-term memory enables long-horizon tasks. - Claims significant outperformance over π₀.₅ in real-world tasks. - It's open-source

World Model meets robot policy! Robbyant's LingBot-VA: unifies video world modeling and robotic policy learning. - A single model generates both future video and the actions to make it real. - Long-term memory enables long-horizon tasks. - Claims significant outperformance over π₀.₅ in real-world tasks. - It's open-source

The Humanoid Hub

17,721 views • 4 months ago

John Major, "One aspect of reducing immigration would be to make it more palatable for migrants to remain in their own countries" "That is why the policy of cutting aid and investment to poor countries is a very poor sighted policy" "It will accelerate the demand from the countries to migrate as their countries get poorer" "It's a policy that acts against our own domestic interest" "It is morally wrong, politically wrong, and heartless"

John Major, "One aspect of reducing immigration would be to make it more palatable for migrants to remain in their own countries" "That is why the policy of cutting aid and investment to poor countries is a very poor sighted policy" "It will accelerate the demand from the countries to migrate as their countries get poorer" "It's a policy that acts against our own domestic interest" "It is morally wrong, politically wrong, and heartless"

Farrukh

40,331 views • 6 months ago

Yay, finally! Introducing Vision Banana🍌 from Google DeepMind, our unified model that outperforms SoTA specialist models on various vision tasks! By treating 2D/3D vision tasks as image generation, we unlock a new foundation for CV. Project page: (1/5)

Yay, finally! Introducing Vision Banana🍌 from Google DeepMind, our unified model that outperforms SoTA specialist models on various vision tasks! By treating 2D/3D vision tasks as image generation, we unlock a new foundation for CV. Project page: (1/5)

Songyou Peng

283,101 views • 1 month ago

CosmicMan A Text-to-Image Foundation Model for Humans We present CosmicMan, a text-to-image foundation model specialized for generating high-fidelity human images. Unlike current general-purpose foundation models that are stuck in the dilemma of inferior quality and

CosmicMan A Text-to-Image Foundation Model for Humans We present CosmicMan, a text-to-image foundation model specialized for generating high-fidelity human images. Unlike current general-purpose foundation models that are stuck in the dilemma of inferior quality and

AK

46,778 views • 2 years ago

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

New Course: Post-training of LLMs Learn to post-train and customize an LLM in this short course, taught by Banghua Zhu, Assistant Professor at the University of Washington University of Washington, and co-founder of @NexusflowX. Training an LLM to follow instructions or answer questions has two key stages: pre-training and post-training. In pre-training, it learns to predict the next word or token from large amounts of unlabeled text. In post-training, it learns useful behaviors such as following instructions, tool use, and reasoning. Post-training transforms a general-purpose token predictor—trained on trillions of unlabeled text tokens—into an assistant that follows instructions and performs specific tasks. Because it is much cheaper than pre-training, it is practical for many more teams to incorporate post-training methods into their workflows than pre-training. In this course, you’ll learn three common post-training methods—Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Online Reinforcement Learning (RL)—and how to use each one effectively. With SFT, you train the model on pairs of input and ideal output responses. With DPO, you provide both a preferred (chosen) and a less preferred (rejected) response and train the model to favor the preferred output. With RL, the model generates an output, receives a reward score based on human or automated feedback, and updates the model to improve performance. You’ll learn the basic concepts, common use cases, and principles for curating high-quality data for effective training. Through hands-on labs, you’ll download a pre-trained model from Hugging Face and post-train it using SFT, DPO, and RL to see how each technique shapes model behavior. In detail, you’ll: - Understand what post-training is, when to use it, and how it differs from pre-training. - Build an SFT pipeline to turn a base model into an instruct model. - Explore how DPO reshapes behavior by minimizing contrastive loss—penalizing poor responses and reinforcing preferred ones. - Implement a DPO pipeline to change the identity of a chat assistant. - Learn online RL methods such as Proximal Policy Optimization (PPO) and Group Relative Policy Optimization (GRPO), and how to design reward functions. - Train a model with GRPO to improve its math capabilities using a verifiable reward. Post-training is one of the most rapidly developing areas of LLM training. Whether you’re building a high-accuracy context-specific assistant, fine-tuning a model's tone, or improving task-specific accuracy, this course will give you experience with the most important techniques shaping how LLMs are post-trained today. Please sign up here:

Andrew Ng

125,146 views • 11 months ago

Before we ship a new model, these teams try to break it. They build with it, push it to its limits, and tell us where it falls short. What they find makes the final model better.

Before we ship a new model, these teams try to break it. They build with it, push it to its limits, and tell us where it falls short. What they find makes the final model better.

Claude

571,498 views • 16 days ago

Eden is building toward a future where physical services are as scalable as software and delivered autonomously, end-to-end, by intelligent agents. Our foundation model Theta is the first step toward that future.

Eden is building toward a future where physical services are as scalable as software and delivered autonomously, end-to-end, by intelligent agents. Our foundation model Theta is the first step toward that future.

Stamatis Floratos

342,723 views • 17 days ago

🧐Applying world models to improve real-world policy on challenging manipulation tasks used to be considered out of reach. 😌After sustained effort, we’re now seeing encouraging progress. 🚀Thrilled to introduce RISE: Self-Improving Robot Policy with Compositional World Model RISE is, to our knowledge, the first work to use a world model as an effective learning environment for challenging real-world manipulation, enabling policy improvement on tasks that demand high dynamics, dexterity, and precision. Incredible teamwork with Kunyang Lin Hongyang Li Yue Xiangyu Hao Zhao Chonghao Sima

🧐Applying world models to improve real-world policy on challenging manipulation tasks used to be considered out of reach. 😌After sustained effort, we’re now seeing encouraging progress. 🚀Thrilled to introduce RISE: Self-Improving Robot Policy with Compositional World Model RISE is, to our knowledge, the first work to use a world model as an effective learning environment for challenging real-world manipulation, enabling policy improvement on tasks that demand high dynamics, dexterity, and precision. Incredible teamwork with Kunyang Lin Hongyang Li Yue Xiangyu Hao Zhao Chonghao Sima

Jiazhi Yang

75,629 views • 4 months ago

Our Robotic Foundation Model, trained on data from real-world warehouse operations and simulated scenarios, enables our robots to identify objects, understand 3D spaces, and predict optimal grasping and placement.

Our Robotic Foundation Model, trained on data from real-world warehouse operations and simulated scenarios, enables our robots to identify objects, understand 3D spaces, and predict optimal grasping and placement.

Covariant

33,508 views • 1 year ago

Humans learn and improve from failures. Similarly, foundation models adapt based on human feedback. Can we leverage this failure understanding to enhance robotics systems that use foundation models? Introducing AHA—a vision-language model for detecting and reasoning over failures in robotic manipulation. Project page: 🧵Thread👇 Aha!

Humans learn and improve from failures. Similarly, foundation models adapt based on human feedback. Can we leverage this failure understanding to enhance robotics systems that use foundation models? Introducing AHA—a vision-language model for detecting and reasoning over failures in robotic manipulation. Project page: 🧵Thread👇 Aha!

Jiafei Duan

48,739 views • 1 year ago

1/ NitroGen: NVIDIA's new image-to-action model! NitroGen, a vision-action foundation model for generalist gaming agents that is trained on 40,000 hours of gameplay videos across more than 1,000 games. Gaming is a significant factor in AI training. Google DeepMind trained AI early on Starcraft 2, and OpenAI on Dota 2. This new product from NVIDIA is therefore extremely important. Why it matters and how it works:

1/ NitroGen: NVIDIA's new image-to-action model! NitroGen, a vision-action foundation model for generalist gaming agents that is trained on 40,000 hours of gameplay videos across more than 1,000 games. Gaming is a significant factor in AI training. Google DeepMind trained AI early on Starcraft 2, and OpenAI on Dota 2. This new product from NVIDIA is therefore extremely important. Why it matters and how it works:

Chubby♨️

55,280 views • 5 months ago