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.

🤖Robots that think ahead and act in real time. LingBot-VA 2.0 — the first embodied-native foundation model. Not fine-tuned from a video generator. Built from scratch for the physical world. ✅ 93.6% success on bimanual tasks ⚡ 150 Hz single-GPU inference 🎯 20 demos to generalize. This is what... show more

Robbyant

3,309 subscribers

895,656 views • 3 months ago •via X (Twitter)

Science & Technology Education

Anya Rossi• Live Now

Private livecam show

35 Comments

Robbyant3 months ago

🚫 Most robot foundation models today? Repurposed video generators — bidirectional, built for pretty frames, not physical causality. Fine-tune them for control and you get forgetting, drift, and brittle transfer. 🔧 LingBot-VA 2.0 is pretrained from scratch, strictly causal, purely forward in time. → Semantic vision-action tokenizer → Causal DiT + sparse MoE backbone → Foresight Reasoning — predicts the next world state while the robot is still executing One model. Imagines the future AND outputs the action to make it happen.

Robbyant3 months ago

📊 RoboTwin 2.0 (bimanual sim benchmark): 🥇 LingBot-VA 2.0: 93.6% Robust across clean & domain-randomized settings — doesn't break when the world changes. 💪 🏆 Real robot: 20 demos → cross-task generalist policy In-context learning: reference video in, zero parameter updates, new task out 🪄

Robbyant3 months ago

🧱 LingBot-VA 2.0 is the capstone of our launch week. This week we shipped 6 models: 🔹 LingBot-Depth 2.0 🔹 LingBot-Vision 🔹 LingBot-VLA 2.0 🔹 LingBot-World 2.0 🔹 LingBot-Video 🔹 LingBot-VA 2.0 Together they form our Embodied-Native Full-Stack — perception, world simulation, prediction, and action generation in one closed loop. One brain, many robots. 🌐 Project Page: 📄 Paper: The era of embodied-native AI starts now. Let's build it together. 🚀

ZenO3 months ago

More innovation like this is always welcome

Kaitee3 months ago

The semantic visual-action tokenizer is probably the architectural detail I'm most interested in here.

Shruti3 months ago

Every time I open X there's a new robotics breakthrough. We're moving so fast it's hard to keep up. The future is going to be fun.

Orikan3 months ago

Native video-action pretraining feels like the right foundation for embodied intelligence.

MR LARK DAVIS 🚀📊3 months ago

The semantic visual-action tokenizer stands out as one of the most impactful architectural innovations, bringing perception and action closer together for robotics

Leila 👩🏻‍💻3 months ago

over 90% of success is crazyyyyy

AI Mastery Guide3 months ago

93.6% success on bimanual tasks from just 20 demos is honestly impressive, native training over fine tuning really shows here

Ebokify2 months ago

Robot Modeling and Control

Vaidehi3 months ago

Nice to see a model built specifically for embodied intelligence.

Anissa3 months ago

Robotics is entering a new era.

Wesley3 months ago

Love seeing models purpose-built for physical causality from day one. Predicting future states while acting in real time feels like the missing piece for robust robot behavior. The few-shot generalization is especially promising for messy real-world use. 🚀

Leonardo3 months ago

LingBot-VA 2.0 feels intentionally built around the demands of embodied intelligence.

Marco | IA3 months ago

Each new generation of models for robotics shows that progress goes far beyond creating content. Seeing AI that perceives, plans and acts in real time is really exciting.

OriginAI3 months ago

Purpose-built video-action foundation models feel like the next logical step for scalable embodied intelligence.

Arti Shah3 months ago

Few-shot adaptation and in-context learning could make robots much more adaptable.

Zoha | Ai creator3 months ago

Really helpful

Nico3 months ago

this is honestly an insane achivement

Manoj Kumar Shah3 months ago

Purpose-built robot foundation models are becoming increasingly important as the field matures.

Dylan Knox3 months ago

Thinking ahead and acting in real time is the future of embodied AI.

Theo Builds AI3 months ago

Generalization is where robotics gets interesting.

SynapseOps AI3 months ago

The asynchronous pipeline is worth reading about.

gus3 months ago

no way it keeps getting crazier

Shruti Codes3 months ago

Native video-action pretraining is a really interesting direction.

Sani Ai Tech3 months ago

Impressive breakthrough with real world robotics from the ground up

Enzo Sanchez | IA3 months ago

La generalización será una de las claves

MR ANDERSON3 months ago

Embodied AI is entering a whole new era. Building natively for the physical world is the real breakthrough. 🤖

Tanmoy Debnath3 months ago

LingBot's focus on long-term consistency could make interactive environments much more reliable.

arc.3 months ago

this drop is crazy

Mark3 months ago

Amazing share

codemarch3 months ago

Foresight Reasoning feels well suited for real-time closed-loop control.

舔狗AI实验室 🛠️🐶3 months ago

20个演示样本训练出来的模型,泛化能力怎么保证啊

FELIX3 months ago

93.6% success on bimanual tasks is an incredible benchmark for native models.

Related Videos

🚀 MiniCPM enters the physical world — enabling robots to understand, remember, and act. We open-source MiniCPM-Robot, our first embodied AI model series, including: 🤖 MiniCPM-RobotManip — a 1.5B general-purpose Vision-Language-Action (VLA) model for robotic manipulation. 🐕 MiniCPM-RobotTrack — a compact model for real-world target tracking. ⚡ PhyAI — a high-performance inference framework built for embodied models. Together, they bring efficient, practical, and open embodied intelligence closer to real-world robots. ⭐ GitHub: 🤗 MiniCPM-RobotManip: 🤗 MiniCPM-RobotTrack:

🚀 MiniCPM enters the physical world — enabling robots to understand, remember, and act. We open-source MiniCPM-Robot, our first embodied AI model series, including: 🤖 MiniCPM-RobotManip — a 1.5B general-purpose Vision-Language-Action (VLA) model for robotic manipulation. 🐕 MiniCPM-RobotTrack — a compact model for real-world target tracking. ⚡ PhyAI — a high-performance inference framework built for embodied models. Together, they bring efficient, practical, and open embodied intelligence closer to real-world robots. ⭐ GitHub: 🤗 MiniCPM-RobotManip: 🤗 MiniCPM-RobotTrack:

OpenBMB

331,849 views • 2 months ago

Demis Hassabis says an “AlphaFold moment” for the physical world could be close—real robots doing useful tasks reliably in the real world. Gemini was built multimodal so a future assistant can understand what you see on a phone or glasses. That same “understand the world” ability is a direct bridge into robotics, says Demis Hassabis. Physical intelligence becomes real when robots stop being demos and start being dependable helpers.

Demis Hassabis says an “AlphaFold moment” for the physical world could be close—real robots doing useful tasks reliably in the real world. Gemini was built multimodal so a future assistant can understand what you see on a phone or glasses. That same “understand the world” ability is a direct bridge into robotics, says Demis Hassabis. Physical intelligence becomes real when robots stop being demos and start being dependable helpers.

Wes Roth

21,658 views • 8 months ago

🚀 We’re excited to announce LingBot-VA, a new state-of-the-art robot policy model from Robbyant ! LingBot-VA is built on a causal, autoregressive video-action world model for generalist robot control. Highlights: (1) First unified autoregressive video-action world model for robot control (2) Low-latency inference with a new asynchronous execution pipeline (3) SOTA on RoboTwin (92.9%, firstever > 90%) and LIBERO (98.5%) (4) +20% over π0.5 on challenging real-world long-horizon & high-precision tasks

🚀 We’re excited to announce LingBot-VA, a new state-of-the-art robot policy model from Robbyant ! LingBot-VA is built on a causal, autoregressive video-action world model for generalist robot control. Highlights: (1) First unified autoregressive video-action world model for robot control (2) Low-latency inference with a new asynchronous execution pipeline (3) SOTA on RoboTwin (92.9%, firstever > 90%) and LIBERO (98.5%) (4) +20% over π0.5 on challenging real-world long-horizon & high-precision tasks

Yinghao Xu

47,604 views • 8 months ago

🌍 What if an AI-generated world didn’t end after a few seconds—but kept running for hours, responded to every move, let you attack, cast spells, shoot, or summon storms, and continued evolving through AI agents? That’s LingBot-World 2.0: our open-source real-time interactive world model, scaling to 720p/60fps in our full setup. Today, we’re open-sourcing 3 more models: 🔹 LingBot-World 2.0 Small (1.3B) 🔹 LingBot-World 2.0 Bidirectional 🔹 LingBot-World 2.0 Causal Pretrain And now, with our 1.3B Small Model, that world can run in real time on a single consumer GPU. 🧵

🌍 What if an AI-generated world didn’t end after a few seconds—but kept running for hours, responded to every move, let you attack, cast spells, shoot, or summon storms, and continued evolving through AI agents? That’s LingBot-World 2.0: our open-source real-time interactive world model, scaling to 720p/60fps in our full setup. Today, we’re open-sourcing 3 more models: 🔹 LingBot-World 2.0 Small (1.3B) 🔹 LingBot-World 2.0 Bidirectional 🔹 LingBot-World 2.0 Causal Pretrain And now, with our 1.3B Small Model, that world can run in real time on a single consumer GPU. 🧵

Robbyant

1,998,545 views • 1 month ago

🦾 From seeing to doing. We're closing the loop between video prediction and real-world action. On the final day of Robbyant Open Source Week, we bring you LingBot-VA—the world's first causal video-action world model for generalist robot control. 🔥 Key Highlights: 🤖 Predicts & Acts: A single model generates both future video and the actions to make it real. 🧠 Remembers the Past: True long-term memory for complex, sequential tasks. ⚡ Learns in a Snap: Masters new skills with just 30-50 real-world demos. The result? 📈 SOTA on RoboTwin (92.9%) and LIBERO (98.5%), +20% over π0.5 on challenging real-world long-horizon & high-precision tasks. More below 👇 #AI #Robotics #EmbodiedAI #WorldModel #OpenSource

🦾 From seeing to doing. We're closing the loop between video prediction and real-world action. On the final day of Robbyant Open Source Week, we bring you LingBot-VA—the world's first causal video-action world model for generalist robot control. 🔥 Key Highlights: 🤖 Predicts & Acts: A single model generates both future video and the actions to make it real. 🧠 Remembers the Past: True long-term memory for complex, sequential tasks. ⚡ Learns in a Snap: Masters new skills with just 30-50 real-world demos. The result? 📈 SOTA on RoboTwin (92.9%) and LIBERO (98.5%), +20% over π0.5 on challenging real-world long-horizon & high-precision tasks. More below 👇 #AI #Robotics #EmbodiedAI #WorldModel #OpenSource

Robbyant

704,157 views • 8 months ago

Robots are getting smarter, but most still fail the same way. They don’t learn from their own mistakes. A new paper proposes something different: a way for robots to self-improve directly from their failures in the real world. It’s called PLD (Probe, Learn, Distill). The idea: instead of collecting endless human demos, let the robot figure out where it fails, learn how to recover, and then distill that knowledge back into its main model. Key takeaways from the research: ✅ Uses residual reinforcement learning to recover from policy failures ✅ Achieves 99% success on LIBERO and 100% on real Franka and YAM arms ✅ Runs hour-long manipulation tasks without human resets ✅ Builds a feedback loop between real-world data and model adaptation Unlike supervised fine-tuning, which relies on humans, PLD learns from the robot’s own experience. By training on its own failure distribution, the model becomes both more efficient and more aligned with the real world. It’s not just a technical shift, it’s a step toward robots that improve themselves through real-world practice. Thanks for sharing, Wenli Xiao !! Paper and demos: —- Weekly robotics and AI insights. Subscribe free:

Robots are getting smarter, but most still fail the same way. They don’t learn from their own mistakes. A new paper proposes something different: a way for robots to self-improve directly from their failures in the real world. It’s called PLD (Probe, Learn, Distill). The idea: instead of collecting endless human demos, let the robot figure out where it fails, learn how to recover, and then distill that knowledge back into its main model. Key takeaways from the research: ✅ Uses residual reinforcement learning to recover from policy failures ✅ Achieves 99% success on LIBERO and 100% on real Franka and YAM arms ✅ Runs hour-long manipulation tasks without human resets ✅ Builds a feedback loop between real-world data and model adaptation Unlike supervised fine-tuning, which relies on humans, PLD learns from the robot’s own experience. By training on its own failure distribution, the model becomes both more efficient and more aligned with the real world. It’s not just a technical shift, it’s a step toward robots that improve themselves through real-world practice. Thanks for sharing, Wenli Xiao !! Paper and demos: —- Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

22,891 views • 11 months ago

Are you kidding me??? It grasps multiple objects with different ways, all at once with… a single hand??? No pauses. 1x speed. GENE-26.5 is Genesis AI’s robotics-native multimodal foundation model. It’s trained on 200,000+ hours of real human hand data (motion, force, touch) and runs on a 54-DoF bimanual system: Scaling that human data 4x lifted real-robot success rates from 16.6% to 65.6% on long-horizon dexterous tasks!! Same model weights, zero fine-tuning for this exact sequence. You know those tiny coordinated movements you do without thinking…? Robots couldn’t reliably do that before. Now they can. Today. This is the video you’ll send to my friends, outside of our bubble, when they say “robots are still just demos.” Congrats to the entire team around Zhou Xian! Credit: Seen at Zu Wang (Zu) Genesis official announcement for the full story + longer demo: ——— Weekly robotics and AI insights. Subscribe free:

Are you kidding me??? It grasps multiple objects with different ways, all at once with… a single hand??? No pauses. 1x speed. GENE-26.5 is Genesis AI’s robotics-native multimodal foundation model. It’s trained on 200,000+ hours of real human hand data (motion, force, touch) and runs on a 54-DoF bimanual system: Scaling that human data 4x lifted real-robot success rates from 16.6% to 65.6% on long-horizon dexterous tasks!! Same model weights, zero fine-tuning for this exact sequence. You know those tiny coordinated movements you do without thinking…? Robots couldn’t reliably do that before. Now they can. Today. This is the video you’ll send to my friends, outside of our bubble, when they say “robots are still just demos.” Congrats to the entire team around Zhou Xian! Credit: Seen at Zu Wang (Zu) Genesis official announcement for the full story + longer demo: ——— Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

204,737 views • 5 months ago

A team tested Pi0, Pi0 Fast, Gr00t, and ACT on real robot arms in manufacturing tasks. (🔖 Bookmark this for later!) The task was precise: place thin rectangular frames from a messy stack into a holder. The team fine-tuned each model on 100 real trajectories and compared training time, inference speed, motion quality, and success rates. ⬇️ Here’s a breakdown of what they found Pi0 (Original) ✅ Strongest overall performance in precise pick-and-place ✅ High success rate even in edge cases ✅ Longest training time (~11 hours, ~$30 per run) ✅ Inference time of 80 ms causes short pauses between actions Despite delays, it handles complex scenarios well… solid for high-precision tasks, but slow to train. Gr00t ✅ Trains fast (~2 hours, ~$5 per run) ✅ Performs almost as well as Pi0 on large-object tasks ✅ Struggles with fine precision; random movement in some trials ✅ More training didn’t fix jitter or random offsets Best suited for tasks where exact precision isn’t critical. Not ready for manufacturing-grade accuracy without more tuning. Pi0 Fast ✅ Promised faster training, but results were underwhelming ✅ Training at 6 hours still showed low success rates ✅ Inference was slower than expected ✅ Not reliable for generalizing even slightly new tasks Currently too unstable for real-world deployment. Doesn’t live up to the “Fast” name yet. ACT (Baseline) ✅ 200MB model—lightweight, but limited ✅ Struggles with stacked objects or ambiguous scenes ✅ Success rates around 70% in best-case setups ✅ Can’t match newer models on precision or generalization Still a solid baseline, but clearly a generation behind in robustness. 🚨 Extra Notes All newer models share a common issue: •Inference takes longer than a frame (80 ms vs 33 ms), so robots “pause” between chunks. •This results in jittery movements, but not a dealbreaker unless tasks are time-sensitive. Language-conditioned tasks also fell short: after training on two labeled tasks, the model couldn’t generalize to a third unseen combination using only text prompts. ✅ The good news? These models adapt well to new robot arms with quick fine-tuning. ❌ The bad news? There’s still no plug-and-play solution for improving performance after deployment. Reinforcement learning or DAgger-style data collection during real-world operation may be the next big step, something many teams in robotics are actively working on.

A team tested Pi0, Pi0 Fast, Gr00t, and ACT on real robot arms in manufacturing tasks. (🔖 Bookmark this for later!) The task was precise: place thin rectangular frames from a messy stack into a holder. The team fine-tuned each model on 100 real trajectories and compared training time, inference speed, motion quality, and success rates. ⬇️ Here’s a breakdown of what they found Pi0 (Original) ✅ Strongest overall performance in precise pick-and-place ✅ High success rate even in edge cases ✅ Longest training time (~11 hours, ~$30 per run) ✅ Inference time of 80 ms causes short pauses between actions Despite delays, it handles complex scenarios well… solid for high-precision tasks, but slow to train. Gr00t ✅ Trains fast (~2 hours, ~$5 per run) ✅ Performs almost as well as Pi0 on large-object tasks ✅ Struggles with fine precision; random movement in some trials ✅ More training didn’t fix jitter or random offsets Best suited for tasks where exact precision isn’t critical. Not ready for manufacturing-grade accuracy without more tuning. Pi0 Fast ✅ Promised faster training, but results were underwhelming ✅ Training at 6 hours still showed low success rates ✅ Inference was slower than expected ✅ Not reliable for generalizing even slightly new tasks Currently too unstable for real-world deployment. Doesn’t live up to the “Fast” name yet. ACT (Baseline) ✅ 200MB model—lightweight, but limited ✅ Struggles with stacked objects or ambiguous scenes ✅ Success rates around 70% in best-case setups ✅ Can’t match newer models on precision or generalization Still a solid baseline, but clearly a generation behind in robustness. 🚨 Extra Notes All newer models share a common issue: •Inference takes longer than a frame (80 ms vs 33 ms), so robots “pause” between chunks. •This results in jittery movements, but not a dealbreaker unless tasks are time-sensitive. Language-conditioned tasks also fell short: after training on two labeled tasks, the model couldn’t generalize to a third unseen combination using only text prompts. ✅ The good news? These models adapt well to new robot arms with quick fine-tuning. ❌ The bad news? There’s still no plug-and-play solution for improving performance after deployment. Reinforcement learning or DAgger-style data collection during real-world operation may be the next big step, something many teams in robotics are actively working on.

Ilir Aliu

21,844 views • 1 year ago

Breaking news: Cosmos 3 is here. They are attempting to do something completely new 🤯 Why is Physical AI much harder than building a chatbot? Understanding the world is not enough, robots need to predict it and act inside it. That's the idea behind NVIDIA Cosmos 3: → Reasoning model understands what's happening from video, images, text, and actions. → World model generates future states of the environment. → Action model generates the actions needed to achieve a goal. Previous systems often stitched these capabilities together using separate models. Cosmos 3 combines them into a single architecture with two components: ▪️ Reasoner Tower Analyzes observations and builds an understanding of objects, motion, interactions, and physical context. ▪️ Generator Tower Uses that understanding to generate future videos and action sequences that obey physical constraints. So Cosmos 3 moves from: Perception → Model A Prediction → Model B Actions → Model C to: Perception + Prediction + Actions → One unified system. The goal is to make robots, autonomous vehicles, and smart environments better at answering three questions: 1. What is happening? 2. What will happen next? 3. What should I do? That's a big shift from today's AI, which mostly focuses on generating text. And check the benchmarks! Physical AI needs to generate decisions that survive contact with the real world. 🚗🤖

Breaking news: Cosmos 3 is here. They are attempting to do something completely new 🤯 Why is Physical AI much harder than building a chatbot? Understanding the world is not enough, robots need to predict it and act inside it. That's the idea behind NVIDIA Cosmos 3: → Reasoning model understands what's happening from video, images, text, and actions. → World model generates future states of the environment. → Action model generates the actions needed to achieve a goal. Previous systems often stitched these capabilities together using separate models. Cosmos 3 combines them into a single architecture with two components: ▪️ Reasoner Tower Analyzes observations and builds an understanding of objects, motion, interactions, and physical context. ▪️ Generator Tower Uses that understanding to generate future videos and action sequences that obey physical constraints. So Cosmos 3 moves from: Perception → Model A Prediction → Model B Actions → Model C to: Perception + Prediction + Actions → One unified system. The goal is to make robots, autonomous vehicles, and smart environments better at answering three questions: 1. What is happening? 2. What will happen next? 3. What should I do? That's a big shift from today's AI, which mostly focuses on generating text. And check the benchmarks! Physical AI needs to generate decisions that survive contact with the real world. 🚗🤖

Turing Post

13,836 views • 4 months ago

Most AI is built to replace humans. We're building AI to understand them. Introducing Atypica's Subjective World Model. A foundation for understanding how people express themselves, tell their stories, make decisions, and behave under real-world constraints. Not just what users say — but why they choose, hesitate, and act. It powers AI Research, AI Personas, AI Panels, and AI Interviews, helping teams turn user research from a quarterly project into a permanent capability. Watch the video to see how it works and what teams are already building with Atypica. Start to explore this AI-native consumer research platform from here:

Most AI is built to replace humans. We're building AI to understand them. Introducing Atypica's Subjective World Model. A foundation for understanding how people express themselves, tell their stories, make decisions, and behave under real-world constraints. Not just what users say — but why they choose, hesitate, and act. It powers AI Research, AI Personas, AI Panels, and AI Interviews, helping teams turn user research from a quarterly project into a permanent capability. Watch the video to see how it works and what teams are already building with Atypica. Start to explore this AI-native consumer research platform from here:

atypica.AI

28,221 views • 2 months ago

Weekends are for research. Inside OptimAI labs, we’ve been pushing agents beyond the screen. A first glimpse of OptimAI Prime. Not just agents that think, but agents that operate. Learning through reinforcement, coordinating across the network, and executing in real-world environments. From digital workflows → to physical actions. This is the next step for the agent-native internet.🤖 Still early. But already in motion.

Weekends are for research. Inside OptimAI labs, we’ve been pushing agents beyond the screen. A first glimpse of OptimAI Prime. Not just agents that think, but agents that operate. Learning through reinforcement, coordinating across the network, and executing in real-world environments. From digital workflows → to physical actions. This is the next step for the agent-native internet.🤖 Still early. But already in motion.

OptimAI Network

20,885 views • 5 months ago

The robot’s “eyes” just received a big upgrade. LingBot-Depth 2.0, a depth-completion model with half the depth error just dropped. 12/16 benchmarks topped. Glass, mirrors, and transparent objects are so easy for us humans, but so hard for robots, because they do not behave like ordinary surfaces in a camera pipeline. A robot that misunderstands a balcony window or a table edge, will have a completely false planning inside a false world. Huge implecation. LingBot-Depth 2.0 takes an RGB image plus a broken depth map from a sensor and then outputs a cleaner depth map and a usable 3D point cloud. Numbers on LingBot-Depth 2.0 - Excels on glass, mirrors & transparent objects — where traditional depth cameras fail - Training data: 3M → 150M (50x scale-up) - 12 out of 16 first-place rankings on depth completion benchmarks - RMSE cut in half: 0.132 → 0.062 on the hardest indoor scenes LingBot-Vision trained on boundaries, because object edges carry the geometry robots need. No human boundary labels are used, which makes this approach easier to scale. The open-sourced LingBot-Vision is the general vision backbone, and LingBot-Depth 2.0 is the depth model built on it.

The robot’s “eyes” just received a big upgrade. LingBot-Depth 2.0, a depth-completion model with half the depth error just dropped. 12/16 benchmarks topped. Glass, mirrors, and transparent objects are so easy for us humans, but so hard for robots, because they do not behave like ordinary surfaces in a camera pipeline. A robot that misunderstands a balcony window or a table edge, will have a completely false planning inside a false world. Huge implecation. LingBot-Depth 2.0 takes an RGB image plus a broken depth map from a sensor and then outputs a cleaner depth map and a usable 3D point cloud. Numbers on LingBot-Depth 2.0 - Excels on glass, mirrors & transparent objects — where traditional depth cameras fail - Training data: 3M → 150M (50x scale-up) - 12 out of 16 first-place rankings on depth completion benchmarks - RMSE cut in half: 0.132 → 0.062 on the hardest indoor scenes LingBot-Vision trained on boundaries, because object edges carry the geometry robots need. No human boundary labels are used, which makes this approach easier to scale. The open-sourced LingBot-Vision is the general vision backbone, and LingBot-Depth 2.0 is the depth model built on it.

Rohan Paul

25,312 views • 3 months ago

VLA-JEPA just dropped in LeRobot 🤖 What makes this model special is that it does not just learn what action to take from a given observation, it also leverages a JEPA world model to learn action-relevant dynamics. During training, the VLA leverages V-JEPA2 by conditioning its predictor. This clever trick adds a world modeling objective to the training, which also allows pretraining on human videos. At inference, the world model is dropped entirely, keeping only a standard VLA architecture: Qwen backbone and action head. The demo here was only fine-tuned on 13 examples, showing great pretraining capability and running in real time on NVIDIA Robotics DGX Spark! VLA-JEPA is the first world model to be ported to LeRobot, and I feel like it won't be the last 🚀 Thomas Wolf clem 🤗

VLA-JEPA just dropped in LeRobot 🤖 What makes this model special is that it does not just learn what action to take from a given observation, it also leverages a JEPA world model to learn action-relevant dynamics. During training, the VLA leverages V-JEPA2 by conditioning its predictor. This clever trick adds a world modeling objective to the training, which also allows pretraining on human videos. At inference, the world model is dropped entirely, keeping only a standard VLA architecture: Qwen backbone and action head. The demo here was only fine-tuned on 13 examples, showing great pretraining capability and running in real time on NVIDIA Robotics DGX Spark! VLA-JEPA is the first world model to be ported to LeRobot, and I feel like it won't be the last 🚀 Thomas Wolf clem 🤗

LeRobot

321,209 views • 4 months ago

Today’s the day🚀 We’re opening the doors to the Daydream platform - API + Playground Experience🌊 We’re the world’s first hosted StreamDiffusion platform: 1⃣ No GPU required 2⃣Sub-second latency 3⃣Advanced controls that let you experiment quickly and scale seamlessly. This is the start of an open ecosystem for real-time video and world models. Your home for open source, real-time video AI 👇🧵

Today’s the day🚀 We’re opening the doors to the Daydream platform - API + Playground Experience🌊 We’re the world’s first hosted StreamDiffusion platform: 1⃣ No GPU required 2⃣Sub-second latency 3⃣Advanced controls that let you experiment quickly and scale seamlessly. This is the start of an open ecosystem for real-time video and world models. Your home for open source, real-time video AI 👇🧵

Daydream

94,889 views • 1 year ago

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Santiago

84,747 views • 2 years ago

Robots might learn better from video than from language! 📼 Most Vision-Language-Action (VLA) models learn what to do from text, but still struggle with how things move in the real world. That makes them data-hungry and slow to train. mimic video takes a different route. Instead of grounding robot control in text, it grounds it in video, using large pre-trained video models that already capture physical motion and dynamics. The idea is straightforward: let the video model handle “what will happen next,” and let a smaller control model focus only on turning that visual plan into robot actions. The result is big gains in practice. Robots trained this way need 10× less data, converge twice as fast, and perform better on both simulated benchmarks and real bimanual manipulation tasks. If robots can “imagine” motion using video, control becomes a much simpler problem. Shoutout to Jonas Pai, Liam Achenbach, Oier Mees, Elvis Nava and the rest of the team! Here's the project page: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Robots might learn better from video than from language! 📼 Most Vision-Language-Action (VLA) models learn what to do from text, but still struggle with how things move in the real world. That makes them data-hungry and slow to train. mimic video takes a different route. Instead of grounding robot control in text, it grounds it in video, using large pre-trained video models that already capture physical motion and dynamics. The idea is straightforward: let the video model handle “what will happen next,” and let a smaller control model focus only on turning that visual plan into robot actions. The result is big gains in practice. Robots trained this way need 10× less data, converge twice as fast, and perform better on both simulated benchmarks and real bimanual manipulation tasks. If robots can “imagine” motion using video, control becomes a much simpler problem. Shoutout to Jonas Pai, Liam Achenbach, Oier Mees, Elvis Nava and the rest of the team! Here's the project page: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →

Lukas Ziegler

50,130 views • 9 months ago

This World Model 'LingBot-World-Infinity (LingBot-World 2.0)' just released from Ant Group looks realy promising. It is an open causal world model with an Agentic harness. Most interactive world models hold together for a few minutes. Then textures smear and geometry warps. That's a video model, not a world — and Robbyant just drew the line at the attention mask. They released LingBot-World-Infinity (LingBot-World 2.0) — a 14B open causal video world model built on Wan2.2, trained with a Mixture of Bidirectional and Autoregressive (MoBA) attention mask, then distilled into a few-step real-time generator with no post-hoc drift filtering anywhere in the stack. Here's what's actually interesting: → Pure teacher forcing overfits — as context grows, the model leans on context instead of predicting frames. MoBA appends a bidirectional full-attention block as a regularizer → Leak-free cross-attention: AR rows attend to background prompt a_B plus chunk prompts a_≤i, lower-triangular. Bidirectional rows see one global prompt a_G → DMD runs over long self-rollout trajectories, not teacher-forced states — the student is optimized on the distribution its own errors induce → Director-Pilot harness: a VLM proposes event cards, the DiT generator renders physical dynamics. Mode B adds a SAM tracking loop for object-centric interaction → One 60-minute uninterrupted session, 20 distinct scenarios, no perceptible decay Full analysis: Paper: Model weight: GitHub Repo: Project: Ant Group Robbyant

This World Model 'LingBot-World-Infinity (LingBot-World 2.0)' just released from Ant Group looks realy promising. It is an open causal world model with an Agentic harness. Most interactive world models hold together for a few minutes. Then textures smear and geometry warps. That's a video model, not a world — and Robbyant just drew the line at the attention mask. They released LingBot-World-Infinity (LingBot-World 2.0) — a 14B open causal video world model built on Wan2.2, trained with a Mixture of Bidirectional and Autoregressive (MoBA) attention mask, then distilled into a few-step real-time generator with no post-hoc drift filtering anywhere in the stack. Here's what's actually interesting: → Pure teacher forcing overfits — as context grows, the model leans on context instead of predicting frames. MoBA appends a bidirectional full-attention block as a regularizer → Leak-free cross-attention: AR rows attend to background prompt a_B plus chunk prompts a_≤i, lower-triangular. Bidirectional rows see one global prompt a_G → DMD runs over long self-rollout trajectories, not teacher-forced states — the student is optimized on the distribution its own errors induce → Director-Pilot harness: a VLM proposes event cards, the DiT generator renders physical dynamics. Mode B adds a SAM tracking loop for object-centric interaction → One 60-minute uninterrupted session, 20 distinct scenarios, no perceptible decay Full analysis: Paper: Model weight: GitHub Repo: Project: Ant Group Robbyant

Marktechpost AI

233,449 views • 3 months ago