Chenhao Li

🌎World models can predict, but controlling real robots from imagination sees a long-standing failure due to hallucination. 🧠Introducing Uncertainty-Aware RWM: a black-box, end-to-end neural dynamics model with long-horizon uncertainty propagation. 🎯sites.google.com/view/uncertain…

@t641769919 @svlevine Yea I wonder if I missed something obvious or it is indeed something crucial.

@breadli428 @svlevine That would be top secret😉

Back in Nov we developed Recap and trained π*-06 with RL. Now, we developed a fast *online* RL method that improves π-06 with as little as 15 min of robot data for precise tasks, using "RL tokens" exposed by our model that can be fed into a small actor-critic method.

✅ Use world models to control ❌ only action-conditioned generation ➡️ Model-Based RL Non-expert data in -> policy out github.com/leggedrobotics…

@MBjelonic @ChongZitaZhang @rivr_tech @amazon Legend

I’m excited to share that @rivr_tech is now part of @amazon, marking the start of an exciting new chapter for our team.  This step will accelerate our vision of building General Physical AI through doorstep delivery, bringing robotics and AI closer to real-world deployment at scale. A huge thank you to the entire RIVR team for the incredible dedication, resilience, and hard work that made this possible. A special thank you to Giorgio Valsecchi, Lorenz Wellhausen, and Alexander Reske for taking the leap together from day one and building RIVR side by side. I’m incredibly proud of what RIVR built, and even more excited about what we will accomplish next at Amazon.

Exactly the point 🔝

@tonyzzhao Congrats!!

We raised $165M at a $1.15B valuation to stop doing demos. 2026 is about 1) deployment and 2) research. We will start shipping Memo with our new frontier models in a few months. Our series-B is led by Coatue, with Thomas Laffont joining the board. ->🧵

@nolan_fey @gabe_mrgl

🤖We updated model training in the Robotic World Model Lite suite, training autoregressive dynamics models in parallel to previously released policy training! 🌐Now, you can do online training with data stream iteratively on both the model and policy! github.com/leggedrobotics…

@GMartius @rlfromlux Nice bot

I am very happy and super proud of Pierre Schumacher (@rlfromlux) for recieving the PhD award from #RIG! It is also a great honor for me and Daniel Haeufle as your PhD supervisors to see (y)our work recognized! All the best for your future! #MachineLearning #Robotics

@MarwaEldiwiny Legend

Throwback: Sang Bae Kim on "Who is going to win, legged vs wheeled?" We did the recording in 2021. Sang Bae Kim led the development of the MIT Cheetah robots. It was fascinating to reflect on his insights from back then.

@robot_in_space2 @UnitreeRobotics 🫡

The G1 robot from @UnitreeRobotics bowing in front of my robot. Robot respect 🤖

@LukasForTech Hope it helps!

@breadli428 Elegant sim-to-real solution. Exactly what I need for UAV aerodynamics where uncertainty breaks pure simulation.

Check out our project page for the paper, more videos and code! sites.google.com/view/uncertain…

.@leggedrobotics introduces Robotic World Model, a neural network simulator for robust policy optimisation in robotics. Read the paper here: arxiv.org/pdf/2501.10100 #humanoids #humanoidrobotics #humanoidrobot #humanoid

Robotic World Model: A Neural Network Simulator for Robust Policy Optimization in Robotics Project: sites.google.com/view/roboticwo… Paper: arxiv.org/abs/2501.10100 Code: github.com/leggedrobotics… ETH just released a joint model-based RL policy and neural dynamics model framework for quadruped and humanoid locomotion task in IsaacLab. - Robotic World Model: in this work, authors define the learned dynamics to support long horizon autoregressive prediction as world model. In implementation, this means next observation and privileged info as output of the World Model. The authors are considering a blind policy which only includes robot proprioceptive info as observation. - Authors evaluated model-based and model-free RL policies in “imagined” rollout from learned dynamics. Conclusion: Model free RL + Robot World Model beats model-based RL methods in all dimensions - prediction accuracy, policy learning and sim-to-real transfer. - While this brings some good news on world model use in real robotics applications, there’re still many limitations of this approach, and one of them is to introduce robust vision in this framework.

@YXWangBot Awesome, thanks for the info!

@breadli428 Yes! We are not actually "extracting" planning policy. Instead, we train imitation learning using generated data. But yes, we will opensource that part!

1/ World models are getting popular in robotics 🤖✨ But there’s a big problem: most are slow and break physical consistency over long horizons. 2/ Today we’re releasing Interactive World Simulator: An action-conditioned world model that supports stable long-horizon interaction. 3/ Key result: ✅ 10+ minutes of interactive prediction ✅ 15 FPS ✅ on a single RTX 4090🔥 4/ Why this matters: it unlocks two critical robotics applications: 🚀 Scalable data generation for policy training 🧪 Faithful policy evaluation 5/ You can play with our world model NOW at #interactive-demo" target="_blank" rel="nofollow noopener">yixuanwang.me/interactive_wo…. NO git clone, NO pip install, NO python. Just click and play! NOTE ⚠️ ALL videos here are generated purely by our model in pixel space! They are **NOT** from a real camera More details coming 👇 (1/9) #Robotics #AI #MachineLearning #WorldModels #RobotLearning #ImitationLearning

@NimaZeighami Great now I see what happened @cixliv

You know how they say don’t put two betta fish in the same tank? Well, we just learned that’s true for fighting robots too.

@YXWangBot Oh I meant extracting a control policy (planning) out of the world model and deploying to real robot. Would this be released too?

@breadli428 Yes! We already open-sourced part of the real robot code and we are still organizing more code!

@YXWangBot Very cool work! Congrats!