Minghuan Liu
@ericliuof97
Prev: Seed Robotics; Ph.D @sjtu1896; Visit @UCSD at @xiaolonw's lab. Robot Learning, Reinforcement Learning, Imitation Learning.
Thrilled to introduce 🦏RHINO: Learning Real-Time Humanoid-Human-Object Interaction from Human Demonstrations! Project: humanoid-interaction.github.io RHINO is our recent attempt on human-robot interaction to bring humanoids into human's real life. To do so, we require highly…
Agree with the decoupled design. Note that our recent work (hugwbc.github.io) exactly shows how a robust whole-body (low-body) controller can support left upper-body control (leaved as an interface for a high-level model).
Everything is learned from data - the network also controls full body motion, including directing the RL lower body controller + arms and spine joints at the same time.
My Takeaways from GR3: Insights on Vision-Language-Action (VLA) Models Although I wasn’t directly involved in the GR3 project, I gained valuable insights into Vision-Language-Action (VLA) models through discussions and observing their performance, particularly their failure…
🚀🚀🚀 Ever wondered what it takes for robots to handle real-world household tasks? long-horizon execution, deformable object dexterity, and unseen object generalization — meet GR-3, ByteDance Seed’s new Vision-Language-Action (VLA) model! GR-3 is a generalizable…
Ep#10 with @RogerQiu_42 on Humanoid Policy ~ Human Policy human-as-robot.github.io Co-hosted by @chris_j_paxton & @micoolcho
The performance looks good! Congrats.
🚀Introducing GMT — a general motion tracking framework that enables high-fidelity motion tracking on humanoid robots by training a single policy from large, unstructured human motion datasets. 🤖A step toward general humanoid controllers. Project Website:…
👀Optimizing Anytime Reasoning via Budget Relative Policy Optimization👀 🚀Our BRPO leverages verifiable dense rewards, significantly outperforming GRPO in both final and anytime reasoning performance.🚀 📰Paper: arxiv.org/abs/2505.13438 🛠️Code: github.com/sail-sg/Anytim…
We bring our H1 robot equipped with the RHINO system to Shanghai Huayu Middle school! Our robot seems robust to the new environment and kids love to interact with him! Thank Jiahang and Wentao for the setup efforts to make this come true!
Thrilled to introduce 🦏RHINO: Learning Real-Time Humanoid-Human-Object Interaction from Human Demonstrations! Project: humanoid-interaction.github.io RHINO is our recent attempt on human-robot interaction to bring humanoids into human's real life. To do so, we require highly…
"Gr00t" vs "Pi0" vs "Pi0 Fast". I compared top open-source robotic models, and here's a detailed overview based on our own experience:
We took a short break from robotics to build a human-level agent to play Competitive Pokémon. Partially observed. Stochastic. Long-horizon. Now mastered with Offline RL + Transformers. Our agent, trained on 475k+ human battles, hits the top 10% on Pokémon Showdown leaderboards.…
Diverse training data leads to a more robust humanoid manipulation policy, but collecting robot demonstrations is slow. Introducing our latest work, Humanoid Policy ~ Human Policy. We advocate human data as a scalable data source for co-training egocentric manipulation policy.⬇️
Thrilled to announce GR00T N1, our open foundation model for generalist humanoid robots! GR00T N1 adopts a dual-system design, leverages the entire data pyramid for model training, and supports various robot embodiments. GR00T N1 embodies years of fundamental research, spanning…
Sharing the slide deck and video recording of my talk "Data Pyramid and Data Flywheel for Robotic Foundation Models" at Princeton Robotics Symposium last November. I discussed the vision of training foundation models on diverse data sources and refining them during deployments.…
🚀The NVIDIA team has released the IsaacLab-based HOVER code to the public! Huge thanks to the team for their incredible effort in making this happen. Now you can play around with the code and train HOVER/OmniH2O! 🤖 🛠️ Code: github.com/NVlabs/HOVER/
Introducing our recent progress to utilize synthetic data—RE3SIM—a Real-to-Sim-to-Real pipeline that integrates Gaussian splatting with NVIDIA Isaac Sim's PhysX engine, improving scene reconstruction and sim-to-real transfer for robotic manipulation tasks. Project:…