Sirui Chen

What missing in RL based humanoid controller from industrial robots are precision and force control. CHIP can do both. We propose a simple recipe to build humanoid impedance controller, which can be used for wiping, carrying large objects and multi-robot collaboration.

Can we learn whole-body mobile manipulation directly from human demonstrations? Introducing Whole-Body Mobile Manipulation Interface (HoMMI) Egocentric + UMI, 0 teleop -> bimanual & whole-body manipulation, long-horizon navigation, active perception hommi-robot.github.io

Excited to release Minimalist Compliance Control! We achieve robust, compliant robot interaction across robot arms, dexterous hands, and humanoids, with NO force sensors or learning. If you’re wondering what remains, please see the thread below😉 Website: …nimalist-compliance-control.github.io

Generalist for dex manipulation!

Finally everyone can benefit from the result of 700h motion + 128 GPU training!

We have seen rapid progress in humanoid control — specialist robots can reliably generate agile, acrobatic, but preset motions. Our singular focus this year: putting generalist humanoids to do real work. To progress toward this goal, we developed SONIC (nvlabs.github.io/GEAR-SONIC/), a Behavior Foundation Model for real-time, whole-body motion generation that supports teleoperation and VLA inference for loco-manipulation. Today, we’re open-sourcing SONIC on GitHub. We are excited to see what the community builds upon SONIC and to collectively push humanoid intelligence toward real-world deployment at scale. 🌐 Paper: arxiv.org/abs/2511.07820 📃 Code: github.com/NVlabs/GR00T-W…

Just like in the video, the year of humanoid is unstoppable!

Long-tail scenarios remain a major challenge for autonomous driving. Unusual events—like accidents or construction zones—are underrepresented in driving data, yet require semantic and commonsense reasoning grounded in control. We propose SteerVLA, a framework that uses VLM reasoning to steer a driving policy via grounded, fine-grained language instructions. Paper: arxiv.org/abs/2602.08440 Website: steervla.github.io

This is amazing! Agility and adaptive compliant usually doesn’t come hand in hand, glad GentleHumanoid also make it works.

How can robots handle fragile, soft everyday objects like humans do, using vision & tactile to regulate force? 🤖🥚 Introducing our full-stack solution: a low-cost ($150) force gripper (0.45~45N), a force-aware teleoperator, and a reactive policy for learning force control.

@robotsdigest Hi thanks for posting our paper CHIP, it appears that the video and image are from HUMI (also a awesome paper). Could you make corrections, thanks

Humanoid robots do have rizz. CHIP shows how to add adaptive compliance without breaking motion tracking. A single controller handles wiping, door opening, box lifting, writing, and even running while carrying objects.

Humanoid robots are agile but stiff. CHIP shows how to add adaptive compliance without breaking motion tracking. A single controller handles wiping, door opening, box lifting, writing, and even running while carrying objects.

🚀 Introducing CHIP: Adaptive Compliance for Humanoid Control through Hindsight Perturbation! Current humanoids face a trade-off: they are either Agile & Stiff OR Slow & Soft. CHIP breaks this barrier. We enable on-the-fly switching between Compliant (wiping 🧼, collaborative holding 📦) and Stiff (lifting dumbbells 🏋️, opening doors 🚪💪) behaviors—all while maintaining agile skills like running! 🏃💨 Website: nvlabs.github.io/CHIP/ Join me for a deep dive on how CHIP enables adaptive control for complex tasks. 🧵↓

Robust humanoid perceptive locomotion is still underexplored. Especially when different cameras see different terrains, paths get narrow, and payloads disturb balance... Introduce RPL, tackling this with one unified policy: • Challenging terrains (slopes, stairs and stepping stones); • Multiple directions; • Payloads; Trained in sim. Validated long-horizon in the real world. Watch the robot walk it all🦿 Details below👇

So cool!

I'm on a singular mission to solve the Physical Turing Test for robotics. It's the next, or perhaps THE last grand challenge of AI. Super-intelligence in text strings will win a Nobel prize before we have chimpanzee-intelligence in agility & dexterity. Moravec's paradox is a curse to be broken, a wall to be torn down. Nothing can stand between humanity and exponential physical productivity on this planet, and perhaps some day on planets beyond. We started a small lab at NVIDIA and grew to 30 strong very recently. The team punches way above its weight. Our research footprint spans foundation models, world models, embodied reasoning, simulation, whole-body control, and many flavors of RL - basically the full stack of robot learning. This year, we launched: - GR00T VLA (vision-language-action) foundation models: open-sourced N1 in Mar, N1.5 in June, and N1.6 this month; - GR00T Dreams: video world model for scaling synthetic data; - SONIC: humanoid whole-body control foundation model; - RL post-training for VLAs and RL recipes for sim2real. These wouldn't have been possible without the numerous collaborating teams at NVIDIA, strong leadership support, and coauthors from university labs. Thank you all for believing in the mission. Thread on the gallery of milestones:

Door opening is hard, especially when the door open inwards. Congrats to the team!

Sim-to-real learning for humanoid robots is a full-stack problem. Today, Amazon FAR is releasing a full-stack solution: Holosoma. To accelerate research, we are open-sourcing a complete codebase covering multiple simulation backends, training, retargeting, and real-world inference.

Collecting data is frastrating, let robot collect its own data!

Introducing Gallant: Voxel Grid-based Humanoid Locomotion and Local-navigation across 3D Constrained Terrains 🤖 Project page: gallantloco.github.io Arxiv: arxiv.org/abs/2511.14625 Gallant is, to our knowledge, the first system to run a single policy that handles full-space constraints — including ground-level barriers, lateral clutter, and overhead obstacles on a humanoid robot. Instead of elevation maps or depth cameras, Gallant uses a voxel grid built directly from raw LiDAR as its perception representation, giving it inherent 3D coverage of the scene. With our custom LiDAR simulation toolkit (github.com/agent-3154/sim…), we model realistic scans, including returns from the robot’s own moving links, which is crucial for sim-to-real transfer. On the control side, we use a target-based training scheme rather than standard velocity tracking. The robot is given a goal and learns to discover its own in-path velocities and trajectories, so no external high-frequency command stream is needed during deployment. The policy itself is intentionally lightweight: just a 3-layer CNN + 3-layer MLP (~0.3M params), running onboard on the Unitree G1’s Orin NX at 50 Hz with no extra compute. Training takes about 6 hours on 8× NVIDIA RTX 4090 GPUs. The resulting policy transfers directly to the real robot and achieves >90% success rate on most tested terrain types. Gallant is our “half-way” step toward robust perceptive locomotion — a problem we believe remains fundamental for humanoid robots. We’re now working toward closing the gap to near-100% reliability and expanding the pipeline further. Code will be fully released soon. Discussion, feedback, and collaboration are very welcome! 🙌