What if a humanoid robot could choose how to interact with the environment 🤖 — soft when it needs compliance, stiff when it needs precision, and force-aware when it must push/pull? That’s exactly what our Heterogeneous Meta-Control (HMC) framework enables. Our new framework

Learning from human video data is the future, the real way to scale up.

Cool results🤖🚀 I'm curious if we could scale such methods beyond quasi-static tasks... kinematic retargeting is great, but for dynamic retargeting it seems like we'd need some human hand/arm sys-id, but maybe average values might be enough too!

Proud to share our latest work In-N-On, with my amazing labmates! Check out the first-author thread below 👇

Egocentric manipulation data at scale allows for following of previously-unseen language instructions -- this is a very important work for indicating the value of egocentric data as a key part of the training pipeline for humanoid robots.

Check out our latest work on scaling egocentric human data🫳 for dexterous manipulation🦾! Compared to teleportation, human data is much easier to scale in quantities, objects, scenes, and tasks, which unlocks key capabilities such as novel language following and enhanced

Humanoids need to mimic humans, and we made a human behavior model for it. Was fun working with Xiongyi Cai

If we want humanoid robots to truly learn to "work like humans," In-N-On may offer a very realistic path. It uses first-person human video (the perspective of a worker wearing glasses) as a large-scale demonstration, supplemented with a small amount of the robot's own task data,

Scale

How do you teach a robot to do something it has never seen before? 🤖 With human data. Our new Human0 model is co-trained on human and humanoid data. It allows the robot to understand a novel language command and execute it perfectly in the wild without prior practice.

Meet ACE-F — a novel, foldable teleoperation platform for collecting high-quality robot demonstration data across robot embodiments. Using a specialized soft-controller pipeline, we interpret end-effector positional deviations as virtual force signals to provide the user with

We are hiring at #NeurIPS2025 Catch me if you can🤩

ACE-F is finally open sourced, with a hardware assembly tutorial and teleoperation code for Franka and Xarm7 robots. Check out our website and more below!Hardware: github.com/ACEFoldable/ac… Software: github.com/ACEFoldable/ac… Webpage: acefoldable.github.io Arxiv:

Can we bridge the Sim-to-Real gap in complex manipulation without explicit system ID? 🤖 Presenting Contact-Aware Neural Dynamics — a diffusion-based framework that grounds simulation with real-world touch. Implicit Alignment: No tedious parameter tuning. Tactile-Driven: