[1/N] Large flow policies (e.g., VLAs) have become extremely popular, unlocking scalability. However, they come at a cost: the robot cannot react quickly. π𝐑² enables them to achieve reactive, real-time control.
Tracking human data offers anthropomorphic motion, but suffers from lack of data. Pure goal reaching achieves generalizability, but the learned motion can be unnatural. Turns out we can simply combine both by setting RL environments to be either former or latter! Led by
What can a robot hand learn from seconds of human motion?
It turns out: dexterous skills that generalize!
Excited to share AdaDex: Learning Adaptable Dexterous Skills via Hybrid Dexterous Mimicking.
From one short hand–object clip per skill, AdaDex learns an adaptable
Attending #ICRA2026 in Vienna this week! Feel free to reach out if you wanna chat.
Thurs: DemoDiffusion (demodiffusion.github.io) at main poster 9-10:30 am.
Fri: DemoDiffusion, Dex4D(dex4d.github.io) at Beyond Teleop workshop.
🤖 Attending ICRA 2026 and interested in dexterous manipulation? Don’t miss the chance to join or submit to our workshop!
📷Checkout our outstanding lineup of speakers, best poster award, and more details on our website: shorturl.at/ojsTo.
Deadline: April 10 (AOE).
Check out Dex4D from @yuxuank02: a framework for zero-shot dexterous manipulation in the real world via a two-level system.🎥 Video generation + 🎯 point tracks act as a high-level planner, while a 🤖 task-agnostic policy trained in simulation serves as the low-level controller.
How to enable dexterous hands with manipulation capabilities that work across diverse objects, tasks, scenes, camera views, and external perturbations?
Excited to share Dex4D, a method for generalizable sim-to-real dexterous manipulation via a task-agnostic point track policy