Project Instinct

Ziwen Zhuang12, Shaoting Zhu12, Mengjie Zhao1, Hang Zhao12,
1IIIS, Tsinghua University 2Shanghai Qi Zhi Institute
YouTube Bilibili Twitter InstinctLab Instinct_RL Instinct Onboard Robot Motion Editor

Instinct-Level intelligences for whole-body control.

Abstract

We present a unified framework from algorithm, environment, dataset curation, and deployment for Instinct-Level intelligence on humanoid robots.

Research Projects



Embrace Collisions: Humanoid Shadowing for Deployable Contact-Agnostics Motions

Ziwen Zhuang†, Hang Zhao,
Conference on Robot Learning (CoRL) 2025
arXiv YouTube Bilibili

Deep Whole-Body Parkour

Ziwen Zhuang†, Shaoting Zhu†, Mengjie Zhao, Hang Zhao,
In Submission
arXiv YouTube Bilibili

Hiking in the Wild: A Scalable Perceptive Parkour Framework for Humanoids

Shaoting Zhu†, Ziwen Zhuang†, Mengjie Zhao, Hang Zhao,
In Submission
arXiv YouTube Bilibili

Auto position-correction from depth sensing

Observed Instinct-Level Behaviors

Depth Onboard

Repeated Tests

Downstream usecase: Distill from tracking policies