Tsinghua IIIS

Thrilled to share one of my favorite works this year: DexNDM! We bridge the Sim2Real gap for dexterous in-hand rotation, achieving a true "0-to-1" advancement. The key? DexNDM learns from biased, real-world data without needing any successful demonstrations. Now a general-purpose dexterous hand can stably rotate large books, long rods, & complex objects around any axis, from any wrist pose. This powerful primitive enables complex, long-horizon tasks like teleoperated screwing and furniture assembly. 📄 Paper: arxiv.org/abs/2510.08556 🌐 Project Page: meowuu7.github.io/DexNDM/

🧩#CVPR2025🌷Introducing Two By Two✌️: The First Large-Scale Daily Pairwise Assembly Dataset with SE(3)-Equivariant Pose Estimation. 🤖2BY2 helps robots master daily 3D assembly tasks—like plugging sockets or arranging flowers—across diverse objects! 🐨Co-lead by @yuqi_Beijing

@taodu_eecs Prof. Tao Du’s team @Tsinghua_IIIS has developed a computational bionics workflow that can automatically design the morphology of #wingedseeds, simulate flight & optimize performance. It will be presented at @SIGGRAPHAsia. #InnovativeTsinghua bit.ly/3Y61gG3

A fun project from our TEA lab! A mobile manipulator robot that can catch objects in flight with a dex hand! I'd buy one of it at home for just seeing it doing the task lol. Everything is achieved through sim2real RL.

When will self-driving cars be a reality? Prof. Hang Zhao's IIIS MARS Lab from @Tsinghua_Uni has teamed up with Li Auto to unveil DriveVLM, a new approach using Visual Language Models to boost autonomous driving. I talked with lead researcher Dr. Xiaoyu Tian about DriveVLM's mechanism, real-world test results, and the future of autonoumous vehicles. 🚗🤖Paper link: tsinghua-mars-lab.github.io/DriveVLM/ #AutonomousDriving #AI #TechInnovation @zhaohang0124

How can we utilize human RGBD video as a data source for robot pretraining? Check out our latest work General Flow!

A new academic year @Tsinghua_Uni always means a fresh batch of brilliant students. Tia Chen, a UC Berkeley graduate from the United States, has joined us in 2024 with a deep passion for embodied AI. In an exclusive interview, she told us about her choice and experience of pursuing a doctoral degree in Robotics at IIIS Tsinghua. iiis.tsinghua.edu.cn/en/show-10622-…

I am so flattered that our paper “The Earth is Flat because...: Investigating LLMs' Belief towards Misinformation via Persuasive Conversation” won the outstanding paper award at #ACL2024 @aclmeeting 🎉🎉 Congrats to all my collaborators!! Photo credit to @shi_weiyan

How well can Large Language models handle misinformation? I interviewed Rongwu Xu, a graduate student at Tsinghua University's IIIS, about his latest research titled "Investigating LLMs' Belief towards Misinformation via Persuasive Conversation". The work has been accepted by the top NLP conference ACL 2024 and scored a perfect Meta 5/5 from the Area Chair. @aclmeeting @Tsinghua_IIIS Paper Link: …s-believe-the-earth-is-flat.github.io News Link: iiis.tsinghua.edu.cn/show-10561-1.h…

Drop the camera, and bang -- the robots do the job without camera calibration or additional data collection. We want robots to manipulate anywhere! Our new work, **Maniwhere**, enables robots to manipulate objects from any camera view with any background. Isn't this the generalization every roboticist is looking for? How do we achieve this with sim2real? You can check this thread x.com/fancy_yzc/stat… from @fancy_yzc ! And the website is here: gemcollector.github.io/maniwhere/

Congrats to Prof.Jianyu Chen’s research team! They received the Outstanding Paper Award Finalist in the RSS 2024. @JianyuChen_THU @wangyenjen

Congrats to Prof.Huazhe Xu’s research team. Their latest work Robo-ABC🤖️has been accepted by ECCV2024! @HarryXu12 @ju_yuanchen @eccvconf

QuasiSim has been accepted to #ECCV2024 ! The code has also been released and welcome to try it out: github.com/Meowuu7/QuasiS…

A robot took part in the commencement ceremony at the Institute for Interdisciplinary Information Sciences at @Tsinghua_Uni. This advanced humanoid robot, developed by a research group at Tsinghua, helped in handing out diplomas and posed for photos with students. #StudyinBeijing

Prof. Luming Duan’s team @Tsinghua_IIIS set a world record in #QuantumComputing by realizing the stable trapping and cooling of a two-dimensional crystal of up to 512 ions and performed a quantum simulation with 300 ion qubits. Their work has been published by Nature (@Nature) and lauded by reviewers as “a milestone to be recognized”. For more: iiis.tsinghua.edu.cn/en/show-10566-…

Prof. Yang Gao (@gao_young)'s research team from Tsinghua IIIS, in collaborate with Pieter Abbeel (@pabbeel)'s group at UC Berkeley, has achieved a breakthrough in robot learning from videos. Their work, Any-point Trajectory Modeling for Policy Learning, was accepted by the top conference Robotics: Science and Systems (RSS) @RoboticsSciSys, receiving a perfect score of 4.0 from all reviewers. This work proposed a novel point-trajectory prediction framework which successfully extracts motion priors from action-free videos and bridges the embodiment gaps, achieving few-shot and cross-embodiment imitation learning. For more: iiis.tsinghua.edu.cn/en/show-10563-… Congrates @ChuanWen15, @Xingyu2017, @johnrso_, @gao_young, and @pabbeel.

A team of physicists, led by Prof. Luming Duan, unlocked new dimensions with trapped #ions, revealing chiral edge modes and #quantum transitions, taking quantum research to new heights. #InnovativeTsinghua For more: bit.ly/3yvTOcU

An article titled “Managing extreme AI risks amid rapid progress” has just been published on @ScienceMagazine. Turing award winners Y. Bengio, G. Hinton, A. Yao and other scholars from US, Canada, Europe, Israel and China collaborated on the critical issue of AI safety, making recommendations that combine technical research with proactive, adaptive governance mechanisms. Link: science.org/doi/10.1126/sc…

Professor Jianyu Chen (@JianyuChen_THU)’s research team from Tsinghua IIIS, in collaboration with @roboterax, has achieved a breakthrough in humanoid robots. Their work titled Advancing Humanoid Locomotion: Mastering Challenging Terrains with Denoising World Model Learning (DWL), received a perfect score of 4.0 from all reviewers and was accepted by Robotics: Science and Systems (RSS) @RoboticsSciSys, one of the most prestigious conferences in the field of robotics. The work proposed a novel denoising learning framework which greatly solves the sim2real problem and significantly enhances the robots' adaptability in complex environments. News article: iiis.tsinghua.edu.cn/en/show-10551-… Congrates @JianyuChen_THU, @wangyenjen, and staff from @roboterax

Congrats to our good friend, Prof. Avi Wigderson, for winning the 2023 #TuringAward🎉! Memorable visit to IIIS in Nov 2023! Come again!