🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

ICCV workshop on Simulation technology for Embodied AI. We are open for submissions .

#ICCV2021 Join us this Saturday Oct 16, 8-17 EDT for the first workshop on Simulation Technology for Embodied AI (SEAI, iccv21-seai.github.io). We feature 10 keynote talks by world-renowned speakers as well as 11 spotlight talks covering topics from simulation and embodied AI.

Congratulations to all #CVPR2022 authors. This is my first CVPR season at Peking University PKU-Center on Frontiers of Computing Studies. We have seven papers in, half of which the leading authors are undergrads. Looking forward to sharing the works soon.

In this [#CVPR2022], my students have two papers on rotation regression: 1) RPMG proposes a manifold-aware gradient layer, yielding the new SOTA and can improve quaternion/6D/9D/10D representation. Paper: arxiv.org/abs/2110.11657 and code: github.com/JYChen18/RPMG.

2) A CVPR oral: FisherMatch considers semi-supervised rotation regression by leveraging matrix Fisher distribution and entropy-based label filtering. Feel free to apply to human/hand/head/object/camera pose. Paper: arxiv.org/abs/2203.15765 and project: yd-yin.github.io/FisherMatch.

Welcome to apply for the international Master and PhD programs at School of Computer Science, Peking University. My Lab, Embodied Perception and Interaction (EPIC) Lab (hughw19.github.io), welcomes students that are interested in Embodied AI, 3D Vision and Robotics to join.

Among 7 accepted papers from our lab, GAPartNet (pku-epic.github.io/GAPartNet/) is selected as a highlight #CVPR2023. The paper tackles cross-category object perception and manipulation via proposing and leveraging GAParts (generalizable and actionable parts, e.g. botton, handle, etc).

Excited to announce the #RSS2023 Workshop on Symmetries in Robot Learning! Submit your work and join us to explore how symmetries can help in robotics with our incredible speakers @andyzengtweets He Wang Hao Su Prof. Tess Smidt Erik Bekkers. Website: sites.google.com/view/rss23-sym.

Our dexterous grasping synthesis work, DexGraspNet (pku-epic.github.io/DexGraspNet/), has been selected as a finalist of ICRA 2023 outstanding manipulation paper. This work proposes a highly efficient method to generate a dataset composed of 1.32 million dexterous grasps for 5K objects.

We will present our #CVPR2023 highlight paper, GAPartNet, in the afternoon of June 20 at West Building Exhibit Halls ABC 282! GAPartNet focuses on cross-category part segmentation, pose estimation and manipulation and provides a large-scale part-based interactive dataset with

Our generalizable dexterous grasping policy learning work, UniDexGrasp++, is awarded as ICCV 2023 best paper finalist🎉 . Congratulations on first authors, Weikang Wan Weikang Wan and Haoran Geng Haoran Geng, and many thanks to our collaborators Li Yi Yaodong.Yang.

🚀Exciting News! We’re thrilled to introduce OpenWBT — the open-source, whole-body VR teleoperation system designed for humanoid across various robot types & virtual/physical worlds! 🤖✨ 💡Developed through collaboration between Galbot and Tsinghua University researchers! 🔥

Proud to announce this landmark JV with Bosch China and BOYUAN Capital! Galbot is committed to turning embodied intelligence into industrial reality. Let’s redefine smart production together. #Galbot #Robotics #EmbodiedAI #Bosch #BOYUAN

China has made great strides in emerging technologies, especially in AI and robotics. Visited the Beijing Humanoid Robotics Innovation Centre (X-Humanoid) to learn more about their cutting-edge R&D and how these innovations can be applied to real-world needs.

🎉🎉🎉

Honored to join the Prime Minister of Singapore Lawrence Wong and fellow business leaders for an insightful lunch. As a representative of the new generation of Chinese entrepreneurs — and the only one pioneering general-purpose humanoid robots — I’m inspired by Singapore’s

If only the intersection area yields a viable data solution for Embodied AI, why do GraspVLA and TrackVLA — trained solely on synthetic action data — work zero-shot in the wild? And isn’t cross-embodiment data basically a spork? Why does it outperform synthetic data as an