I am a Research Engineer at Google DeepMind in London, where I work on building generalist foundation models and embodied intelligence for the physical world—including Gemini Robotics 2, Gemini Robotics 1.5, Gemini 2.5, and RoboCat.
Prior to joining DeepMind in 2018, I completed my Ph.D. in Computing at Imperial College London under the supervision of Prof. Stefanos Zafeiriou, researching statistical deformable models, high-speed 3D mesh decoding, and third-person visual imitation learning.
Bridging large multimodal foundation models, reinforcement learning, and 3D visual perception to enable general-purpose physical intelligence.
Scaling multimodal Vision-Language-Action (VLA) architectures with embodied reasoning, long-horizon planning, cross-embodiment motion transfer, and self-improving data generation loops across diverse robotic hardware.
Learning dexterous manipulation and contact-rich skills (such as diverse-shape RGB stacking) via manipulator-independent third-person imitation, offline RL with policy kickstarting, and self-supervised sim-to-real adaptation.
Statistical deformable shape modelling, convolutional mesh decoders capable of >2,500 FPS 3D face reconstruction, adversarial UV map completion, and dense 3D pose estimation in unconstrained environments.
Over a decade of research and engineering across Google DeepMind, Google Brain, and Imperial College London.
Core contributor to flagship robotics foundation models (Gemini Robotics 2, Gemini Robotics 1.5, Gemini Robotics, Gemini 2.5, and RoboCat). Researching cross-embodiment manipulation, visual imitation learning, and large-scale reinforcement learning for real-world robots.
Joint research internship across Google Brain and DeepMind Robotics, focusing on deep reinforcement learning, third-person visual imitation learning, and robotic manipulation.
Led small-group tutorials, laboratory sessions, and coursework assessment for undergraduate computing courses covering Haskell, C++, Java, Compilers, and the Pintos Operating System kernel.
Member of the Bloomberg Mobile Professional team, building and shipping real-time financial analytics workflows for Bloomberg Anywhere across hybrid web and native mobile platforms.
Intelligent Behaviour Understanding Group (iBUG), supervised by Prof. Stefanos Zafeiriou.
Thesis: Dense Pose Estimation of Deformable Objects — Statistical deformable models, dense correspondence estimation without landmarks, 3D mesh decoding, and imitation learning.
Four-year integrated Master of Engineering in Computing.
Master's Thesis: Predictive Remote Execution for Mobile Code Offloading; group lead for Pintos OS kernel & Microsoft Research GPUVerify Eclipse IDE plugin.
Peer-reviewed conference papers, journal articles, and technical reports. Equal contribution is denoted by *.
Next-generation multimodal embodied reasoning and generalist control architecture designed to generalize across diverse robotic embodiments and complex real-world manipulation tasks.
Introduces interleaved embodied thinking and cross-embodiment motion transfer, enabling generalist robots to reason through multi-stage physical tasks before acting.
@misc{geminiroboticsteam2025geminirobotics15pushing,
title={Gemini Robotics 1.5: Pushing the Frontier of Generalist Robots with Advanced Embodied Reasoning, Thinking, and Motion Transfer},
author={Gemini Robotics Team and others and Zhou, Yuxiang},
year={2025},
eprint={2510.03342},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2510.03342}
}
Foundational technical report detailing the Gemini 2.5 model family, native thinking capabilities, multimodal understanding, and agentic reasoning across digital and physical domains.
@article{comanici2025gemini25,
title={Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities},
author={Comanici, Gheorghe and Bieber, Eric and Schaekermann, Mike and others and Zhou, Yuxiang},
journal={arXiv preprint arXiv:2507.06261},
year={2025}
}Adapts Gemini's multimodal world understanding into physical actions, achieving zero-shot and few-shot dexterous manipulation across bimanual and humanoid robot platforms.
@article{team2025gemini,
title={Gemini Robotics: Bringing AI into the Physical World},
author={Team, Gemini Robotics and Abeyruwan, Saminda and Ainslie, Joshua and Alayrac, Jean-Baptiste and Arenas, Montserrat Gonzalez and others and Zhou, Yuxiang},
journal={arXiv preprint arXiv:2503.20020},
year={2025}
}
The first visual goal-conditioned foundation agent capable of controlling multiple robot embodiments and autonomously generating new training data to self-improve on novel tasks from as few as 100 demonstrations.
@article{bousmalis2023robocat,
title={RoboCat: A Self-Improving Foundation Agent for Robotic Manipulation},
author={Bousmalis, Konstantinos and Vezzani, Giulia and Rao, Dushyant and Devin, Coline and Lee, Alex X and Bauza, Maria and Davchev, Todor and Zhou, Yuxiang and Gupta, Agrim and Raju, Akhil and others},
journal={Transactions on Machine Learning Research (TMLR)},
year={2023}
}
Demonstrates how side-network and adapter architectures can inject task-specific visuomotor control features into frozen pretrained vision backbones without catastrophic forgetting.
@inproceedings{sharma2023lossless,
title={Lossless Adaptation of Pretrained Vision Models For Robotic Manipulation},
author={Sharma, Mohit and Fantacci, Claudio and Zhou, Yuxiang and Koppula, Skanda and Heess, Nicolas and Scholz, Jon and Aytar, Yusuf},
booktitle={International Conference on Learning Representations (ICLR)},
year={2023}
}@inproceedings{lee2022spend,
title={How to Spend Your Robot Time: Bridging Kickstarting and Offline Reinforcement Learning for Vision-based Robotic Manipulation},
author={Lee, Alex X and Devin, Coline and Springenberg, Jost Tobias and Zhou, Yuxiang and Lampe, Thomas and Abdolmaleki, Abbas and Bousmalis, Konstantinos},
booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
year={2022}
}
Introduces the RGB-Stacking benchmark and trains a vision-based generalist policy via sim-to-real RL and offline data distillation to balance and stack geometrically complex objects.
@inproceedings{lee2021beyond,
title={Beyond Pick-and-Place: Tackling Robotic Stacking of Diverse Shapes},
author={Lee, Alex X and Devin, Coline Manon and Zhou, Yuxiang and Lampe, Thomas and Bousmalis, Konstantinos and Springenberg, Jost Tobias and Byravan, Arunkumar and Abdolmaleki, Abbas and Gileadi, Nimrod and Khosid, David and others},
booktitle={5th Annual Conference on Robot Learning (CoRL)},
year={2021}
}
Learns disentangled, embodiment-agnostic visual state representations that allow a robot to imitate skills directly from third-person demonstrations performed by humans or different robot arms.
@inproceedings{zhou2021manipulator,
title={Manipulator-Independent Representations for Visual Imitation},
author={Zhou, Yuxiang and Aytar, Yusuf and Bousmalis, Konstantinos},
booktitle={Robotics: Science and Systems (RSS)},
year={2021}
}
@inproceedings{zambelli2020learning,
title={Learning rich touch representations through cross-modal self-supervision},
author={Zambelli, Martina and Aytar, Yusuf and Visin, Francesco and Zhou, Yuxiang and Hadsell, Raia},
booktitle={Conference on Robot Learning (CoRL)},
pages={1415--1425},
year={2020}
}
@inproceedings{jeong2020learning,
title={Learning Dexterous Manipulation from Suboptimal Experts},
author={Jeong, Rae and Springenberg, Jost Tobias and Kay, Jackie and Zheng, Daniel and Zhou, Yuxiang and Galashov, Alexandre and Heess, Nicolas and Nori, Francesco},
booktitle={Conference on Robot Learning (CoRL)},
year={2020}
}
@inproceedings{jeong2020self,
title={Self-supervised sim-to-real adaptation for visual robotic manipulation},
author={Jeong, Rae and Aytar, Yusuf and Khosid, David and Zhou, Yuxiang and Kay, Jackie and Lampe, Thomas and Bousmalis, Konstantinos and Nori, Francesco},
booktitle={2020 IEEE International Conference on Robotics and Automation (ICRA)},
pages={2718--2724},
year={2020},
organization={IEEE}
}
Unifies face box prediction, 2D facial landmark localization, and dense 3D mesh regression in a single-stage multi-task network; one of the most widely adopted face detectors in computer vision.
@inproceedings{deng2020retinaface,
title={RetinaFace: Single-stage dense face localisation in the wild},
author={Deng, Jiankang and Guo, Jia and Zhou, Yuxiang and Yu, Jinke and Kotsia, Irene and Zafeiriou, Stefanos},
booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2020}
}
Proposes joint graph/mesh convolutional decoders that reconstruct dense 3D facial geometry and albedo texture from a single in-the-wild image at over 2,500 frames per second.
@inproceedings{zhou2019dense,
title={Dense 3d face decoding over 2500fps: Joint texture \& shape convolutional mesh decoders},
author={Zhou, Yuxiang and Deng, Jiankang and Kotsia, Irene and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={1097--1106},
year={2019}
}
@article{cheng2019meshgan,
title={Meshgan: Non-linear 3d morphable models of faces},
author={Cheng, Shiyang and Bronstein, Michael and Zhou, Yuxiang and Kotsia, Irene and Pantic, Maja and Zafeiriou, Stefanos},
journal={arXiv preprint arXiv:1903.10384},
year={2019}
}
@inproceedings{deng2018uv,
title={Uv-gan: Adversarial facial uv map completion for pose-invariant face recognition},
author={Deng, Jiankang and Cheng, Shiyang and Xue, Niannan and Zhou, Yuxiang and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={7093--7102},
year={2018}
}
@inproceedings{zhou2018improve,
title={Improve accurate pose alignment and action localization by dense pose estimation},
author={Zhou, Yuxiang and Deng, Jiankang and Zafeiriou, Stefanos},
booktitle={2018 13th IEEE International Conference on Automatic Face \& Gesture Recognition (FG 2018)},
pages={480--484},
year={2018},
organization={IEEE}
}
@inproceedings{deng2018cascade,
title={Cascade multi-view hourglass model for robust 3d face alignment},
author={Deng, Jiankang and Zhou, Yuxiang and Cheng, Shiyang and Zaferiou, Stefanos},
booktitle={2018 13th IEEE International Conference on Automatic Face \& Gesture Recognition (FG 2018)},
pages={399--403},
year={2018},
organization={IEEE}
}
@article{deng2019joint,
title={Joint multi-view face alignment in the wild},
author={Deng, Jiankang and Trigeorgis, George and Zhou, Yuxiang and Zafeiriou, Stefanos},
journal={IEEE Transactions on Image Processing},
volume={28},
number={7},
pages={3636--3648},
year={2019},
publisher={IEEE}
}
@inproceedings{deng2017marginal,
title={Marginal loss for deep face recognition},
author={Deng, Jiankang and Zhou, Yuxiang and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops},
pages={60--68},
year={2017}
}
@inproceedings{zhou2017deformable,
title={Deformable Models of Ears in-the-wild for Alignment and Recognition},
author={Zhou, Yuxiang and Zaferiou, Stefanos},
booktitle={2017 12th IEEE International Conference on Automatic Face \& Gesture Recognition (FG 2017)},
pages={626--633},
year={2017},
organization={IEEE}
}
@inproceedings{zhou2016estimating,
title={Estimating correspondences of deformable objects 'in-the-wild'},
author={Zhou, Yuxiang and Antonakos, Epameinondas and Alabort-i-Medina, Joan and Roussos, Anastasios and Zafeiriou, Stefanos},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={5791--5801},
year={2016}
}
@inproceedings{zhang2016semi,
title={Semi-autonomous data enrichment based on cross-task labelling of missing targets for holistic speech analysis},
author={Zhang, Yue and Zhou, Yuxiang and Shen, Jie and Schuller, Bj{\"o}rn},
booktitle={2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages={6090--6094},
year={2016},
organization={IEEE}
}
@inproceedings{liao2012fingertips,
title={Fingertips detection algorithm based on skin colour filtering and distance transformation},
author={Liao, Yun and Zhou, Yuxiang and Zhou, Hua and Liang, Zhihong},
booktitle={2012 12th International Conference on Quality Software},
pages={276--281},
year={2012},
organization={IEEE}
}
No publications match your current filter or search query.