构建具身智能的
下一代基础模型

学术背景

刘扬博士毕业于美国密歇根大学，随后在哈佛大学从事博士后研究。他在人工智能和基础模型领域拥有十余年的研究经验，研究方向涵盖大语言模型对齐、强化学习、数据质量和世界模型，并在相关领域取得了多项具有重要影响力的研究成果。目前刘扬博士担任上海创智学院副教授职位。

研究与领导经历

刘扬博士在 NeurIPS、ICML、ICLR 等国际顶级人工智能学术会议及《Nature》等权威期刊上发表论文超过 50 篇，Google Scholar 学术引用量近 14,000 次，具有广泛的国际学术影响力。他曾任职于国际顶尖学术机构及科技企业，如亚马逊和字节跳动。并获得美国加州大学圣克鲁兹分校（UC Santa Cruz）终身教职。2024 年，刘扬博士加入天工 AI，作为基础模型负责人深度参与并推动了天工大模型从 3.0 到 5.0 版本的研发，同时主导了 Matrix 系列世界模型项目。他还曾担任 2050 Research 研究负责人、天工首席科学家等重要领导职务。

Academic background

Dr. Yang Liu earned his Ph.D. from the University of Michigan and subsequently conducted postdoctoral research at Harvard University. He has more than a decade of research experience in artificial intelligence and foundation models, with research spanning large language model alignment, reinforcement learning, data quality, and world models. He currently serves as an Associate Professor at the Shanghai Innovation Institute.

Research impact

Dr. Liu has authored more than 50 papers at leading artificial intelligence conferences, including NeurIPS, ICML, and ICLR, as well as in leading journals including Nature. His work has received nearly 14,000 citations on Google Scholar. He has held positions at leading academic institutions and technology companies, including Amazon and ByteDance, and was awarded tenure at the University of California, Santa Cruz.

Leadership experience

In 2024, Dr. Liu joined Skywork AI as Head of Foundation Models. He played a key role in advancing the Skywork model series from version 3.0 to 5.0 and led the Matrix series of world-model projects. He has also held leadership roles including Head of Research at 2050 Research and Chief Scientist at Skywork AI.

研究与经历

李阳光博士毕业于香港中文大学多媒体实验室（MMLab），现任黎曼动力（Riemann Dynamics）联合创始人，负责面向下一代机器人智能的具身世界动作模型（Embodied World Action Model）研发。此前，他曾担任天工（Skywork AI）研发总监及世界模型团队负责人，主导研发了 Matrix-Game 实时交互式世界模型系列。在此之前，他是 VAST 的创始团队成员、研发总监及基础模型负责人，主导构建了 Tripo3D 系列基础模型，并推动打造了全球领先的 AI 3D 生成平台之一。更早之前，他曾任职于商汤科技，负责大规模视觉预训练、多模态基础模型及自动驾驶感知等方向的研究。

学术与开源成果

李阳光博士累计发表学术论文 70 余篇，Google Scholar 学术引用量超过 4,300 次，主导及参与的开源项目在 GitHub 上累计获得超过 26,000 颗 Star。他的研究涵盖世界模型、3D 生成式人工智能及多模态基础模型，相关成果发表于 ICLR、NeurIPS、CVPR、ICCV 等国际顶级人工智能学术会议及期刊。

技术专长

李阳光博士在大规模数据基础设施建设、基础模型研发、分布式训练、推理优化及产品落地等方面拥有丰富的技术积累与实践经验，致力于构建面向机器人领域的通用人工智能。

Research & experience

Dr. Yangguang Li earned his Ph.D. from The Chinese University of Hong Kong, where he conducted research at the Multimedia Laboratory (MMLab). He is Co-Founder and CTO of Riemann Dynamics, where he leads the development of embodied world-action models for next-generation robotic intelligence.

Previously, Dr. Li served as R&D Director and Head of the World Model Team at Skywork AI, where he led the development of the Matrix-Game series of real-time interactive world models. Before that, he was a founding-team member, R&D Director, and Head of Foundation Models at VAST. There, he led the development of the Tripo3D foundation model series and helped build one of the world’s leading AI-powered 3D generation platforms. Earlier in his career, he worked at SenseTime on large-scale visual pretraining, multimodal foundation models, and autonomous-driving perception.

Research & open source

Dr. Li has authored more than 70 research papers and received more than 4,300 citations on Google Scholar. Open-source projects that he has led or contributed to have collectively received more than 26,000 GitHub stars. His work spans world models, 3D generative AI, and multimodal foundation models, with publications at leading venues including ICLR, NeurIPS, CVPR, and ICCV.

Technical expertise

He brings extensive experience in large-scale data infrastructure, foundation model development, distributed training, inference optimization, and product deployment. His work focuses on advancing general-purpose embodied intelligence for robotics.

研究方向

王耀晖博士毕业于法国国家信息与自动化研究所（Inria），自 2018 年起持续深耕视频生成领域。现任黎曼动力（Riemann Dynamics）联合创始人兼首席科学家，专注于面向具身智能的大规模世界模型研发。王耀晖博士在 CVPR、NeurIPS、ICLR、IEEE TPAMI 等国际顶级学术会议及期刊上发表论文超过 40 篇，学术成果累计被引用超过 10,000 余次。

研发经历

此前，他曾任上海人工智能实验室研究科学家，主导研发了 Vchitect——全球最早的大规模视频扩散生成系统之一。他还曾担任哔哩哔哩（Bilibili）视频生成团队负责人，带领团队开展交互式及全模态视频生成、以人物为中心的叙事生成，以及面向长视频生成的世界模型等方向的研究与开发。

开源成果

王耀晖博士主导及参与的代表性开源项目包括 Latte、LaVie、AnimateDiff 和 VBench，在 GitHub 上累计获得超过 17,000 颗 Star。其中，代表性成果 Latte 是全球首个基于 Diffusion Transformer（DiT）架构的开源视频生成模型，发布早于 OpenAI Sora。

学术认可

王耀晖博士曾入选上海市海外高层次人才，并获评斯坦福大学与爱思唯尔（Stanford–Elsevier）联合发布的“全球前 2% 顶尖科学家”榜单。

Research focus

Dr. Yaohui Wang completed his doctoral research at Inria in France and has worked in video generation since 2018. He is Co-Founder and Chief Scientist at Riemann Dynamics, where he leads research and development in large-scale world models for embodied intelligence.

Research impact

Dr. Wang has authored more than 40 papers at leading conferences and in journals including CVPR, NeurIPS, ICLR, and IEEE TPAMI. His research has received more than 10,000 citations.

Leadership experience

Previously, he was a Research Scientist at the Shanghai AI Laboratory, where he led the development of Vchitect, one of the early large-scale video diffusion generation systems. He also led the Video Generation Team at Bilibili, directing research and development in interactive and omnimodal video generation, human-centric narrative generation, and world models for long-form video generation.

Open-source work

Dr. Wang has led or contributed to open-source projects including Latte, LaVie, AnimateDiff, and VBench, which have collectively received more than 17,000 GitHub stars. Latte was among the early open-source video-generation models based on the Diffusion Transformer (DiT) architecture and was released before OpenAI introduced Sora.

Recognition

He has been recognized through Shanghai’s high-level overseas talent program and included in the Stanford–Elsevier Top 2% Scientists list.