Jim Fan

Jim Fan (also known in academic publications as Linxi Fan) is a leading researcher in artificial intelligence, robotics, and embodied intelligence. He currently serves as Director of AI and Distinguished Research Scientist at NVIDIA, where he is responsible for advancing research in Physical AI and general-purpose robotics. He is the co-lead of Project GR00T, NVIDIA’s major initiative to build foundation models for humanoid robots, and a co-founder and leader of the GEAR Lab (Generalist Embodied Agent Research). His work focuses on solving Physical AGI by enabling machines to learn, reason, and act in both real and simulated environments.

Before his current role, Jim Fan was Senior Manager and Principal Research Scientist at NVIDIA (2024–2025), where he founded the GEAR research team. Earlier, he served as Senior Research Scientist (2023–2024) and Research Scientist (2021–2023), during which he led major projects including Voyager, MineDojo, Eureka, VIMA, Prismer, and World of Bits. His paper “MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge” won the NeurIPS Best Paper Award and later the Outstanding Paper Award at NeurIPS 2022. His research on AI agents and robotics has been widely covered by global media including The New York Times, Forbes, MIT Technology Review, Wired, TechCrunch, and VentureBeat.

Jim Fan earned his PhD in Computer Science from Stanford University (2016–2021), advised by Professor Fei-Fei Li, focusing on deep reinforcement learning, robotics, computer vision, and large-scale systems. He completed his Bachelor’s degree in Computer Science at Columbia University (2012–2016) as Valedictorian, graduating with a perfect GPA and receiving the Illig Prize.

He has held research internships at OpenAI, NVIDIA, Baidu AI Labs, Mila (Quebec AI Institute), Google Cloud AI, and Stanford University, working closely with researchers including Andrew Ng, Yoshua Bengio, Aaron Courville, Ilya Sutskever, Andrej Karpathy, and Dario Amodei. Notably, he was OpenAI’s first intern in 2016 and co-developed the OpenAI Universe Initiative.

Jim Fan, also known in academic publications as Linxi Fan, is one of the leading researchers working on artificial intelligence, robotics, and embodied intelligence. His work focuses on building generally capable AI agents that can operate in physical environments such as robots and in virtual environments such as games and simulations. He currently serves as Director of AI and Distinguished Research Scientist at NVIDIA, where he leads major research efforts to advance Physical AI and general-purpose robotics. He is a co-lead of Project GR00T, NVIDIA’s large-scale initiative to develop foundation models for humanoid robots, and a co-founder and leader of the GEAR Lab, which stands for Generalist Embodied Agent Research. His long-term mission is to make Physical AI a reality by enabling machines to learn, reason, and act in the real world.

Jim Fan began his academic journey at Columbia University, where he studied Computer Science from 2012 to 2016. During his undergraduate studies he demonstrated strong academic performance, graduating as Valedictorian of the Class of 2016 with a perfect grade point average. He also received the Illig Prize, one of the highest academic honours awarded by Columbia University. While at Columbia, he carried out research work in computer vision and natural language processing, serving as a research assistant under the supervision of Professor Shree Nayar, Professor Daniel Hsu, Professor Michael Collins, Professor Venkat Venkatasubramanian, and IBM senior researcher Brian Kingsbury. He contributed to projects related to computer vision, chemical engineering knowledge systems, and speech recognition, including work on the IARPA-funded Babel project.

During this period, he also completed several important research internships. In 2014 he worked on the Stanford Autonomous Driving Project under Professor Andrew Ng. In 2015 he joined Baidu USA as a research intern, collaborating with Andrew Ng and Dario Amodei on the DeepSpeech large-scale end-to-end speech recognition system for both English and Mandarin. From 2015 to 2016 he worked at the Mila – Quebec Artificial Intelligence Institute in Montreal as a research assistant, advised by Yoshua Bengio and Aaron Courville on deep semi-supervised learning.

In 2016, Jim Fan joined OpenAI as a research intern. He was the first intern in the organisation’s history. During this time, he co-developed the OpenAI Universe Initiative and co-authored World of Bits, published at ICML 2017. This work produced one of the first AI agents capable of controlling a web browser using pixel input and keyboard and mouse actions, years before large language models became the dominant research focus.

After completing his undergraduate degree, Jim Fan began his doctoral studies at Stanford University in 2016. He pursued a PhD in Computer Science at the Stanford Vision Lab under the supervision of Professor Fei-Fei Li. His doctoral research focused on deep reinforcement learning, robotics, computer vision, and large-scale distributed learning systems. Throughout his PhD, he continued to collaborate with industry and academic research groups, including a research internship at NVIDIA in 2020 where he worked on deep reinforcement learning.

He completed his PhD in 2021 with a perfect academic record and soon after joined NVIDIA as a Research Scientist. From 2021 to 2023, his work focused on building embodied general intelligence. During this period, he led the development of MineDojo, a large-scale platform for open-ended agent learning that uses internet-scale knowledge. The MineDojo paper won the Best Paper Award and later the Outstanding Paper Award at NeurIPS 2022, becoming one of the most influential works in embodied AI research.

From 2023 to 2024, Jim Fan served as Senior Research Scientist at NVIDIA, where he led several major projects including Voyager, Eureka, Prismer, VIMA, and MineDojo. Voyager became known as the first AI agent capable of playing Minecraft at a high level while continuously improving its own abilities. Eureka demonstrated highly precise robotic manipulation, including a five-finger robot hand performing complex tasks such as pen spinning. VIMA became one of the earliest multimodal foundation models for robot manipulation.

In 2024, he co-founded the GEAR research team at NVIDIA and was appointed Senior Manager and Principal Research Scientist. The team’s goal is to build what they describe as the Foundation Agent, a generally capable AI that can act skillfully across many different physical and virtual worlds. He led research programmes covering robotics, gaming AI, and simulation, working towards a future where autonomous machines and intelligent agents become widely deployed.

In 2025, Jim Fan was appointed Director and Distinguished Research Scientist at NVIDIA. In this role he spearheads Project GR00T, focusing on foundation models and techniques for general-purpose humanoid robotics. He co-leads the GEAR Lab and continues to guide major research directions in Physical AI, robotics, multimodal learning, and embodied intelligence. His recent publications include NVIDIA Isaac GR00T N1, an open foundation model designed for humanoid robots.

Alongside his research, Jim Fan actively shares insights on AI development and industry progress through public lectures, academic publications, and social media. His work has been featured by global media including The New York Times, Forbes, MIT Technology Review, Wired, TechCrunch, and VentureBeat. He continues to publish under the name Linxi Fan in academic journals and conferences and remains closely involved in shaping the next generation of intelligent machines that operate in both the physical and digital world.

Jim Fan’s vision is to build generally capable artificial intelligence that can operate reliably in the physical world as well as in digital environments. He believes that the next major step in AI development is the creation of Physical AI, where machines understand their surroundings, learn from experience, and perform useful tasks alongside humans. His long-term goal is to develop foundation models for robotics that allow machines to move, perceive, reason, and act safely and effectively in real-world settings. Through projects such as GR00T and the GEAR Lab, he works towards a future where intelligent robots and autonomous systems become practical tools across industries and everyday life.

Jim Fan has received several important academic and research honours for his contributions to artificial intelligence and robotics. His work MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge won the Outstanding Paper Award at NeurIPS 2022, one of the most respected conferences in AI research. Earlier in his academic career, he graduated as Valedictorian of the Class of 2016 at Columbia University with a perfect academic record and received the Illig Prize, awarded for outstanding academic achievement. His research has been widely recognised and featured in major global publications including The New York Times, Forbes, MIT Technology Review, Wired, TechCrunch, and VentureBeat.