Linxi Fan

Also known as Jim Fan

Linxi Fan, who goes by Jim, is an NVIDIA researcher who builds generalist AI agents for video games and robots. He graduated as Columbia University's valedictorian in 2016, interned at Baidu in 2015 and at OpenAI in 2016, and completed a PhD at Stanford Vision Lab under Fei-Fei Li in 2021. At NVIDIA, which he joined in December 2021, he led the MineDojo project, whose paper won a NeurIPS 2022 Outstanding Paper Award, and in 2023 he co-authored the Minecraft agent Voyager and the reward-design system Eureka. In 2025, he co-authored GR00T N1, an open foundation model for humanoid robots. He co-leads NVIDIA's Generalist Embodied Agent Research group with Yuke Zhu.

Now at: NVIDIA
Badges: Bridge

Timeline

2013–2016

Valedictorian at Columbia

He works as a research assistant in Columbia's NLP Group under Michael Collins from 2013 to 2014, studying kernel methods for speech recognition. He graduates from Columbia University in 2016 as valedictorian of the class and receives the Illig Medal.
2015–2016

DeepSpeech 2 and World of Bits

As an intern at Baidu Silicon Valley AI Lab in 2015, he co-develops DeepSpeech 2, an end-to-end speech recognition system for English and Chinese. In 2016, he is OpenAI's first intern and co-designs World of Bits, a platform for teaching AI to use a web browser.
2016–2021

PhD at Stanford Vision Lab

He studies for a PhD at Stanford Vision Lab from 2016 to 2021, advised by Fei-Fei Li, and writes the thesis Training and Deploying Visual Agents at Scale. During a 2018 internship at Google Cloud AI, he creates SURREAL, an open-source distributed reinforcement learning framework for robot learning.
December 2021

Research scientist at NVIDIA

He joins NVIDIA as a research scientist in December 2021 to work on foundation models for general-purpose autonomous agents. He leads the MineDojo effort for open-ended agent learning in Minecraft.
2022

MineDojo wins at NeurIPS

In 2022, he is first author of MineDojo, a Minecraft framework with thousands of open-ended tasks and an internet-scale knowledge base of videos, tutorials, wiki pages and forum discussions. It wins an Outstanding Paper Award in the Datasets and Benchmarks track at NeurIPS 2022.
2023

Voyager and Eureka

In 2023, he co-authors Voyager, a Minecraft agent that explores continuously, acquires skills and makes discoveries using GPT-4, an effort that grows out of MineDojo. He also co-authors Eureka, in which large language models such as GPT-4 write reward functions for reinforcement learning and outperform human experts on 83% of the tasks in a suite of 29 environments.
2024–2025

GR00T for humanoid robots

By 2024, he leads NVIDIA's Generalist Embodied Agent Research (GEAR) team, whose flagship project GR00T aims at foundation models for humanoid robots. In 2025, he is a coauthor, with Yuke Zhu and the NVIDIA GEAR team, of GR00T N1, an open foundation model with a dual-system vision-language-action design.
As of October 2026

Co-leading NVIDIA's GEAR group

He co-leads NVIDIA's Generalist Embodied Agent Research (GEAR) group with Yuke Zhu. The group's mission is to build foundation models for embodied agents in virtual and physical worlds, and in 2026 he coauthors HumanoidMimicGen, a data-generation method for loco-manipulation via whole-body planning.