PeopleofSuperintelligence

Researcher

Kevin Xie

Also known as Cheng Xie

Kevin Xie is a research scientist at NVIDIA who builds generative models of 3D objects, human motion and simulated worlds for autonomous physical agents. He studied Engineering Physics at the University of British Columbia, where he coauthored the PLAID skill-distillation paper with Michiel van de Panne for ICLR 2018, and did his PhD at the University of Toronto and the Vector Institute, co-supervised by Florian Shkurti and Sanja Fidler. As first author of a 2021 ICCV paper, he used differentiable physics to correct human poses estimated from monocular video. In 2024, he was co-first author of LATTE3D, which turns text into textured 3D meshes in 400 milliseconds, and coauthored the L4GM model for animated 4D objects. His recent work centers on world models, including NVIDIA's Cosmos-Transfer1 in 2025 and the OmniDreams world model for autonomous vehicle simulation in 2026.

Now at
NVIDIA
Works
22
Citations
341
h-index
8
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2018

    Skill distillation at UBC

    He studies Engineering Physics at the University of British Columbia, where he works with Michiel van de Panne. With Glen Berseth, Paul Cernek and van de Panne, he coauthors PLAID at ICLR 2018, a method that merges bipedal locomotion skills by extending policy distillation to continuous actions.

  2. 2020–2021

    Planning with learned models

    His doctoral work in computer science at the University of Toronto and the Vector Institute is co-supervised by Florian Shkurti and Sanja Fidler. In April 2020, he, Homanga Bharadhwaj and Shkurti post a model-predictive control method that interleaves the cross-entropy method with gradient descent steps. Latent Skill Planning, posted in November 2020 and published at ICLR 2021, joins a learned world model with skills chosen by online planning.

  3. 2021

    Physics-corrected human motion

    In September 2021, he posts Physics-based Human Motion Estimation and Synthesis from Videos as first author, with Tingwu Wang, Umar Iqbal, Yunrong Guo, Sanja Fidler and Florian Shkurti of the University of Toronto and NVIDIA. The ICCV 2021 paper corrects image-based pose estimates with differentiable physics constraints and trains generative models of human motion from monocular RGB video.

  4. 2023

    ATT3D for text-to-3D

    He coauthors ATT3D with Jonathan Lorraine, James Lucas, Sanja Fidler and colleagues at NVIDIA, the University of Toronto and the Vector Institute, presented at ICCV 2023. The method trains one model on many text prompts at once, sharing computation across the prompt set and generalizing to unseen prompts.

  5. 2024

    LATTE3D in 400 milliseconds

    As co-first author with Jonathan Lorraine and Tianshi Cao, he posts LATTE3D in March 2024. The model produces detailed textured 3D meshes from text in a single forward pass, generating objects in 400 ms, and appears at ECCV 2024.

  6. 2024

    L4GM for animated 4D objects

    He is second author, after Jiawei Ren, of L4GM, a NeurIPS 2024 paper. Billed as the first 4D large reconstruction model, it turns a single-view video into an animated object in one feed-forward pass that takes about a second.

  7. 2025

    VideoPanda and Cosmos-Transfer1

    In April 2025, he posts VideoPanda as first author; it adds multi-view attention to a video diffusion model to synthesize 360-degree videos from text or a single-view video. He is also among the authors of NVIDIA's Cosmos-Transfer1, an open-source conditional world generation model posted in March 2025 for robotics Sim2Real and autonomous vehicle data enrichment.

  8. As of October 2026

    World models at NVIDIA

    He is a research scientist at NVIDIA working on generative world models for autonomous physical agents. In NVIDIA's DAIR research group, his research covers learning-based simulation and control, including generative modeling, model-based reinforcement learning and character control. In June 2026, he is among the authors of NVIDIA OmniDreams, a real-time generative world model for closed-loop autonomous vehicle simulation trained on 21k hours of driving scenarios.

Career

  1. Since 2020NVIDIAResearch scientist

Education

  1. 2021University of TorontoPhD
  2. 2021Vector InstitutePhD
  3. 2018University of British Columbia

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Physics-based Human Motion Estimation and Synthesis from Videos2021 · IEEE/CVF International Conference on Computer Vision (ICCV) · 87 citationsWith Sanja Fidler
  2. ATT3D: Amortized Text-to-3D Object Synthesis2023 · IEEE/CVF International Conference on Computer Vision (ICCV) · 50 citationsWith Ming-Yu Liu, Sanja Fidler and Tsung-Yi Lin
  3. gradSim: Differentiable simulation for system identification and visuomotor control2021 · 49 citationsWith Miles Macklin and Sanja Fidler
  4. KAMA: 3D Keypoint Aware Body Mesh Articulation2021 · International Conference on 3D Vision (3DV) · 34 citations
  5. L4GM: Large 4D Gaussian Reconstruction Model2024 · neural information processing systems · 30 citationsWith Antonio Torralba and Sanja Fidler

From OpenAlex, as of 21 October 2026.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Kevin Xie. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. Google has the highest combined h-index among AI organizationsIncludes NVIDIA. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes NVIDIA. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes NVIDIA. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Kevin Xie yet. Send us a free-licensed photo.

  1. Kevin Xie kevincxie.github.io. Checked 20 October 2026.
  2. Kevin Xie | DAIR @ NVIDIA Research research.nvidia.com. Checked 20 October 2026.
  3. [2004.08763] Model-Predictive Control via Cross-Entropy and Gradient-Based Optimization arxiv.org. Checked 20 October 2026.
  4. [2011.13897] Latent Skill Planning for Exploration and Transfer arxiv.org. Checked 20 October 2026.
  5. [2109.09913] Physics-based Human Motion Estimation and Synthesis from Videos arxiv.org. Checked 20 October 2026.
  6. Physics Human Motion research.nvidia.com. Checked 20 October 2026.
  7. ATT3D: Amortized Text-to-3D Object Synthesis research.nvidia.com. Checked 20 October 2026.
  8. LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesis research.nvidia.com. Checked 20 October 2026.
  9. [2403.15385] LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesis arxiv.org. Checked 20 October 2026.
  10. L4GM: Large 4D Gaussian Reconstruction Model research.nvidia.com. Checked 20 October 2026.
  11. [2504.11389] VideoPanda: Video Panoramic Diffusion with Multi-view Attention arxiv.org. Checked 20 October 2026.
  12. [2503.14492] Cosmos-Transfer1: Conditional World Generation with Adaptive Multimodal Control arxiv.org. Checked 20 October 2026.
  13. [2606.03159] NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation arxiv.org. Checked 20 October 2026.