Wufei Ma

I am a Research Scientist at Meta Superintelligence Labs, working on post-training of the Muse models with a focus on long-horizon agentic RL, synthetic data, and RL sandboxes. I have contributed to agentic post-training and evaluation of Muse Glimmer, Muse Spark, and Muse Wonder series models.

I received my Ph.D. in Computer Science from Johns Hopkins University in 2026, advised by Bloomberg Distinguished Professor Dr. Alan Yuille, where my research focused on multimodal LLMs, 3D spatial reasoning, and video understanding. Before that, I received my B.S. in Computer Science and Mathematics, summa cum laude, from Rensselaer Polytechnic Institute in 2020, where I was fortunate to work with Prof. Bülent Yener and Prof. Lirong Xia.

I’ve also spent great time at Google Research, Reality Labs, Research Asia, Frontier AI & Robotics (FAR) and AWS CV Science, Research, and collaborated with many exceptional researchers.

news

Oct 2026 The Muse Glimmer 30B technical report is out.
Jun 2026 Presented LychSim at the 3D-LLM/VLA Workshop at CVPR 2026.
May 2026 Organizing the 2nd Embodied Spatial Reasoning (ESR) Workshop at NeurIPS 2026.
Feb 2026 Joined Meta Superintelligence Labs, working on agentic RL.
Dec 2025 Presented SpatialReasoner at NeurIPS 2025.

selected publications

see all publications here

Muse Glimmer 30B: An On-Device Agentic Model

Meta Superintelligence Labs

arXiv preprint, 2026

LLM

arXiv
UnrealSpace: Analyzing Spatial Understanding and Reasoning in Controllable Simulation

Wufei Ma^*, Sky Cen^*, Jianzhi Shen^*, Rex Lee^*, León Begiristain, Yan Zhuang, Jiawei Peng , Zhifei Yu, Tianao Song, Xinyuan Qi, Tianmin Shu, Adam Kortylewski, and Alan Yuille

In Findings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2026

Dataset 3D Vision Vision-Lanugage
LychSim: A Controllable and Interactive Simulation Framework for Vision Research

Wufei Ma , Chloe Wang , Siyi Chen, Jiawei Peng , Patrick Li, and Alan Yuille

In 3D-LLM/VLA Workshop at CVPR , 2026

Dataset 3D Vision Vision-Lanugage

Abs Webpage arXiv Code

While self-supervised pretraining has reduced vision systems’ reliance on synthetic data, simulation remains an indispensable tool for closed-loop optimization and rigorous out-of-distribution (OOD) evaluation. However, modern simulation platforms often present steep technical barriers, requiring extensive expertise in computer graphics and game development. In this work, we present LychSim, a highly controllable and interactive simulation framework built upon Unreal Engine 5 to bridge this gap. LychSim is built around three key designs: (1) a streamlined Python API that abstracts away underlying engine complexities; (2) a procedural data pipeline capable of generating diverse, high-fidelity environments with varying out-of-distribution (OOD) visual challenges, paired with rich 2D and 3D ground truths; and (3) a native integration of the Model Context Protocol (MCP) that transforms the simulator into a dynamic, closed-loop playground for reasoning agentic LLMs.
SpatialReasoner: Towards Explicit and Generalizable 3D Spatial Reasoning

Wufei Ma, Yu-Cheng Chou, Qihao Liu , Xingrui Wang, Celso M de Melo, Jianwen Xie, and Alan Yuille

In Advances in Neural Information Processing Systems , 2025

3D Vision Vision-Lanugage

Abs Webpage arXiv

We introduce SpatialReasoner, a novel large vision-language model (LVLM) that address 3D spatial reasoning with explicit 3D representations shared between stages – 3D perception, computation, and reasoning. Explicit 3D representations provide a coherent interface that supports advanced 3D spatial reasoning and enable us to study the factual errors made by LVLMs.
3DSRBench: A Comprehensive 3D Spatial Reasoning Benchmark

Wufei Ma , Haoyu Chen, Guofeng Zhang, Jieneng Chen, Celso M de Melo, and Alan Yuille

In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , Oct 2025

3D Vision Vision-Lanugage

Abs Webpage arXiv

We present 3DSRBench, a comprehensive 3D spatial reasoning benchmark.
SpatialLLM: A Compound 3D-Informed Design towards Spatially-Intelligent Large Multimodal Models

Wufei Ma, Luoxin Ye, Celso M de Melo, Jieneng Chen, and Alan Yuille

In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , Oct 2025

(Highlight, 3.0%)

3D Vision Vision-Lanugage

Abs Webpage arXiv

We systematically study the impact of 3D-informed data, architecture, and training setups and present SpatialLLM, an LMM with advanced 3D spatial reasoning abilities.
ImageNet3D: Towards General-Purpose Object-Level 3D Understanding

Wufei Ma, Guofeng Zhang, Qihao Liu, Guanning Zeng, Adam Kortylewski , Yaoyao Liu, and Alan Yuille

In Advances in Neural Information Processing Systems , Oct 2024

Dataset 3D Vision

Abs Webpage arXiv Data Code

We present ImageNet3D, a large dataset for general-purpose object-level 3D understanding.
Rethinking Video-Text Understanding: Retrieval from Counterfactually Augmented Data

Wufei Ma, Kai Li, Zhongshi Jiang, Moustafa Meshry, Qihao Liu, Huiyu Wang, Christian Häne, and Alan Yuille

In European Conference on Computer Vision , Oct 2024

(Strong Double Blind)

Dataset Vision-Lanugage

Abs Webpage arXiv Data

We propose a novel task, retrieval from counterfacually augmented data, and a dataset, Feint6K, for video-text understanding.
Generating Images with 3D Annotations Using Diffusion Models

Wufei Ma^*, Qihao Liu^* , Jiahao Wang^* , Angtian Wang, Xiaoding Yuan , Yi Zhang, Zihao Xiao, Guofeng Zhang , Beijia Lu, Ruxiao Duan, Yongrui Qi, Adam Kortylewski , Yaoyao Liu, and Alan Yuille

In The Twelfth International Conference on Learning Representations , Oct 2024

(Spotlight, 5%)

Dataset 3D Vision

Abs arXiv Code

We propose 3D-DST that generates synthetic data with 3D groundtruth by incorporating 3D geomeotry control into diffusion models. With our diverse prompt generation, we effectively improve both in-distribution (ID) and out-of-distribution (OOD) performance for various 2D and 3D vision tasks.