Jionghao Wang

Currently looking for full-time/internship opportunities.

I am a 3rd-year Ph.D. student at Texas A&M University, advised by Prof. Wenping Wang and Prof. Xin Li. I also work closely with Prof. Yue Wang. My goal is to bring intelligence to the real world through Robot Foundation Models, Physics-based Motion, and 3D Real-to-Sim.

Before that, I received my B.Eng. and M.Eng. from Shanghai Jiao Tong University with Prof. Li Song. People call me John.

Email Scholar GitHub Twitter LinkedIn

News

NOV 2026 Released Flex-RSI, a benchmark for recursive self-improvement of frontier agents! 🤖
OCT 2026 DiTaR is out on arXiv: learning dense robot rewards from action-free videos! 📄
MAY 2026 Started as a research intern at WorldEngine AI! 🌍
AUG 2025 Presented PDT at SIGGRAPH 2025 in Vancouver, Canada! 🇨🇦
AUG 2025 Wrapped up my amazing time with Sony! 🎮
MAY 2025 Started as a research intern at Sony in San Jose, CA! 🎮
OCT 2024 Presented SO-SMPL at ECCV 2024 in Milan, Italy! 🇮🇹
AUG 2024 Started as a Ph.D. student at Texas A&M! Gig 'em! 🎓
MAR 2024 Obtained my Master's degree from SJTU! Grateful for the amazing time! 🎓
FEB 2024 Started as a research intern at miHoYo in Shanghai! ✨

Publications (highlighted research is in pale green; * equal contribution)

Flex-RSI: Benchmarking Recursive Self-Improvement of Frontier Agents

Jionghao Wang*, Jingdong Zhang*, Zhengming Yu, et al.

Tech Report 2026 / Project/ Blog/ GitHub/ Leaderboard

A budget-aware benchmark where agents rewrite their own harness between rounds to improve at a task. We ran 18 frontier models on 24 vision and robotics tasks to see where self-improvement pays off.

Steerable Humanoid Loco-Manipulation VLA

Jionghao Wang* (co-first author), et al.

In Submission

A humanoid policy steered by future goal images: a predictor turns subtask prompts into goal images, and a planner drives whole-body actions toward them across multi-task routines.

DiTaR: Disentangling Task-Relevant Latent Dynamics for Reward Modeling

Dayou Li*, Jionghao Wang*, Peihao Wang, Jiuzhou Lei, Hao Wang, Jichen Chai, Hezhen Hu, Minghui Zheng, Zhangyang Wang, Xin Li, Wenping Wang, Zhiwen Fan

arXiv 2026 / Paper/ Project/ GitHub

We learn dense rewards from action-free videos by modeling task-relevant latent dynamics, so the reward tracks task progress rather than visual distractors such as lighting.

Spatial-UMI-Dream: Scaling World Action Models with Metric Geometry for UMI Manipulation

Weiduo Yuan*, Ze Yuan*, Sihan Liu*, Jionghao Wang, Liang Heng, Zhenyu Zhao, Yue Wang

arXiv 2026 / Paper/ Project/ GitHub

A 5B spatial World Action Model pretrained on 605 hours of RGB-D UMI data to jointly predict future RGB, metric depth and actions, evaluated on RoboTwin-UMI, a twenty-task benchmark for UMI policy generalization.

Learning What Matters: Adaptive Information Theoretic Objectives for Robot Exploration

Youwei Yu, Jionghao Wang, Zhengming Yu, Wenping Wang, Lantao Liu

RSS 2026/ Paper/ Project/ GitHub

Developed QOED to enhance robot exploration by isolating identifiable model parameters and suppressing unidentifiable 'nuisance' directions in high-dimensional systems.

PDT: Point Distribution Transformation with Diffusion Models

Jionghao Wang*, Cheng Lin*, Yuan Liu, Rui Xu, Zhiyang Dou, Xiao-Xiao Long, Hao-Xiang Guo, Taku Komura, Wenping Wang, Xin Li

SIGGRAPH 2025 / Paper/ Project/ GitHub/ Twitter Post

PDT is a novel diffusion-based framework for transforming dense point distribution into a target distribution that is semantically meaningful.

Disentangled Clothed Avatar Generation from Text Descriptions

Jionghao Wang*, Yuan Liu*, Zhiyang Dou, Zhengming Yu, Yongqing Liang, Cheng Lin, Rong Xie, Li Song, Xin Li, Wenping Wang

ECCV 2024 / Paper/ Project/ GitHub

In this paper, we introduced a novel text-to-avatar generation method that separately generates the human body and the clothes and allows high-quality animation on the generated avatar.

360-Degree Panorama Generation from Few Unregistered NFoV Images

Jionghao Wang*, Ziyu Chen*, Jun Ling, Rong Xie, Li Song

ACM Multimedia 2023 / Paper/ GitHub/ Video

We proposed a novel diffusion-based pipeline that generates complete 360 panoramas using one or more unregistered NFoV images captured from arbitrary angles.