LifePlanner: Evaluating LLM Agents for Geo-spatial Planning with Social Media Data
A benchmark that connects maps with local social-media evidence to evaluate LLM agents across four geo-spatial planning tasks and three difficulty levels.
Ways of seeing
Exploring how machines see, understand, and interact with our world.
A different perspective is one click away.
A room. More than meets the eye.Scroll to discover
Info
I study how machines perceive
and understand the world in 3D.
I am a master’s student at LIESMARS, Wuhan University , supervised here by Zhen Dong and Bisheng Yang , and co-supervised by Yuan Liu and Haiping Wang at HKUST .
My research spans 3D scene understanding, reconstruction, and large language models. I am part of WHU-USI3DV .
Education
2024 — present
M.S. · Photogrammetry & Remote Sensing
Wuhan University
2020 — 2024
B.S. · Wuhan University
Recognition
- 2025Excellent Academic Performance Scholarship, Wuhan University
- 2025Excellence Award for Graduate Students, Wuhan University
- 2023National Scholarship for Undergraduates · Top 3%
- 2023Innova Student Innovation Achievement Award, Wuhan University
- 2022First Prize, China Software Cup · Rank 1
News
Works
From observations
to understanding.
The Neural City: A Next-generation Spatio-Temporal Intelligence Paradigm for Urban Holistic Governance
A framework connecting raw urban observations with holistic urban governance through the “6W+4R” paradigm.
Fine-Resolution Forest Height Estimation by Integrating ICESat-2 and Landsat 8 OLI Data with a Spatial Downscaling Method for Aboveground Biomass Quantification
Mapping forest height and aboveground biomass at 15-m resolution by combining satellite imagery and spaceborne LiDAR.
* Equal contribution · † Corresponding author
ByteDance
Research Intern · Hangzhou, China
Developed a Reinforcement Learning post-training pipeline for autoregressive 3D mesh generation models, designing reward functions to systematically optimize the geometric fidelity, topological correctness, and visual aesthetics of the generated meshes.
Built a scalable pre-training data curation pipeline by developing geometry-based filtering operators for targeted failure modes and a VLM-based auto-annotation framework through instruction tuning and prompt engineering.
Arts
Another way
of seeing.
Between the shape of the land
and the traces we leave.
Three ways of lookingSelected photographs
Two passing seasonsA different light. A different hour.
Photographs by Yuning Peng Light, space, and the intervals between.