Haoyu Han
Haoyu Han is a Harvard computational robotics researcher associated with the XM reconstruction project. With Heng Yang, he coauthored Building Rome with Convex Optimization in 2025. The method uses learned depth to form a scaled bundle-adjustment problem and solves a convex relaxation with a GPU-based optimizer. The project combines these steps in a structure-from-motion pipeline and presents reconstruction demonstrations. It received the RSS 2025 Outstanding Systems Paper award.
- Now at
- Harvard University
- Works
- 46
- Citations
- 545
- h-index
- 12
Publication numbers from OpenAlex, as of 14 October 2026.
Timeline
-
February 2025
A reconstruction method
Han and Heng Yang submitted Building Rome with Convex Optimization. The method starts with image keypoints and learned depth estimates. It turns them into a scaled bundle-adjustment problem, which is part of recovering a scene and its camera positions.
-
2025
Optimization with a certificate
The researchers designed a convex relaxation of the problem and a CUDA-based optimizer called XM. The paper describes solving the scaled formulation to certifiable global optimality in its tested setting.
-
2025
Showing complete scene reconstructions
XM powers a structure-from-motion pipeline that demonstrates reconstructions from multiple datasets and uses the camera estimates in three-dimensional rendering. Han shares the work with Yang.
-
2025
Outstanding Systems Paper Award
RSS awarded the Outstanding Systems Paper Award to Han and Yang for Building Rome with Convex Optimization.
Sources and credits
Photo: hyhan0118.github.io; Haoyu Han; creator not stated. License: Not stated by source. Source: hyhan0118.github.io. Cropped to a square and resized. Ask us to remove or replace this photo.