Papers
12
Total Citations
72
H-Index
5
About
Haonan Chang is a robotics researcher whose work spans robot perception, manipulation, and scene understanding, with a particular focus on solving real-world challenges that hinder the deployment of autonomous systems. He is perhaps best recognized for his pioneering contributions to transparent object manipulation, most notably through the GlassLoc framework, which leverages plenoptic imaging to detect grasp poses amid transparent clutter — a notoriously difficult problem that has accumulated nearly 30 citations across related venues. His scene estimation work, including GeoFusion and a novel object-prior-free tracking and reconstruction system, demonstrates a sustained commitment to building robust semantic maps in cluttered, occluded environments. More recently, Chang has extended his research into language-guided robotics and foundation model integration, developing systems such as LGMCTS for natural-language-driven object rearrangement, OVIR-3D for open-vocabulary 3D instance retrieval, and UniAff for unified affordance representation using vision-language models. His 2025 work on autoregressive action sequence learning reflects growing ambitions toward universal robot policy architectures. Collectively, Chang's research addresses the full manipulation pipeline — from perception and scene modeling to planning and control — making him a versatile and impactful contributor to modern robotics research.
Research Focus
Key Achievements
Top Papers
- 1GlassLoc: Plenoptic Grasp Pose Detection in Transparent Clutter22 citations · 2019
- 2GeoFusion: Geometric Consistency Informed Scene Estimation in Dense Clutter11 citations · 2020
- 3Autoregressive Action Sequence Learning for Robotic Manipulation8 citations · 2025
- 4
- 5
- 6GlassLoc: Plenoptic Grasp Pose Detection in Transparent Clutter5 citations · 2019
- 7Scene-level Tracking and Reconstruction without Object Priors4 citations · 2022
- 8
- 9
- 10