Yongyuan Cheryl Liang

I am a PhD student at UMD CS and was fortunate to work with Jianwei Yang and Huazhe Xu. I received my B.S. degree in Mathematics from Sun Yat-sen University.

I have interned and conducted research at NVIDIA, Adobe, and Microsoft Research.

I'm currently on the full-time job market.

Research Highlight

I study how to build large agentic models that can generalize across tasks and environments while interacting with the virtual and physical world over time. My research focuses on three capabilities needed to move beyond one-shot perception-to-action mappings toward adaptive, closed-loop agents:

Memory Maintain and update task-relevant information across interactions, viewpoints, and open-ended environments.

Interaction memory: TraceVLA; Spatial memory: MomaGraph, SAW-Bench, Spatial Memory in Frontier Models

Reasoning Learn a structured understanding of the world and reason prospectively about the consequences of actions.

Cross-model reasoning: ROVER; With 3D modality: Lemon; As training objectives: Magma

Planning Compose decisions over long horizons and adapt them as new observations and execution feedback emerge.

Action planning: MomaGraph; Task planning: Plan with Vibe Agents

Together, these capabilities are deeply intertwined within a shared generalist agentic model, where memory, reasoning, and planning jointly shape decisions that are grounded through flexible action interfaces, while interaction feedback continuously updates all three.

Research overview: generalist agentic models interacting with the physical world

Aug' 26	We drop DanceOPD and DiffusionOPSD.
May' 26	SAW-Bench won the Best Paper Award Runner-Up at CVPR 2026 WMAS.
Apr' 26	SAW-Bench to appear in ICML 2026 as Spotlight (2%).
Apr' 26	Drop a new blog post about Vibe Agents Step Into the Real World.
Mar' 26	We release a new blog post about Spatial Memory in Frontier Models.
Feb' 26	Three papers to appear in CVPR 2026 (2 main track and 1 findings).
Jan' 26	MomaGraph selected as Oral presentation (1%) in ICLR 2026.
Jan' 26	One papers to appear in ICRA 2026.
Jan' 26	Two papers to appear in ICLR 2026 (ROVER and MomaGraph).
Sept' 25	One paper to appear in NeurIPS 2025.
Feb' 25	Magma to appear in CVPR 2025.
Jan' 25	Two papers to appear in ICLR 2025.
Jan' 25	Start to update Awesome-Generalist-Agents.
Sept' 24	Make-An-Agent to appear in NeurIPS 2024.
June' 24	ACE selected as Oral presentation (1%) in ICML 2024.
May' 24	Two papers to appear in ICML 2024.
Jan' 24	Three papers to appear in ICLR 2024, including two spotlights and one poster.