EngineeringAI Lab
Projects
The four project groups below correspond to the lab's shared research directions. Each group expands into concrete papers and code status.
Selected Projects
Compact and controllable 3D asset representations for reconstruction, editing, decomposition, and state prediction.
A compact 3D mesh tokenization project for efficient VAE reconstruction and foundation representation learning.
A topology-aware prior for decomposing 3D meshes into meaningful parts for structure-aware manipulation.
Agentic pipelines and engine-grounded tools for coherent 3D scenes, simulation data, and spatial evaluation.
EngineeringAI Lab's first team-led ECCV paper, building editable multi-shot 3D story worlds with spatial-temporal consistency.
A world visual attention framework for viewpoint selection and camera trajectories in dynamic 3D story worlds.
Expressive avatars, human-scene interaction, and multi-person motion generation for dynamic digital worlds.
A controllable portrait animation framework for identity-preserving, expressive, and synchronized talking faces.
A 3D talking-avatar system unifying identity, lip sync, emotion, and spatial dynamics.
A multi-person motion generation project focused on social coherence, interpersonal relations, and plausible dynamic interaction.
LLM-based reasoning, language-guided planning, reinforcement learning, and decision systems for agents in 3D worlds.
A scalable in-context decision learning framework for adapting policies from task context.
Offline meta-reinforcement learning from natural language supervision for decision-making agents.
Evolutionary decoding for mathematical reasoning in diffusion LLMs.