ExFace
ExFace: Expressive Facial Control for Humanoid Robots with Diffusion Transformers and Bootstrap Training
This paper presents a novel Expressive Facial Control (ExFace) method based on Diffusion Transformers, which achieves precise mapping from human facial blendshapes to bionic robot motor control.
Method
ExFace maps human facial blendshapes to bionic robot motor control with a Diffusion Transformer. A model bootstrap training strategy improves the quality, accuracy, and temporal smoothness of the generated facial expressions.
Data Collection
The ExFace dataset is driven by synchronized human facial data and robot motor states, providing paired control signals for expressive facial motion generation.
Robot Performance
Real-time and movie-driven applications transfer a performer's facial motion to different humanoid robots. These demonstrations evaluate both generalizability across platforms and the naturalness of expression rendering in performance and human-robot interaction.
Evaluation
Experimental results show improvements over previous methods in motor-control accuracy, frames per second, and response time.
