Humanoid Imitation of Facial Expressions through Deep Learning

Supervisor: Zhiwu Huang · Second Examiner: David Thomas

A fine-grained emotion labelling framework that maps human facial motion to humanoid robot control and classifies each frame into 22 nuanced emotion categories.

Keep a straight face when it starts — the first few frames lock your "neutral" expression as the baseline. If you launch mid-yawn, the robot will think that's your resting face.

R Recalibrate P Pause +- Smoothing Q Quit

Install & Run downloads a one-click script (~1.3 GB weights). Requires Python 3.10+, a CUDA GPU, and a webcam.

Three-Stage Pipeline

A

Motion Transfer

LivePortrait transfers human facial motion onto a robot source image via keypoint warping and SPADE generation.

B

X2Control Regression

A ResNet-18 mapping network converts each synthesised frame into a 30-dimensional servo control vector.

C

Emotion Classification

Control vectors are mapped to AU intensities, then classified into 22 fine-grained emotions via a hybrid classifier.

22-Class Fine-Grained Emotion Taxonomy

A hierarchical taxonomy grounded in FACS Action Units, extending beyond the 7 basic emotions to capture nuanced compound expressions.

Action Units (AUs) are the building blocks of facial expressions defined by the Facial Action Coding System (FACS). Each AU corresponds to a specific facial muscle movement — for example, AU12 is the lip corner pull that produces a smile. Every emotion below is defined by its unique combination of AUs.

Tier 1 Basic Emotions (7)

Neutral
No AU activation
Happy
AU6 + AU12 + AU25
Cheek raise, lip corner pull, lips part
Sad
AU1 + AU4 + AU15
Inner brow raise, brow lower, lip corner depress
Fearful
AU1 + AU4 + AU20 + AU25
Brow raise, brow lower, lip stretch, lips part
Angry
AU4 + AU7 + AU24
Brow lower, lid tightener, lip press
Surprised
AU1 + AU2 + AU25 + AU26
Brow raise, outer brow raise, lips part, jaw drop
Disgusted
AU9 + AU10 + AU17
Nose wrinkle, upper lip raise, chin raise

Tier 2 Compound Emotions (10)

Combinations of two basic emotions, exhibiting AU patterns from both constituent emotions simultaneously.

Sadly Fearful
AU1 + AU4 + AU15 + AU20
Sadly Angry
AU1 + AU4 + AU7 + AU15
Sadly Surprised
AU1 + AU4 + AU15 + AU25
Sadly Disgusted
AU4 + AU9 + AU10 + AU15
Fearfully Angry
AU1 + AU4 + AU7 + AU20
Fearfully Surprised
AU1 + AU2 + AU4 + AU20 + AU25
Fearfully Disgusted
AU1 + AU4 + AU9 + AU20
Angry Surprised
AU4 + AU7 + AU25 + AU26
Angrily Disgusted
AU4 + AU7 + AU9 + AU10
Disgust Surprised
AU1 + AU2 + AU9 + AU25

Tier 3 Additional States (5)

Complex cognitive-emotional states that go beyond simple combinations.

Awed
AU1 + AU2 + AU5 + AU25
Wide eyes, raised brows, open mouth
Appalled
AU4 + AU9 + AU10
Lowered brows, nose wrinkle, shocked disgust
Hatred
AU4 + AU7 + AU10
Brow lower, lid tighten, lip raise (contempt)
Confused
AU1 + AU2 + AU4 + AU7
Mixed brow signals, asymmetric tightening
Skeptical
AU4 + AU14 + AU7
Brow lower, dimpler, lid tighten

Acknowledgements

This project builds upon the X2CNet framework and the X2C dataset [1], which provides the three-stage pipeline for transferring human facial expressions to humanoid robots via 30-dimensional control vectors. Our fine-grained emotion taxonomy extends this pipeline with a semantic interpretation layer.

The emotion classification system is grounded in the Facial Action Coding System (FACS) [3], which defines the Action Unit vocabulary used to characterise each of our 22 emotion categories. The AU-based pattern matching stream of our hybrid classifier follows the conventions established by EmotioNet [2], which demonstrated that compound facial expressions can be reliably annotated through AU activation patterns at scale.

References

[1]
P. Li, L. Cao, X.-M. Wu, R. Yang, and X. Yu, “X2C: A Dataset Featuring Nuanced Facial Expressions for Realistic Humanoid Imitation,” arXiv preprint arXiv:2505.11146, 2025. arXiv
[2]
C. F. Benitez-Quiroz, R. Srinivasan, and A. M. Martinez, “EmotioNet: An Accurate, Real-Time Algorithm for the Automatic Annotation of a Million Facial Expressions in the Wild,” in Proc. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5562–5570, 2016. PDF
[3]
iMotions, “Facial Action Coding System (FACS) – A Visual Guidebook,” iMotions Research Fundamentals, 2024. Link