Galatea responds
Character, conversation and emotional state shape the response.
Personality + memory← The Avatar projectThe same character.
FUTURE · PHYSICAL EXPRESSION
The same character.
A face you can meet.
Today, Galatea gives a digital character a voice and a face. Our next direction: carry that expression into a physical robot.
A smile starts in software. Small motors can bring it into the real world.
See how it works ↘Development direction · robot integration still to be validated
Kismet · MIT. Hardware illustration; not a Galatea prototype or evidence of blendshape control.
Photo: Rama / Wikimedia Commons · CC BY-SA 3.0 FR · Image unchanged.
From a response to a real smile.
“Blendshapes” are expression controls: how much to smile, blink or open the mouth. They describe a face. A robot-specific translator turns those values into movement.
Expression signals
Software describes the intended smile, blink and mouth movement.
Blendshape valuesTranslate for the robot
A calibrated mapping converts expression signals into commands for that particular face.
Robot-specific integrationThe face moves
Motors pull cables or move linkages beneath the skin.
A physical expressionProposed Galatea integration. The research below demonstrates the expression-to-motor principle; it does not demonstrate a completed Galatea deployment.
WHY THIS MATTERSKeep the character.
Keep the character.
Change how it appears.
One software foundation
The opportunity is to carry Galatea’s personality, memory and expression logic from a screen to a physical face.
Work with robot makers
Galatea provides the conversational character. A hardware partner provides the robot and the controls needed to move it.
A practical next step
Match the expression controls, calibrate the motors, then test movement limits, timing and expression quality on real hardware.
ENGINEERED ARTS · ROBOTS IN ACTION
See the hardware in action.
Start with Ameca and Mesmer from Engineered Arts: physical robots showing facial movement and the mechanisms behind it. The videos demonstrate expression; they do not establish the exact blendshape-to-motor mapping.
▶
Ameca
Human facial motion captured with ARKit and mapped to a robot.
Video owner: Engineered Arts · Open on YouTube ↗▶Mesmer
Two moving heads: one with skin, one showing the mechanism.
Video owner: Engineered Arts · Open on YouTube ↗▶Emo
A physical robot face that anticipates and mirrors a smile.
Video owner: Columbia Engineering · Open on YouTube ↗This bridge exists in research.
ExFace explicitly maps blendshape signals to physical robot motors. Its authors report tests on two different robotic faces.
Micheal
A cable-driven robotic face. The study maps expression values to facial motor commands.
Read the hardware description ↗Hobbs
A second robot using articulated linkages. The authors report validating the same approach on this different mechanism.
Read the two-robot evidence ↗“mapping from human facial blendshapes to bionic robot motor control”— Zhang et al., ExFace, abstract (2025) [1]
A little more technical detail
Each robot needs a mapping between expression values and its own motors. Those values are not universal motor commands. Calibration must account for the face’s mechanics, available movements and operating limits. A completed Galatea-to-robot integration has not been demonstrated here.
Evidence you can check.
[1] Dong Zhang, Jingwei Peng, Yuyang Jiao, Jiayuan Gu, Jingyi Yu and Jiahao Chen. ExFace: Expressive Facial Control for Humanoid Robots with Diffusion Transformers and Bootstrap Training. ShanghaiTech University, 2025. arXiv:2504.14477v1. See the abstract, Introduction and § III. Paper ↗
[2] Apple — ARKit, ARFaceAnchor.blendShapes. Documents expression values for features such as blinking and jaw opening. Documentation ↗
Robot names and third-party videos belong to their respective owners. Videos are linked to their original publishers; no third-party footage or research figures are reproduced here. The Kismet photograph by Rama is reused unchanged under CC BY-SA 3.0 FR, with its source and licence linked beside the image. The process diagram is an original conceptual illustration. References do not imply a partnership, endorsement or permission to use the researchers’ software commercially.
Discuss a robot integration ↗ See Galatea’s current demonstrations ↗