Facial expressions in sex robots: motors, eyes and AI
A smile may be a motor function. A conversation may be a cloud service. Assess them separately before assessing the whole.
Eyes, brows and a moving mouth can change the experience of a physical AI head. But a good product photograph shows only one position. It does not tell you how movements start, how smooth they are or how they align with speech. We therefore look at both the described mechanism and the software features intended to coordinate it. We use specific manufacturer examples and distinguish their feature information from our assessment. We have not measured movement or conducted a physical conversation test with the heads discussed.
Read the movement list alongside the DoF figure
SEDOLL describes Annika as having seven degrees of freedom and movement in the eyes, eyelids, brows and mouth. The manufacturer also shows named facial expressions. Such a figure is useful when accompanied by a specific list. It is not in itself a rating of naturalness. We therefore do not read more expressions as evidence of better conversation or more accurate expressions. Look for which parts move, whether they can be seen in a demonstration and which hardware version the information concerns. [1]
Mouth movement and speech synchronisation are separate questions
RealDoll's robot overview describes both opening and closing the mouth and lip synchronisation with speech sounds. That says more than a still image with an open mouth. But the description is not our measurement of how accurately synchronisation works. Watch a longer speech sequence with natural pauses and changes between sentences. Notice whether the mouth keeps moving when the sound stops, and whether the same quality is demonstrated in the conversation mode and language you actually want to use. [2]
A displayed expression does not tell you how it was selected
A robot head may display an expression without the video explaining how it was selected. Ask whether the demonstration uses a preselected pattern, a remote command or a response from the conversation system. All three may be relevant to know. The crucial distinction is what has been shown. We do not use terms such as emotional understanding as a technical guarantee. A surprised face in a gallery documents a displayed expression; it does not establish that the system has understood a particular situation.
The facial surface is part of the finished design
A technical photo without a wig or with exposed mechanics can be useful, but should not be confused with disassembly instructions. RealDoll's robot manual says users should not remove the face and should contact the manufacturer if this seems necessary. Describing a platform as modular therefore does not mean every part is intended for owner replacement. Clarify which replacements are designed for ordinary use, who performs them and what comes with an additional design. [3]
Conversation has its own requirements
The Annika guide describes connecting to both a network and a server before activating conversation. A motor in the head therefore does not automatically make conversation local or independent of a service. Examine the two layers separately: what controls does the physical unit provide, and what account, connection and software does voice require? Ask for answers covering the current version. This is more useful than assuming a head continues unchanged simply because its mechanics can still receive power. [4]
Use a simple assessment of the whole experience
View the model in ordinary light from both the front and side. Assess whether its gaze, mouth movement and pauses work together in the same recording, and confirm which version you are seeing. Save the product's movement list and note which aspects you have only seen in short clips. Noise, response time, robustness and movement quality require actual testing for a reliable judgement. An honest product assessment leaves these points open rather than assigning an invented test score.
Our assessment
The most informative robot head is not necessarily the one with the most advertised expressions. It is the one whose movements, software requirements and controls can be linked to a precise model. Assess facial animation from documentation and a continuous demonstration, and separate it from promises of understanding.