On September 16, 2026, SexTechGuide published a detailed look at Somnia Lab, a Hong Kong-incorporated startup with a build team in Shenzhen that is using motion capture technology to train a forthcoming AI sex robot called Model L to perform sex positions. The company has posted video of a performer in a mocap suit thrusting on a gym mat, and in late August put out a since-removed callout for motion capture actors arguing that "billions of hours of adult content contain almost no usable motion data." Somnia Lab announced a $10 million angel round in April 2026. Estimated delivery for Model L is Q3 2027.

Why It Matters

Strip away the render gallery and there's one genuinely novel idea here: using human motion capture as a training corpus for robotic intimate movement. The observation behind the actor callout is correct and non-obvious — adult video is abundant but contains almost no usable motion data, because footage is not joint-position telemetry. If Somnia Lab actually builds a mocap library of intimate movement, that dataset is a more durable asset than the robot, and it's the kind of thing a larger manufacturer eventually licenses or acquires. Most sex robots today can barely move at all; Chinese doll maker Evas Doll recently made news simply by adding motorized legs that respond to voice command on a $2,999 model. Everything else is the sector's standard credibility problem. A disabled pre-order button, renders standing in for hardware, position counts that change depending on who's asking, and a 2027 ship date are the exact pattern that has preceded a decade of sex robot announcements that never shipped. Investors and journalists should treat the $10M angel round and the Hong Kong showroom as the only verified assets and the rest as marketing until hardware is independently handled. The privacy design deserves harder scrutiny than the hardware: a cloud video upsell attached to an always-on camera in an intimate setting is a data architecture decision, and it is being made before anyone has shipped a unit.

The claimed hardware: four tiers running from a 165cm, 22kg Basic to a 175cm, 55kg Founder Edition, all in humanlike form, all with touch sensors on breasts, hips and vagina, all with binocular cameras in the eyes and a built-in speaker for AI conversation. Body heat is rationed by price — the two cheaper tiers heated to 30–37°C in what the company calls key areas, the top two to 37°C throughout. Facial expressions scale from none on the Basic to twelve on the Founder Edition. Model L will not walk; Somnia Lab says it deliberately skipped walking and dexterous hands because both add weight, noise, power draw and cost.

The numbers are where it gets slippery, and SexTechGuide's reporting is appropriately skeptical. The advertised "100+" figure is labelled "sexual interactions" on the product page above a note that more positions are rolling out — the Basic ships with four, expandable over the air, and missionary, cowgirl and doggy are the only three named anywhere. Founder and chief executive Xara told Hong Kong outlet Dim Sum Daily in August that the robot can switch between 165 distinct intimate movements, and gave height and weight figures matching none of the four published tiers. Much of the humanoid imagery appears to be AI or CGI renders rather than hardware; an older clip shows a mannequin-style head with wires coming out of a metallic neck. The pre-order page quotes a $300 refundable deposit, then disables the button that would take it. Xara has put initial pricing in the low five figures USD.

There is also a privacy structure worth naming plainly. Model L carries cameras in its eyes, and Somnia Lab's memory plan charges extra for cloud video — a paid tier for uploading footage from a camera-equipped device in a customer's bedroom. Separately, the engineering team sits in a country that has moved to ban AI companions cultivating emotional dependency while apparently leaving sex robots outside the rules, though Somnia Lab does not plan to sell Model L there.

Sources


Update — 2026-09-18

Initial entry — story first created.