Elva is a humanoid robot face that holds eye contact, responds with genuine expression, and remembers — built for long-term companionship and care.
Not renders. A living archive of real faces, and thousands of hours of real, dynamic expression — the only proof that a humanoid robot face can move like a person.
Across five continents, all ethnicities, all ages. Expressiveness that does not collapse into a single face of the world.
More than sixty muscle action units annotated on each face. The paths of attention, of rest, of a polite smile and a sincere one — named, labelled, and taught to the motors.
One face: twelve in the mouth, twelve in the eyes, three in the head and neck, the cheek pair of a genuine smile, and the lift of the nose. Not a mask. A mechanism of presence.
Companion. Care. Educate. Research.
Not a walking body. Not a cartoon on plastic. Not a screen that talks. Elva is a super-bionic humanoid robot face — built to sit across from someone, and to be believed.
The product is presence at the distance of a conversation. Elva does not walk the floor and does not replace a kiosk. It is a humanoid robot face for rooms that already pay for a person to sit there — care, training, research, a desk that greets.
Three million real human faces, across five continents and forty-plus countries. Thousands of hours of real, dynamic expression. Sixty-plus action units named on each. Expressiveness is not a library of poses. It is earned from how faces actually move.
Thirty degrees of freedom in one face: twelve in the mouth, twelve in the eyes, three in the head and neck — and the cheek pair of a genuine smile, and the lift of the nose. People can tell. Elva puts that difference into muscle, so words and face agree.
A near-human circadian rhythm: bright by day, eyes closed at night. At rest, it does not sense. Raw camera images are not uploaded; expression runs on-device. The detail that feels most alive is also why a face can be left in a private room.
Identity, knowledge and personality are set in plain language and held. Care routines, family stories, institutional guidance: Elva draws only from what has been entrusted — not from the open web. A person is never meeting a stranger twice.
Every line below is a real capability of Elva — head, perception, and the data that taught it how a face actually moves.
Every cell below stands for a real human face, overlaid with blendshape guide lines — 3 million and more, across five continents and forty-plus countries, with sixty-plus action units annotated on each. Not renders. Real data.
A living archive of real faces. Each square is a specimen of structure and expression — the lines that teach a machine how a face actually moves.
A voice from a box cannot hold a long hour of care. A cartoon on plastic cannot be believed when it smiles. A humanoid robot that will sit with someone — patiently, one to one — needs a face: attention is shown there, a smile is believed there, and rest is understood there.
Not to perform. To give a presence emotional depth and contextual influence — so a person is met, not addressed.
Three ways of being in a room. Companion, training, rest — chosen for the setting, not for show.
One to one. The face engages when it is addressed, holds the conversation, and remembers the last mood. Built for the patient hours of companionship and care — a listener in the room, not a display at the front of it.
Predictable, non-judging. Expression clarity and gaze intensity can be set so a social-training partner stays readable, never a stare. Turn-taking and emotion-naming become possible because the face does not overwhelm.
Eyes close. At rest, it does not sense. The detail that feels most alive — a near-human circadian rhythm — is also the privacy switch: a face that sleeps does not watch.
Steady eye contact, attention that turns toward you, micro-expressions that shift with the moment — so a person feels genuinely seen and heard.
A sincere smile and a polite one are not the same — and people can tell. Elva puts that nuance into every facial muscle.
The last conversation, the last mood, the story told before. Companionship is not a one-off performance but a relationship that lasts.
It rests, and it sleeps. Elva carries a near-human circadian rhythm — bright by day, eyes closed at night. The detail that feels most alive is also the most natural guardian of privacy: at rest, it does not sense.
Describe how this humanoid robot face should meet a person — in plain language. Identity, knowledge and personality stay consistent, so companionship is a relationship, not a one-off performance.
A name, a language, a place in the room — companion, listener, training partner. The identity is designated once and held, so a person is never meeting a stranger twice.
Designate the central source of data and information. Care routines, family stories, institutional guidance: Elva draws only from what has been entrusted — not crawled from the open web. The source of truth is chosen.
Gaze intensity, expression clarity, warmth and pace. Personality is how the face agrees with the words — a sincere smile and a polite one are not the same. Adjustable, never theatrical.
An endlessly patient companion — for reminiscence, gentle routine and easing loneliness, helping close the caregiver gap. Companionship only; not diagnosis, not a substitute for medical staff.
Sitting room · Late afternoon
A predictable, non-judging partner with adjustable expression clarity and gaze intensity, supporting emotion-naming and turn-taking as a training aid.
Two chairs · A readable hour
A listener that reads real emotion and does not judge — a private outlet that can hold heavy feelings.
Evening lamp · Rain on the glass
A genuine human-robot interaction platform for universities and institutions — for study, teaching and demonstration.
Studio bench · Daylight
Another presence at the desk — concierge, lobby, brand space — not a kiosk and not a cartoon. A face that can greet, guide and hold still when the room is busy.
Lobby desk · Afternoon
Elva is not a cheaper chatbot seat. It is a humanoid robot face for operators, trainers, researchers and front-of-house teams who already fund hours of human presence — and still cannot fill them.
The caregiver gap is a budget line, not a slogan. Elva sits for reminiscence, gentle routine and easing loneliness — so a person in the room is not left with a screen.
What is bought: presence that lasts. Not diagnosis. Not a substitute for medical staff.
Hours of turn-taking and emotion-naming are already paid to people. Elva is a predictable, non-judging training aid, with gaze and expression set to stay readable.
What is bought: repeatable practice. Not therapy. Not a clinical claim.
Human-robot interaction needs a physical face, not a render. Elva is a platform for study, teaching and demonstration — identity, knowledge and personality held constant across sessions.
What is bought: a research and teaching instrument. On-device, GDPR-ready.
Lobbies already pay for people who greet, wait and represent. A humanoid robot face holds that hour without becoming a screen or a mascot — and can rest when the room does.
What is bought: dwell, greeting, brand presence. Not a walking body. A trusted face.
On-site, or at WCAG-2026 in Berlin. A real face in a real room — not a deck.
One setting, a designated knowledge source, identity and personality locked. The test is whether people dare to leave it there.
Tailored for that operator: language, role, rest cycle, GDPR by design. Regional partnerships welcome.
A face that can be trusted overnight is a face that can sleep. Privacy is not a policy bolted on afterwards — it is how Elva behaves in the room.
Im Ruhezustand nimmt es nichts wahr.
Eyes close. At rest, the face does not perceive. The near-human circadian rhythm — bright by day, still at night — is also the most natural guardian: a face that sleeps does not watch.
The expression layer runs on the device, in real time. Raw camera images are not uploaded. What is seen in the room stays in the room.
Care routines, family stories, institutional guidance: Elva draws only from what has been entrusted by the operator — not crawled from the open web.
Built to meet European expectations as a starting condition, not a later patch. On-device, no raw upload, a resting state that does not sense.
For care operators, universities, regional partners and tailored deployments — in person or at WCAG-2026 in Berlin.
Book a demo