A round display as a face: expression engine design for desktop robots

A round display becomes a face when it stops being an image and starts being a system. The distinction shows immediately in use: an image changes when something happens, a face keeps living between things happening. Building that system is mostly about layers, gaze and restraint — and about accepting that detail matters less than timing at face scale.

Four layers beat one canvas

A workable expression engine separates the face into independent layers: the base face — eyes and mouth in their neutral state; the gaze layer — pupil position and eyelid state, driven by attention logic; the effect layer — blinks, blushes, sparkles, one-shot and self-clearing; and a mood tint — a subtle palette and brightness shift that colours every other layer without redrawing them. When the layers are independent, changing mood costs a tint change; on a single-canvas design, every mood needs its own full redraw and the asset library explodes.

Gaze is attention made visible

The most life-giving system in a robot face is not the mouth — it is the gaze logic. The eyes should move toward whoever is speaking, hold briefly, and drift away, exactly like human conversational attention. Full tracking — eyes locked on a person wherever they move — crosses from attentive into surveillance within seconds. The design that reads right: glance, hold two or three seconds, drift, occasionally glance at nothing. Combine with the head gimbal where one exists: large angles go to the neck, small corrections to the pupils, and the compound motion reads as natural attention.

Idle is the majority state

Faces spend most of their time not interacting, and idle behaviour decides whether the robot feels alive or paused. A good idle set includes wandering gaze, occasional slow blinks at irregular intervals, a rare stretch or squint, and micro-saccades — tiny eye jitters that faces make constantly in life. All of these are cheap to render and expensive to fake; a face that goes static between interactions reads as a screensaver.

The rendering budget decides the art style

Expression engines run on modest processors, and the budget favours flat, layered design with pre-rendered assets over photoreal faces. At desktop-robot screen sizes, photorealism reads uncanny anyway; stylised faces with strong silhouettes carry mood at two metres. The budget conversation belongs in the art brief: assets designed for the render budget produce smooth faces; assets designed for a poster produce stutter.

We build desktop robots with the face as a designed system — layers, gaze, idle sets, render budget. The OEM/ODM page shows how we structure development, or write to us with the character you want to bring to life.