Should Your Robot Have a Face? A Practical Answer to a 100-Year-Old Question

Look at the machines currently walking out of factories and into the world. Tesla's Optimus has a smooth, dark panel where a face would go. Figure's robots wear a black glass visor. Boston Dynamics gave the electric Atlas a ring of light. Agility's Digit spent its first years with no head at all. The most advanced humanoid robots ever built, machines that can fold laundry and stack totes, mostly cannot smile, frown, or meet your eye. This is not an oversight. It is a decision, made separately by some of the best-funded design teams on the planet.

And yet the same industry keeps drifting back toward faces. Digit eventually got a head with animated eyes. Optimus's blank panel is, by Tesla's own description, a screen meant to display useful information. 1X ships Neo, its home robot, in soft knit with a face deliberately gentler than anything bound for a warehouse. The question "should a robot have a face" turns out to be less settled than the visors suggest.

We make robot faces for a living, so you know where we stand. But the honest answer is more interesting than our sales pitch. Here it is, both sides argued properly.

The question is older than robots

When Karel Čapek staged R.U.R. in 1921 and gave the world the word "robot," his artificial workers looked exactly like people, which was the whole horror of it. Six years later, Fritz Lang's Metropolis gave its Maschinenmensch a serene metal face, expressive enough to deceive a city. From the very beginning, fiction understood that the face was the contested territory: the part of the machine that decides whether we read it as a colleague or a threat. A century on, the engineers building real humanoids have inherited the same question, minus the special effects budget.

What faces are for

Humans are face-reading machines. We extract intent, mood, and attention from faces in a fraction of a second, involuntarily, from across a room. It is our oldest interface, running long before language.

Which means a face on a robot is not decoration. It is a display protocol for intent. This is precisely the reasoning Agility Robotics gave when it finally added a head to Digit in 2023, after years of shipping it headless into warehouses. Jonathan Hurst, the company's co-founder, put it plainly to CNET:

"You really want it to be intuitive. You don't want to have to train people that this light means that and this sound means this. We have the opportunity of using body language and expression to share intent from the robot."

Agility's own design writing goes further: a simple, expressive face communicates the robot's status, its direction of movement, its overall awareness, "preventing surprises and building trust with human co-workers." Note what happened there. The most pragmatic company in the industry, one that sells robots by the fleet to logistics operators, concluded that a face is a safety feature. People who can read a machine's intent do not get startled by it. That argument only gets stronger when the machine moves from a warehouse aisle to your hallway.

The case against: sensors, cost, and honesty

The faceless camp deserves a fair hearing, because its arguments are good.

• Sensors want the head. The top of a humanoid is prime real estate: elevated, clear line of sight over the shoulders, the natural home for cameras, depth sensors, and antennas. Agility lists sensor placement as the head's first job, before any social function. Every millimeter given to an eyebrow is a millimeter not given to a lens.
• Faces are expensive. Actuated facial features mean motors, seals, points of failure, and cleaning problems. A screen showing cartoon eyes is cheaper, which is why screens are the industry compromise. A sealed visor is cheaper still.
• Blankness is honest. This is the strongest version of the argument. A robot does not have feelings. A face that smiles at you implies an inner life the machine does not possess, and some designers consider that a small ongoing lie. A matte visor makes no promises. There is integrity in that.

The trouble is that "no face" is not a neutral choice. Humans read blankness too, and what we read into it is rarely warmth. A six-foot machine with a void where the face should be does not register as honest hardware. It registers as withholding. We wrote about why that is in our piece on why humanoid robots look creepy: the short version is that the uncanny valley has two slopes, and the featureless side is not the safe one.

What the research says people actually prefer

Human-robot interaction researchers have been testing face preferences for years, and the findings are more specific than "faces good."

The study we come back to most is by Akanksha Prakash and Wendy Rogers at Georgia Tech, published in the International Journal of Social Robotics. Younger and older adults rated robot faces across a spectrum of human-likeness, imagining the robot helping with tasks at home. The pattern that emerged: the morphed, almost-human "mixed" faces were liked less than either the clearly human or the clearly robotic ones. Participants compared them to aliens and spacemen. Meanwhile the frankly mechanical faces won people over whenever they read as, in the participants' own words, cute, friendly, and trustworthy. For chores and social tasks, a likable robot face beat the near-human option comfortably.

Two details from that study matter for anyone designing, or dressing, a robot. First, task context changed everything: people wanted more humanness from a robot giving advice, less from one doing dishes. A face is not one decision; it is a decision about a relationship. Second, when asked what features a home robot's face should have, both age groups wanted eyes and a mouth and were happy to skip the nose and ears. People do not want a replica of a human face. They want just enough face to read.

That finding, abstraction over realism, shows up across the HRI literature, and it quietly indicts both extremes the industry ships today: the photoreal android and the blank visor. The evidence points at the middle. Simple, legible, deliberately not-human.

Faces in the home are different from faces in the warehouse

Everything above gets amplified by context. A warehouse robot is equipment. You badge in, work a shift alongside it, and go home. If Digit needed eyes to keep trained logistics staff from being startled, consider the setting where nobody signed a safety briefing: your kitchen.

A home robot is not equipment in any lived sense. It is a presence. It stands in your hallway at 2am on its charger. It is there at 7am, before coffee, at your most unguarded. Agility, whose entire business is industrial, says the home is the destination but the bar is far higher, quoting the memorable standard that a robot is not going into the home until you can prove it will not fall on a baby. Capability and safety are the engineers' problem. But there is a third bar nobody certifies: whether the thing is bearable to live with. A blank faceplate that reads as "industrial" in a fulfillment center reads as "unsettling roommate" at breakfast.

If you build robots and you are wrestling with exactly this gap between factory-ready and living-room-ready, that is the conversation on our partners page.

A face you can take off

Here is the position we have arrived at, after sitting with both camps: the robot should stay honest, and the face should be a choice.

A removable mask splits the difference in a way neither a bolted-on face nor a bare visor can. Underneath, the machine remains exactly what it is: sensors, actuators, a dark panel, no pretense of an inner life. The manufacturer never lied to you. On top, a soft knit face, chosen by the person who actually lives with the robot, cut so that every camera and sensor keeps its full field of view. If you own a printer, you can even print one yourself for free. The warmth is added by the owner, the way we have always domesticated our tools. Nobody accuses a dog's bandana of dishonesty.

It also answers the research better than anything molded at the factory. If preferences vary by task, by age, by household, then no single face shipped from a production line will be right for everyone. A face you can change, or remove entirely when the robot is doing rough work, is the only design that takes the HRI findings seriously. You can see what this actually looks like on a real machine in our before and after comparison: same robot, same hardware, a different presence entirely.

What owners choose when given the choice

We will end with what we see in our own small corner of this, offered as anecdote rather than data, because that is what it is. When people render faces for their robots in our Mask Lab, they do not reach for realism. Almost nobody tries to make the robot look human. They pick the simple, abstract, friendly patterns: soft eyes, a suggestion of a mouth, colors that match the kitchen. Given a completely open canvas, people independently land where the research pointed all along, somewhere warm and legible and honestly artificial.

You can see what that looks like in practice in the Mask Market, where every design was prompted by an owner and ranked by what people actually print.

So, should your robot have a face? The visor camp is right that it does not need one to work. The HRI literature is right that the wrong face is worse than none. And your household is the only party qualified to make the final call. Our suggestion: make it reversible. Try a face on your robot in the Mask Lab, and see what it does to the 7am encounter. Worst case, you take it off. That is rather the point.