Plastic People

The robots are coming. We’ve heard this claim frequently over the past 30 years: that someday soon robots will be ironing our clothes, washing our windows, and serving our morning coffee. In fact, the nearest we’ve come to achieving this vision of domestic automation is embodied by the iRobot Roomba, a puck-shaped robotic vacuum cleaner that does decent work on tile and hardwood, but won’t venture near pile.

As a working roboticist, however, I can attest that the vision of domestic robotics is finally, if incrementally, becoming a reality. Robots will not be serving our coffee any time soon, but they will be entertaining our children and caring for our – hopefully not my – elderly relatives. And the likely form of these robots is decidedly humanoid. But what should a humanoid robot look like?

Industrial designers are facing this question with recurring frequency. Since humanoid robots have, until recently, been the stuff of science fiction, there are few guidelines for their design, or commercial examples to study for inspiration. (Most humanoids in the robotic research community appear to have drawn their design inspiration from 1950s science fiction movies, or supermarket displays of canned vegetables.) Recently, however, a coherent set of design guidelines has begun to emerge, drawn from both aesthetic principles and scientific research.

But one could also reasonably answer the question with a tautology: a humanoid robot should look like a human. As it happens, however, it’s difficult to design a robot that looks, moves and behaves in a human-like manner. And when faced with objects that are nearly (but not quite) human, many of us are touched by feelings of unfamiliarity, unease and even disgust.

This negative emotional response has been dubbed ‘the uncanny valley’ and was first described by roboticist Masahiro Mori in a 1970 paper, as part of an analysis of human psychological reactions to robot design. Mori posited that as robots become more human-like in appearance and movement – as they progress, for example, from crude assembly-line manipulators to elegant bipedal walkers with articulated limbs – our reaction becomes increasingly positive.

However, as a robot’s appearance nears but does not fully achieve that of a human being, our reaction becomes sharply negative, to the point of revulsion. It’s been shown that even minor deviations from a standard human form can change a person’s perception of a figure from beautiful to disturbing. Skin colour and tone, feature asymmetry, hair texture – subtle differences in any of these cues can negatively skew our reactions. Humans are especially sensitive to changes in the human face: studies have shown that shifting features even one millimetre can affect our judgment of facial attractiveness.

This sensitivity to ‘humanness’ applies to movement as well. Hiroshi Ishiguro at Osaka University has built a series of humanoid robots that are exact visual replicas of living humans (using his daughter, himself, and a Japanese Broadcasting Corporation anchor as models). Ishiguro has found that when a robot’s movements are inhuman – when they’re unnaturally jerky, or lack ‘micro motions’ induced by natural physiological processes – an observer’s perception of the robot changes. The observer views the robot as uncanny.

These findings create a conundrum for designers. Should their designs fall safely on the ‘inhuman’ side of the uncanny valley, or should they reach for the grail of verisimilitude? And even if a designer succeeds in creating a humanoid with an appearance that’s indistinguishable from a human, will the robot’s movements and behaviour be correspondingly human-like?

The creation of indistinguishably human-like movement and behaviour is, however, still beyond the current state of the art. (Artificial intelligence researchers have been trying to crack the behaviour problem for 50 years, without success.) This would seem to deepen that valley to an infinite ravine.

Luckily, not all humanoids need look like humans. A designer might quite happily aim for another region of the appearance space, which ranges from realistic to completely abstract. Generally, the more iconic the robot’s appearance, the greater the chance that a viewer will be able to project their own experiences and emotions onto the robot. (Whether or not this is desirable is probably dictated by the robot’s function.) The properties of this space have been studied by numerous roboticists, who have exploited the crucial insight that human aesthetic preferences generally transfer to nonhuman objects and beings – and therefore, preferences for human faces should apply to robotic faces as well.

Recent studies have analyzed the effect of facial feature properties on a subject’s perception of a robotic head. It was found that the head’s aspect ratio (i.e. the ratio of length to width) and the presence of multiple facial features (specifically the nose, mouth, and eyelids) strongly influence a subject’s perception of a robot as human. Heads with small aspect ratios – that is, landscape rather than portrait orientations – and sparse facial features are perceived as robotic. The same researchers suggested that humanoid robot design should take cues from consumer appliance design, so that individuals perceive robots as consumer items, and thus maintain a sense of control over the system.

Other design cues can be taken from classical studies of facial attractiveness. Clear skin, well-groomed hair and large, expressive features are generally considered appealing. Large eyes and forehead, and a small nose and jaw are associated with infants, and are thus thought of as endearing. It’s well known that ‘average’ faces are considered attractive. Each of these feature sets inspires a unique response. A child-like appearance inspires nurturing, while attractiveness inspires both friendship and sexual attraction.

Armed with this knowledge, a designer can tailor a humanoid’s design to create a specific human-robot interaction experience. The design goal would not be perfect realism, in the Turing sense – where the human believes he or she is interacting with another human being – but sufficient expressiveness to allow for rich interaction, while avoiding discomfort arising from uncanniness in appearance, movement, or behaviour.

Humanoid robot designers could do worse than to look to films for instructive examples. The Ur-film in this regard is Star Wars (1977), where we find examples of an iconic humanoid in C-3PO alongside an abstract R2-D2. Both robots exhibit jerky, robotic movements, though neither strikes the viewer as uncanny. R2-D2’s squat iconicity is somehow even lovable.

Like any product, however, design decisions for humanoid robots will ultimately be driven by their intended use. For robots that will assist us with household tasks, an aesthetic derived from consumer appliance design might dominate. For robot caregivers, forms that engender feelings of tenderness and empathy will likely prevail. And for humanoid love-bots – yes, they’re on the way – attractiveness and sensuality will trump all.