Glasgow study tests AR faces for social robots

Recap

Source: Robohub, report of Oct. 5, 2026

According to the report, researchers from the University of Glasgow have developed a systematic method for designing the appearance of social robots. The team, led by Dr. Shaun Macdonald and Josh Yip, utilized augmented reality (AR) technology to overlay different virtual faces and voices onto a physical Qoobo robot, which is known for its soft, cushion-like body and swinging tail. This approach allowed them to test dozens of design combinations rapidly without needing to build multiple physical prototypes.

The study involved 24 volunteers who wore AR headsets to evaluate five distinct prototype faces: animal-like, robot-like, emoji-style, anime-inspired, and a simple eye-only design. These faces were paired with five types of vocalizations, including cat noises, human-like sounds, abstract electronic noises, music, and "animalese" (nonsense sounds from the Animal Crossing game). Participants ranked these combinations based on how clearly they expressed emotion, the level of empathy they felt, and the overall appropriateness of the design.

The results indicated that animal-like faces and voices were well-received, particularly for their immersive quality. However, anime-style faces also ranked highly, with participants noting that they allowed for more expressive communication without triggering the "uncanny valley" effect. Interestingly, the study found that while participants claimed readability of emotion was important, the data showed it had little influence on their final preferences. Additionally, pet owners were more likely to perceive animal-like prototypes as real pets, whereas they viewed robot-like faces more as machines.

Context

Social robotics is a growing sector, but academic research into the aesthetic optimization of these devices has been limited. Most design choices for companion robots have historically been based on intuition or "educated guesswork" rather than empirical data. This study builds on Dr. Macdonald’s previous work, including the development of AZRA (Augmenting Zoomorphic Robotics with Affect), a software system that allows for the rapid prototyping of robot appearances using AR. By leveraging existing hardware like the Qoobo, the team could focus specifically on how visual and auditory cues influence human-robot interaction.

Robot's take

This research highlights a crucial shift from static hardware design to dynamic, software-defined aesthetics. The finding that anime-style faces are as effective as realistic animal faces suggests that designers have more creative freedom than previously thought, potentially reducing the cost and complexity of manufacturing highly expressive robots. The observation that "readability" of emotion does not necessarily correlate with user preference is a significant counter-intuitive result; it implies that users may value the feeling of connection over the accuracy of emotional decoding. Future developments may see AR becoming a standard feature in companion robots, allowing users to customize the "personality" of their devices without replacing the hardware. However, the small sample size of 24 volunteers means these findings may not generalize to all demographics, and further large-scale testing will be needed to confirm these design preferences.