AI Avatar Design: User Study within KARLI research project

How should AI interact with you during a relaxed, everyday drive? What kind of personality would you like while cruising down a long, uneventful highway? And, most importantly, how should AI respond when you find yourself in an unexpected emergency situation? The answers to these questions are invaluable for designing an AI that not only acts according to programming but also adjusts to your needs and preferences.

Our team is committed to the pioneering research project KARLI, funded by the government. Our goal? To develop a smart and user-friendly interaction system for the cars of the future. This system won't just utilize cutting-edge technology but will also take into account the wishes and preferences of the people inside the vehicle.

We recently conducted a user study in our design studio using a Virtual Reality (VR) setup to immerse participants in a simulated driving experience. Within this VR environment, we had passengers go through different driving situations, starting from a friendly greeting to handling a potentially risky scenario. The crucial part in improving the driving experience was the questions we posed to the participants during these scenarios.

Creating an Immersive VR Environment for Study Participants

We established an immersive VR environment using a specially equipped chair featuring a steering wheel and pedals. This setup allowed us to immerse 20 participants from diverse genders and age groups into the VR world created using Unity. Video data was delivered through VR headsets, providing an engaging and lifelike experience.

Exploring the Impact of Different AI Avatar Designs on Users

We created three distinct AI avatar designs that significantly differ in appearance and behavior. Our goal was to assess how these various AI avatars influence users.

Design A, known as "Abstract", is a completely abstract avatar that dynamically transforms itself based on the situation. It forms a mouth when speaking and turns into an exclamation mark in moments of danger. During interactions and pauses, the avatar pulsates to indicate its availability.

Design B, named "Robotic", features a faceless, robot-like avatar with pulsating lines that shift to convey activity and availability.

Design C, "Humanoid", represents a human-like avatar with a facial representation created from a point cloud. This avatar can display different facial expressions and exhibits regular blinking during inactivity.

For each AI alternative, we designed three scenarios: Welcoming the driver; encouraging the driver to take over control and reacting to a critical situation when the driver fails to respond.

During scenario evaluations, participants rated various aspects, including clarity of understanding, scariness, usefulness, visual appeal, social impression, and emotional connection of the avatar. They answered questions like 'How well did you understand the avatar?' or How scary did it appear to you' on a scale from 0 to 10.

Results: Avatar Design Preferences

In the study, Design B, "Robotic," emerged as the leading choice in terms of ranking, Net Promoter Score (which measures the likelihood of consumers recommending a product or service), and overall questionnaire assessment. Following closely were Design A, "Abstract," and Design C, "Humanoid."

Younger participants, particularly those with experience in artificial intelligence, tended to prefer Design B, "Robotic." On the other hand, Design C, "Humanoid," found favor mainly among participants over 40 years old.

Notably, the most popular choice overall was "Robotic." Interestingly, as participants' age increased, the preference for "Humanoid" grew, while "Abstract" became less popular.

Results: Understanding and Aesthetics

Participants found the "Robotic" avatars to be the easiest to understand and visually appealing. The "Abstract" avatars struck a balance, being both understandable and aesthetically pleasing. However, the "Humanoid" avatars were seen as less understandable and not visually appealing.

Results: Creepiness and Usefulness

"Abstract" avatars were not perceived as creepy and were considered useful. The "Robotic" avatars, while not creepy, had a slight unsettling factor but were still viewed as useful. On the other hand, "Humanoid" avatars were thought to be extremely creepy.

Summary

In summary, the study found that the "Robotic" design was both highly understandable and somewhat visually appealing. In contrast, the "Abstract" design was seen as the most useful and least creepy among the options, while the "Humanoid" avatars were generally considered to be creepy and less understandable.

Moreover, in terms of overall preference, "Robotic" emerged as the top choice, particularly among younger participants with AI experience. In contrast, "Humanoid" found favor with older participants, and "Abstract" saw a decrease in popularity as participants' age increased.

What’s next?

In the next phases, we're excited to optimize the AI Avatar based on the study results and smoothly integrate it into our VR application. Our research and the findings from this user study are essential steps in creating more intelligent, adaptable, and responsive AI interactions for future vehicles. The future of driving will be a blend of technology and user-centered design. We're committed to ensuring that this future provides a comfortable and safe driving experience for every passenger.

Research project KARLI is funded by the German Federal Ministry for Economic Affairs and Climate Protection. More about KARLI at karli-projekt.de

All partners:

Continental Ford Motor Company, AUDI AG, Allround Team GmbH, Universit?t Stuttgart, Hochschule der Medien Stuttgart, TWT GmbH Science & Innovation, paragon semvox GmbH, Fraunhofer IAO, Fraunhofer IOSB, INVENSITY and branmatt Rechtsanwaltsgesellschaft mbH