Dine in or take out dataset: user behavior in an interactive virtual reality café

We present the Dine In or Take Out Dataset, a virtual reality dataset collected using a Meta Quest Pro, to understand how predominantly Gen Z individuals immersed in a virtual café opt to make the choice of dining in or taking food to go. In our dataset, participants completed four different treatments: no wait for maître d' with no wait for food, no wait for maître d' with 3-minute wait for food, 3-minute wait for maître d' with no wait for food, and 3-minute wait for maître d' with 3-minute wait for food, with treatment orders being randomized for each participant. Our dataset consists of 35 participants interacting with a virtual maître d' non-playable character (NPC) in a virtual cafe environment to order food for dine-in or take-out. Our virtual café was designed using Unity and consists of standing and seated sections found in a typical real-world café and includes four NPCs, the interacting maître d' and 3 seated patrons. To replicate the sounds of a typical café, we generate background sounds of clacking utensils and plates, pouring of liquid, distant conversations, ambient music along with the dialog between the NPC and the participant. Participants provided their demographic information and completed the standard Frustration Distraction Scale, Revised UCLA-20 Loneliness Scale, Social Phobia Inventory, State Trait Anxiety Inventory, which measure frustration and discomfort intolerance, loneliness, social phobia, and anxiety respectively, prior to immersion. After each treatment, participants retake the State Trait Anxiety Inventory to measure changes in anxiety and rate the proximity, noise level, frustration due to delays, frustration due to a sense of isolation, frustration due to a feeling of anxiousness, and the likelihood of making the same table choice in the real-world using a 5-point Likert scale. After completing all four treatments, participants are re-administered the Frustration Discomfort Scale to understand any post-study changes in frustration and discomfort, NASA Task Load Index to measure workload, System Usability Scale for perceived usability, and the Virtual Reality Sickness Questionnaire to assess any simulator sickness experienced by the participant. Additionally, we also record head, left-hand, and right-hand position and orientation, and eye-gaze data from the VR headset. Our dataset enables researchers to understand how participant-reported anxiety using the State Trait Anxiety Inventory changes between treatments when compared to the baseline provided before the immersion. Participant-reported loneliness from the Revised UCLA-20 in conjunction with the wait time for the maître d can be used by researchers to understand how decisions to dine in or take out change. Participant eye movements can be studied prior to, during, and after participants opt to dine in or take out to understand whether differences in fixations and saccades are present for participants who opt to dine in versus those that choose take out. The dataset can be used in conjunction with pre-trained times series-based deep learning models such as TEMPO, a Transformer-based model, to predict participant responses after each treatment, such as the closeness of people, noise levels, frustration due to the delay, isolation, and anxiousness.