Abstract
We explored human-virtual agent collaboration during a layout design task in a virtual reality environment. Specifically, we developed a human-in-the-loop optimization-based method that drives the decision-making of the virtual agent. Our algorithm accounts for spatial constraints in furniture placement by evaluating boundary proximity, collision costs, and relationships between furniture items in real-time. It also considers the current configuration of the living room, as modified by the user during the co-design process, to guide the virtual agent's furniture placement decisions in the virtual living room. We compared our method (i.e., optimization) against two other co-design strategies (i.e., template and random) following a within-group ($N=24)$ study design. We found the proposed optimization co-design strategy significantly enhanced perceived collaboration compared to the other two co-design strategies. Moreover, our participants attributed higher private and public awareness to the virtual agent in the optimization condition. In addition, the analysis of the logged data showed that participants placed more furniture items and made fewer corrections when codesigning the living room with a virtual agent whose decisions were based on the optimization method. Our results demonstrate that a virtual agent's behavior, which dynamically responds to user actions while maintaining spatial coherence, creates more effective collaborative experiences in an immersive co-design task.