Comfort-Aware Trajectory Optimization for Immersive Human-Robot Interaction

Abstract

In human-robot cohabited environments, generating socially acceptable and human-like trajectories is critical to fostering safe, comfortable, and intuitive interactions. This paper presents a trajectory prediction framework that emulates human walking behavior by incorporating social dynamics and comfort-driven optimization, specifically within immersive virtual environments. Leveraging the Social Locomotion Model (SLM), our framework captures inter-personal interactions and spatial preferences, modeling how humans implicitly adjust paths to maintain social norms. We further introduce a Nelder-Mead-based optimization process to refine robot trajectories under these constraints, ensuring both goal-directedness and human-likeness with efficiency and applicability. To evaluate the perceptual realism and spatial comfort of the generated trajectories, we conduct a user study in a virtual reality (VR) setting, where participants experience and assess various robot navigation behaviors from a first-person perspective. Subjective feedback indicates that the trajectories optimized by our model are perceived to be significantly more natural and comfortable than those generated by baseline approaches. Our framework demonstrates strong potential for deployment in virtual human-robot interaction systems, where social legibility, responsiveness, and computational efficiency are all critical.