The Socially Invisible Robot: Navigation in the Social World using Robot\n Entitativity

Aniket Bera, Tanmay Randhavane, Emily Kubin, Austin Wang, Dinesh Manocha, Kurt Gray

Abstract

We present a real-time, data-driven algorithm to enhance the social-invisibility of robots within crowds. Our approach is based on prior psychological research, which reveals that people notice and—importantly—react negatively to groups of social actors when they have high entitativity, moving in a tight group with similar appearances and trajectories. In order to evaluate that behavior, we performed a user study to develop navigational algorithms that minimize entitativity. This study establishes a mapping between emotional reactions and multi-robot trajectories and appearances and further generalizes the finding across various environmental conditions. We demonstrate the applicability of our entitativity modeling for trajectory computation for active surveillance and dynamic intervention in simulated robot-human interaction scenarios. Our approach empirically shows that various levels of entitative robots can be used to both avoid and influence pedestrians while not eliciting strong emotional reactions, giving multi-robot systems socially-invisibility.

Introduction

As robots transition from industrial settings to shared social spaces, they face the challenge of being "too noticeable." In applications like environmental monitoring or surveillance, robots must navigate through human crowds without causing alarm or altering natural human behavior. Traditional navigation algorithms prioritize collision avoidance but often ignore the social cues that humans are evolutionarily tuned to detect.

Psychological research indicates that humans react strongly to groups with high entitativity—collections of actors that appear uniform, move in sync, and maintain close proximity. High entitativity often triggers a threat response, leading to unease or social reactance. This research proposes a method to bridge the gap between physical safety and social "invisibility" by quantifying and controlling robot entitativity.

Methodology

The proposed framework utilizes a data-driven approach to map physical robot behaviors to human emotional responses. The process involves four primary stages:

  1. Entitativity Modeling: Defining group cohesion through appearance uniformity, common movement, and proximity.
  2. User Study: A web-based study with 100 participants was conducted to rate simulated robot group movements on a 4-D emotional vector: Friendliness, Creepiness, Comfort, and Unnervingness.
  3. Entitativity Data-driven Mapping (EDM): A mathematical model was developed to relate motion parameters (neighbor distance, personal space, group cohesion) to perceived emotional responses.
  4. Navigation Integration: The EDM is integrated with Generalized Velocity Obstacles (GVO) to solve for optimal trajectories that minimize negative social impact while reaching target destinations.
The navigation pipeline: from input crowd tracking to selecting social interaction levels and outputting group robot navigation.

Key Insights and Results

The study revealed that the more a collection of robots acts as a single unit, the more threatening it becomes. Interestingly, a "sweet spot" exists for social invisibility; while high cohesion is unnerving, extremely low cohesion can be perceived as "creepy."

Comparison of high and low entitativity robot groups and their associated emotional ratings for friendliness, creepiness, comfort, and unnervingness.

Key findings include:

  • Emotional Variance: Principal Component Analysis (PCA) showed that two components explain over 96% of the variance in human emotional perception of robot groups.
  • Stealthy Navigation: In active surveillance modes, robots successfully blended into crowds by maintaining minimum entitativity levels ($E_{min}$).
  • Dynamic Intervention: Conversely, by intentionally increasing entitativity, robots can act as a "social wall" to subtly influence crowd movement without physical force.
  • Computational Efficiency: The algorithm operates at interactive rates, with trajectory computation times ranging from $1.0 \times 10^{-5}$ to $1.0 \times 10^{-3}$ milliseconds.

Conclusion and Future Work

This work demonstrates that entitativity modeling allows multi-robot systems to navigate social environments with minimal psychological impact on humans. Future research will expand this model to include more detailed robot aesthetics and explore how individual human personality traits influence the perception of robot group dynamics.

Summary updated

Abstract

We present a real-time, data-driven algorithm to enhance the\nsocial-invisibility of robots within crowds. Our approach is based on prior\npsychological research, which reveals that people notice\nand--importantly--react negatively to groups of social actors when they have\nhigh entitativity, moving in a tight group with similar appearances and\ntrajectories. In order to evaluate that behavior, we performed a user study to\ndevelop navigational algorithms that minimize entitativity. This study\nestablishes a mapping between emotional reactions and multi-robot trajectories\nand appearances and further generalizes the finding across various\nenvironmental conditions. We demonstrate the applicability of our entitativity\nmodeling for trajectory computation for active surveillance and dynamic\nintervention in simulated robot-human interaction scenarios. Our approach\nempirically shows that various levels of entitative robots can be used to both\navoid and influence pedestrians while not eliciting strong emotional reactions,\ngiving multi-robot systems socially-invisibility.\n
BibTeX
@inproceedings{bera2018socially,
  author = {Aniket Bera and Tanmay Randhavane and Emily Kubin and Austin Wang and Dinesh Manocha and Kurt Gray},
  title = {{The Socially Invisible Robot: Navigation in the Social World using Robot\textbackslash\{\}n Entitativity}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2018},
  doi = {10.1109/iros.2018.8593411},
}
Figure