Editorial: Social human-robot interaction (sHRI) of human-care service robots

Minsu Jang, Jongsuk Choi, Ho Seok Ahn, Chung Hyuk Park

Introduction

As global demographics shift toward aging populations and the prevalence of single-person households increases, the demand for service robots is transitioning from simple utility to complex care. While the first generation of service robots focused on "non-interactive" physical tasks—such as floor cleaning or logistics—the next frontier lies in "human-care" robots. These machines must navigate the intricate social and physical nuances of human environments, moving beyond being mere "chore-bots" to becoming socially intelligent "care-bots." This editorial synthesizes recent research that integrates machine learning, hierarchical motion planning, and user-centric psychology to bridge the gap between mechanical execution and social engagement.

Addressing the Social Gap

The primary challenge in modern robotics is the "social gap." Most current systems lack the capacity for meaningful, socially intelligent interaction required for roles such as reception, medical intervention, or emotional companionship. To be effective in a domestic or healthcare setting, a robot must do more than avoid obstacles; it must understand human engagement, adapt its behavior to user preferences, and perform "social manipulation"—interacting with objects and people in a way that feels natural and supportive rather than purely mechanical.

Methodological Frameworks for Social Intelligence

The research featured in this collection employs diverse methodologies to address different facets of social human-robot interaction (sHRI).

Multimodal Engagement and Role-Switching

One significant study focuses on therapeutic settings, specifically for children with autism. By utilizing computer vision for pose estimation and facial expression recognition, combined with Reinforcement Learning (RL) and Long Short-Term Memory (LSTM) models, researchers developed a system capable of predicting engagement levels. This allows the robot to autonomously switch between "leader" and "follower" roles during activities like dancing, ensuring the interaction remains therapeutic and responsive.

Deep Reinforcement Learning in Social Contexts

Training robots to make social decisions based on estimated emotional states requires vast amounts of data, which is often difficult to obtain in real-world settings. To solve this, researchers utilized specialized simulators like SimDRLSR, which provides a risk-free environment for robots to learn social perception and navigation through high-volume synthetic training data. This framework, known as Social Deep Q-Networks (SocialDQN), allows for the development of complex social decision-making policies.

Hierarchical Control and Task Planning

Bridging the gap between high-level social goals and low-level motor execution is a persistent hurdle. The introduction of "Combined Task and Motion Planning" (CTAMP) architectures uses symbolic task planners to generate action sequences, which are then verified through geometric reasoning at the metric level. This ensures that social tasks are grounded in physical reality, allowing robots to operate reliably even under sensor uncertainty and environmental noise.

User-Centric Personality Studies

Beyond technical execution, the psychological impact of robot behavior is critical. Video-based user studies have been used to quantify how robot personality (e.g., "serious" vs. "cheerful") and interaction modalities (e.g., storytelling vs. physical movement) affect long-term engagement. Findings highlight that user preference is not universal but depends heavily on the robot's perceived personality and the user's familiarity with the technology.

Key Insights and Results

The synthesis of these studies reveals several critical insights for the future of sHRI:

  • Dynamic Social Roles: Effective interaction is not purely reactive. Robots that can fluidly transition between driving an interaction and responding to a user are significantly more engaging, particularly in specialized care settings.
  • The Role of Simulation: Training social intelligence in the real world is slow and carries the risk of frustrating or alienating users. High-fidelity social simulators are essential "sandboxes" for developing AI-driven social behaviors before real-world deployment.
  • Personality as a Variable: A robot’s "personality" and the modality of its interaction are primary determinants of user retention. A one-size-fits-all approach to robot behavior is insufficient for diverse human populations.
  • Symbolic and Physical Integration: Robots must simultaneously reason about abstract social symbols and concrete physical metrics to be effective in unpredictable care environments.

Future Directions

While these advancements represent the current state-of-the-art in sHRI, several challenges remain. Transitioning models from controlled simulations to the unpredictable reality of private homes is a major hurdle. Furthermore, future research must address long-term adaptation—how a robot can evolve its social strategies over months or years of cohabitation—and the scalability of social intelligence into a general-purpose "social brain." The path forward for human-care robots lies in this intersection of technical robustness and psychological sensitivity.

Summary updated

BibTeX
@article{jang2022editorial,
  author = {Minsu Jang and Jongsuk Choi and Ho Seok Ahn and Chung Hyuk Park},
  title = {{Editorial: Social human-robot interaction (sHRI) of human-care service robots}},
  journal = {Frontiers Robotics AI},
  year = {2022},
  doi = {10.3389/frobt.2022.1064440},
}
Figure