TLDR: A new research paper addresses the inconsistent use of “social context” in Human-Robot Interaction (HRI) research. It surveys existing definitions and proposes a conceptual model that defines social context as attributes of agents, environments, and their associations that influence an interaction. The paper provides detailed taxonomies for these attributes, covering environmental factors, agent characteristics (actions, demographics, physical and mental states), and inter-entity relationships. This model aims to provide a standardized framework for planning, developing, and analyzing human-robot interactions, fostering clearer communication and more robust robot designs.
The field of Human-Robot Interaction (HRI) is rapidly growing, with robots becoming increasingly integrated into our daily lives. A crucial aspect of designing effective and appropriate robot behavior is understanding the “social context” of an interaction. However, a new research paper highlights a significant challenge: the term “social context” is used in many different ways across HRI research, leading to confusion and making it difficult for researchers to connect their work.
To address this, researchers Sydney Thompson, Kate Candon, and Marynel Vázquez from Yale University have conducted a comprehensive survey of existing HRI literature. Their goal was to identify how “social context” is currently defined and used, and then to propose a clear, conceptual model for describing it in human-robot interactions. This model aims to provide a common language and framework for the HRI community.
Understanding the Current Landscape
The paper reveals that while many studies acknowledge the importance of social context, very few explicitly define it. Among the handful that do, definitions vary widely. Some researchers consider social context to include the physical environment (like a home or workplace), the organization of the interaction (one-on-one or group), and the roles or statuses of participants. Others focus more narrowly on the physical behavior of agents, such as their positions and velocities.
Beyond explicit definitions, the term is used broadly to describe problem areas or domains (like healthcare robotics) or more specifically as a synonym for social norms. The researchers categorized 257 sentences from 58 papers, finding common usages referring to the state of society, specific domains, tasks, social settings (beliefs, norms, roles), physical settings, or a combination of social and physical settings.
A New Conceptual Model for Social Context
The core contribution of this research is a new conceptual model designed specifically for human-robot interactions. It defines a human-robot interaction as an exchange between at least one human and one robot, involving a sequence of actions related to a task within a given environment. The social context of this interaction is then defined as the set of attributes of the relevant agents (humans, robots, or even other entities like pets), their environments, and the associations between them, all of which influence the interaction.
This means that anything that affects how humans and robots act during their interaction is considered part of the social context. For example, in a scenario where a firefighter teleoperates a robot in a disaster area, the robot’s environment (fire, debris) and the firefighter’s environment (noise, distractions) both influence the interaction. If another person is found, their attributes also become part of the social context, demonstrating that social context is dynamic and can change over time.
Categorizing Socially Contextual Information
To make their model practical, the authors developed detailed taxonomies for different types of “socially contextual information” – the specific attributes that make up the social context. These are divided into three main categories:
-
Environment Attributes: This includes the physical location (public space, specific room), objects present (task-related items, furniture, obstructions), behavior constraints (safety rules, social norms of the place), and physical properties (room size, layout, time of day, brightness, volume).
-
Agent Attributes: These describe the characteristics of humans and robots. They include actions (verbal communication like utterances and tone, nonverbal actions like gaze and gestures, and strategic actions like goals and planned trajectories), common population characteristics (age, gender, race, education, diagnoses), physical state (location, body pose, motion), appearance (robot design, color, morphology, screen display), and mental states (cognitive factors like workload and abilities, social/behavioral attributes like roles and experiences, motivational states like goals and needs, and feelings like stress and emotions).
-
Association Attributes: These describe the relationships between entities. They cover physical relationships (distance between agents or objects) and social relationships (partner status, group membership, authority). Also included is the medium through which an agent experiences an environment (in-person, teleoperation, virtual reality), and associative mental states, which are internal states involving more than one entity, such as expectations, beliefs, perceptions, trust, novelty, interest, and attitudes towards others.
Also Read:
- Unpacking Social Dynamics: A New Framework for Evaluating Digital Human Behavior
- Bridging the Language Gap: How AI is Enabling Robots to Understand Human Instructions
Implications for the Future of HRI
This conceptual model offers significant benefits for HRI practitioners and researchers. It can serve as a checklist for planning human-robot interactions, helping designers anticipate novel situations and challenges. By considering a wide range of potential socially contextual information, researchers can better prepare for how robots will be adopted and perceived in the real world.
For robot behavior generation, the model provides a framework for developing autonomous systems that can adapt to different social contexts, leading to improved user experiences. While computationally challenging, the authors suggest that leveraging relational abstractions and advanced machine learning models could help robots understand and respond to complex social cues.
Finally, the model aids in post-interaction analyses, helping researchers understand why certain robot behaviors were appropriate or not, identify potential confounds in experiments, and define what “generalization” truly means in HRI. Ultimately, a clearer understanding of social context is crucial for building social robots that can behave ethically and effectively in novel, real-world situations.
For more details, you can read the full research paper here.
