PaperlayerView exact revision on arXiv

Source-linked AI summary

Embodiment in Socially Interactive Robots

Eric Deng, Bilge Mutlu, Maja Mataric

arXiv:1912.00312v1cs.ROcs.HC

TL;DR

Socially interactive robots often achieve their goals without physical work, raising questions about when physical embodiment improves social interaction over virtual agents. This paper reviews and systematizes embodiment research, finding evidence that embodiment can enhance communication and perceptions while identifying methodological and design limitations.

  • Problem

    Because most socially interactive robots do not physically interact with their environments, the benefits of physical embodiment over less expensive and complex virtual presence remain unclear.

  • Method

    The paper thoroughly reviews existing embodiment work, analyzes 65 experiments, and organizes the design and interaction space through taxonomies of embodiment, social roles, and human-robot tasks.

  • Results

    Physical embodiment can provide additional communication channels and improve perceptions of robots as more trustworthy, helpful, and engaging than disembodied agents.

  • Takeaways & Limitations

    The review provides a framework for discussing prior research, identifying knowledge gaps, and informing future socially interactive robot design.

  • Takeaways & Limitations

    Embodiment studies are constrained by integrated robotic systems that make individual design variables difficult to isolate, alongside small samples that can produce underpowered findings.

Abstract

from arXiv · show

Physical embodiment is a required component for robots that are structurally coupled with their real-world environments. However, most socially interactive robots do not need to physically interact with their environments in order to perform their tasks. When and why should embodied robots be used instead of simpler and cheaper virtual agents? This paper reviews the existing work that explores the role of physical embodiment in socially interactive robots. This class consists of robots that are not only capable of engaging in social interaction with humans, but are using primarily their social capabilities to perform their desired functions. Socially interactive robots provide entertainment, information, and/or assistance; this last category is typically encompassed by socially assistive robotics. In all cases, such robots can achieve their primary functions without performing functional physical work. To comprehensively evaluate the existing body of work on embodiment, we first review work from established related fields including psychology, philosophy, and sociology. We then systematically review 65 studies evaluating aspects of embodiment published from 2003 to 2017 in major peer-reviewed robotics publication venues. We examine relevant aspects of the selected studies, focusing on the embodiments compared, tasks evaluated, social roles of robots, and measurements. We introduce three taxonomies for the types of robot embodiment, robot social roles, and human-robot tasks. These taxonomies are used to deconstruct the design and interaction spaces of socially interactive robots and facilitate analysis and discussion of the reviewed studies. We use this newly-defined methodology to critically discuss existing works, revealing topics within embodiment research for social interaction, assistive robotics, and service robotics.

1 Introduction

Socially interactive robots primarily use social capabilities rather than functional physical work, making the value of physical embodiment over virtual agents an open question. The paper frames this question through the embodiment hypothesis and reviews related research and design implications.

  • Embodiment can add proxemics, oculesics, and gestures as communication channels, potentially increasing perceived trustworthiness, helpfulness, and engagement.
  • Most socially interactive robots achieve their goals without physically interacting with the environment, so the benefits of embodiment over cheaper virtual presence are unclear.
  • The embodiment hypothesis proposes that physical embodiment has a measurable effect on performance and perception in social interactions.
  • The review connects robotics research with philosophy, neuroscience, psychology, sociology, and human-computer interaction perspectives on embodiment and social interaction.
  • The paper introduces terminology and an embodiment taxonomy, then uses the surveyed literature to discuss the hypothesis and implications for future robot design.

2 What is Embodiment?

Embodiment is treated as a cross-disciplinary concept relevant to socially interactive robotics. The paper reviews its treatment across several fields to clarify its relevance to robotic social interaction.

  • Embodiment is studied across philosophy, psychology, neuroscience, communications, and engineering.
  • The review uses these disciplinary perspectives to understand embodiment’s past, present, and future relationship to socially interactive robotics.

2.1 Embodiment in Philosophy and Ethics

Philosophical and ethical perspectives treat embodiment as shaping cognition, affordances, expectations, and limitations. Robot embodiment can support engagement while also creating risks from misleading expectations and attachment.

  • Philosophy: Embodied cognition examines how aspects of the body beyond the brain affect human experience and cognitive processes.
  • Philosophy: An agent’s embodiment constrains its capabilities and determines affordances that shape how it can be used.
  • Ethics: Engaging robot designs can produce undesirable influence, unrealistic expectations, perceived deception, disappointment, or emotional discomfort.
  • Ethics: Attachment to a robot may lead to grief and anxiety when the robot is removed, while misleading designs can encourage inappropriate use and potential injury.
  • Ethics: Classical philosophy establishes foundations for embodiment, while ethics highlights negative consequences associated with some design choices.

2.2 Embodiment in Psychology and Communication

Psychology and communication research examines how physical and virtual cues, presence, and telepresence shape social interaction. These perspectives inform the study of embodiment in social agents and robots.

  • Communication: Presence research links effective embodied-robot design with increased social presence and desired affordances.
  • Communication: Recent communication research focuses on presence in virtual reality and telepresence, alongside explorations of physical embodiment in social agents.
  • Communication: Embodiment studies in robotics address the importance of social presence in both human-human and human-robot interaction.

2.3 Embodiment in Robotics and Design

Robotics has increasingly valued physical embodiment not only for functional manipulation but also for communication in socially interactive systems. This shift creates a design space centered on interaction rather than physical function.

  • Artificial-agent research has developed physically disembodied communicative systems, while socially interactive robots use physical embodiment to improve interaction.
  • Sensorimotor capabilities connect an agent to its environment and provide a richer experiential context spanning biology, psychology, and culture.
  • In HRI, physical components such as independently controlled fingers can acquire communicative value by enabling more complex gestures.
  • The paper characterizes this interaction-centered design space to inform robot designers and future embodiment decisions.
  • Socially interactive robots such as Pepper, Kiwi, and Cozmo exemplify embodiments developed for social interaction rather than traditional robotic functions.

Defining Embodiment for Interactive Agents

The paper defines embodiment for socially interactive robots through their physical relationship with the environment and its connection to sociability and presence. It situates this definition within established embodiment theories and related work on virtual agents, colocation, and collaboration.

  • Ziemke’s six notions distinguish structural, historical, physical, organismoid, organismic, and social embodiment.
  • Structural coupling describes embodiment through bidirectional perturbatory channels between an agent and its environment.
  • Social embodiment treats bodily states such as postures, arm movements, and facial expressions as central to social information processing.
  • The paper combines social embodiment with situated structural coupling to define embodiment for socially interactive robots.
  • Virtual agents use on-screen or immersive virtual embodiments, while physical robots enable colocation that can support collaboration.
  • Socially interactive robots use embodiment primarily for communication, acceptance, and engagement while perceiving and generating communicative signals.

Social Performance and Social Presence in Embodied Robots

Socially interactive robots rely on human-like social characteristics and embodied cues to support interaction. Existing studies report that physically embodied agents possess greater social presence than virtual counterparts.

  • Seven social characteristics include high-level dialogue, model learning, relationship maintenance, natural cues, personality, and social-competency development.
  • Physically embodied agents have been shown to possess social presence to a greater extent than virtual counterparts.
  • Social presence is defined as the degree to which perceptions of an agent or robot shape interaction with it.
  • Physical embodiment provides channels for conveying internal states and intentions through intuitive, human-like communication.
  • Embodied social interaction links perceived stimuli, bodily mimicry, affective states, and performance effectiveness.

Types of Socially Interactive Robots

Socially interactive robots can be organized into seven categories that range from socially evocative and receptive systems to socially intelligent robots with human-level social intellect. The categories distinguish their social activity, environmental understanding, and technological complexity.

  • Socially evocative robots elicit nurturing or caring responses through anthropomorphized agents.
  • Social-interface robots generate and often understand social cues and communication modalities familiar to human users.
  • Socially receptive robots are socially passive and limited in the cues they can learn through their embodiments.
  • Sociable robots proactively interact with humans to complete internal goals.
  • Socially situated robots understand and react to their social environments, while socially embedded robots additionally couple structurally with those environments and understand interactional structures.
  • Socially intelligent robots are defined as the most complex and technologically capable class, possessing human-level social intellect.

2.4 Summary

Embodiment is understood differently across disciplines, with disagreement over how essential physical instantiation is to artificial agents. The paper therefore reviews socially interactive robotics studies to assess the embodiment hypothesis.

  • Philosophers, psychologists, communication theorists, and social scientists have examined embodiment in relation to cognition, human experience, and social expectations.
  • Researchers disagree about embodiment’s importance for artificial agents, ranging from limited benefits to claims that intelligence requires a physical body.
  • The paper reviews socially interactive robotics studies from the past decade and a half to evaluate the current state of the embodiment hypothesis.

3 The Design Space for Socially Interactive Robots

The design space for socially interactive robots is structured by task, social role, embodiment form, and design metaphor. These dimensions connect physical design with expected capabilities and interaction outcomes.

  • 3 The Design Space for Socially Interactive Robots: Robot design spaces define variable elements of appearance, behavior, and system makeup to support constructive discussion and systematic experimentation.
  • 3.1 Social Context: Task and role are core contextual factors because interaction is shaped by the activity performed and the robot’s position in that activity.
  • 3.1.1 Tasks: Reviewed tasks are classified into eight octants using McGrath’s dimensions of generate-negotiate and execute-choose.
  • 3.1.1 Tasks: The task taxonomy includes decision-making, intellective, creative, planning, performance/action, contest, mixed-motive, and cognitive-conflict tasks.
  • 3.1.2 Social Roles: Social roles span subordinate to superior, with peers positioned between them and associated with different perceived abilities and interaction goals.
  • 3.2 Design Paradigms: Embodiment may be physical, virtual, or disembodied, and studies commonly compare physical robots with virtual counterparts to assess interaction outcomes.
  • 3.2 Design Paradigms: Virtual and physical embodiments create different interaction frames: virtual agents offer crafted environments, while physical agents are co-situated with users.
  • 3.2 Design Paradigms: Design metaphors establish interaction expectations, while embodiments vary continuously in abstraction and stylization within those metaphors.

3.3 Behavior Design

Behavior design for socially interactive robots comprises multimodal embodied cues whose timing, form, and coordination shape social interaction. The design space includes gestures, posture, gaze, and facial expressions, but embodiment differences and expression mismatches constrain effective implementation.

  • Behavior Design: Embodied cues are behavioral elements whose specific forms and timing provide design variables for effective socially interactive robots.The design space is grounded in human-human multimodal communication and includes cues performed through combinations of embodied features.
  • Limb-Based Gestures: Limb-based gestures communicate ideas and support interaction through iconic, metaphoric, beat, cohesive, and deictic categories.These gestures use hand, arm, or head movements and can communicate semantic or abstract content and establish joint attention.
  • Posture: Posture conveys attitude, status, and internal state, but robot hardware constrains expressive gestures and makes cross-embodiment gesture mapping an open challenge.Different embodiments affect posture and gesture realization, limiting straightforward transfer of semantic gestures across robot forms.
  • Gaze: Gaze communicates attention and mental or emotional states while supporting turn-taking, joint attention, and conversational regulation.Effective gaze cues are associated with improved information recall, conversational-floor management, and task-collaboration efficiency.
  • Facial Expressions: Facial expressions strongly influence perception, but designing them is challenging because incongruence with speech or inappropriate timing can harm interaction.Ill-timed smiles and unrealistic or mismatched expressions can invoke the Uncanny Valley phenomenon or confuse interaction partners.
  • Design Space: The section characterizes embodied behavior as a design space spanning industrial design, interaction design, and animation, with carefully normed cues supporting rich and effective interactions.This characterization is intended as a tool for designers and researchers developing embodied interactive agents.

4 Embodiment Study Outcomes and Design Implications

The review finds strong overall support for physical embodiment, while showing that benefits depend on task type, social role, and whether outcomes concern agent perception or task performance. Its taxonomies and measurement framework are used to identify design implications and explain neutral or negative findings.

  • Review scope: 65 experiments were reviewed, comparing physically or strongly embodied agents with comparable virtual or weakly embodied agents across varied embodiment designs and social roles.The review included 50 two-embodiment comparisons, 11 three-embodiment comparisons, and 4 comparisons involving more than three embodiments.
  • Overall outcomes: 63.1% of combined results were solely positive, while 15.4% were mixed positive, 15.4% neutral, 1.5% mixed negative, and 4.6% solely negative.Overall, 63.1% showed improvements in interaction and performance, compared with 6.1% showing negative results.
  • Task fit: Embodiment benefits are task-specific: disembodied agents may suffice for information transmission or personal disclosure, whereas relationship-oriented tasks benefit from increased social presence.Observed preferences and ratings did not always align, reinforcing the need to analyze task performance and agent perception together.
  • Agent perception: 43 of 57 experiments (75.4%) found physical embodiment superior for improving user perceptions of the agent, with every task category containing a majority favoring physical embodiment.The review interprets these findings as strong support for embodiment in perceived social competence.
  • Design implications: Neutral and negative findings clustered by social role within task categories, indicating that embodiment design must appropriately match a robot’s role to the task.Creative tasks appeared especially sensitive to social role, while peer-like roles could provide insufficient confidence or support in some performance-and-actions tasks.
  • Task performance: 37 of 57 task-performance experiments (71.15%) reported positive results for physically embodied robots over virtual or disembodied agents.Examples include higher negotiation scores, greater compliance with unusual requests, and improved learning outcomes when interacting with physical robots.

5 Recommendations for Future Embodiment Studies

Future embodiment studies should combine better-matched study designs, broader settings, richer measurement, and stronger attention to integrated robot systems and sample-size constraints.

  • Study Designs: Studies should balance laboratory control against ecological validity by expanding from controlled experiments toward in situ and naturalistic settings.Laboratory studies support control but limit generalization; future work should select methods that fit the deployment context.
  • Study Designs: System-level designs can model many design variables simultaneously, enabling finer-grained links between robot features and interaction outcomes than isolated manipulations.One study found that each standard-deviation increase in pointing gestures predicted information-recall increases of 0.123 and 0.623 standard deviations for females and males, respectively.
  • Measurements: Measurement should combine behavioral and physiological observations with validated self-report instruments, while recognizing that attitudes and experiences require participant reports.Existing robot-interaction surveys often lack established validity and reliability, and qualitative approaches remain underused despite their value in naturalistic settings.
  • Hypotheses: Researchers should align hypothesis development with prior research, pilot data, and explicit design goals before selecting variables and outcomes.Controlled laboratory and in situ studies use a priori independent and dependent variables to construct testable hypotheses.
  • Limitations and Open Issues: Future studies must address integrated-system constraints, small samples, and limited generalizability when isolating low-level embodiment features.Robotics introduces perception, action, repeatability, robustness, hardware-cost, and sampling challenges; small samples can produce underpowered findings.

6 Implications for Designing Embodiment

The paper proposes using prior study results and a formalized task, social-role, and embodiment design space to guide socially interactive robot selection and development. Its process orders decisions from task and social role through embodiment design and behavior implementation.

  • Interaction requirements: Socially interactive robots must combine human-oriented perception, natural HRI, readable social cues, and real-time performance to support effective interaction.These requirements include interpreting human behavior, following social norms, communicating internal states, and operating at human interaction rates.
  • Design process: Past studies can inform selecting, modifying, or building robot platforms for new socially interactive robotics studies and deployments.The paper frames aggregated prior results as support for robot-selection and design processes.
  • Evidence for design: The review’s results reveal benefits of physical embodiment across task contexts and patterns that can guide socially interactive robot design.The authors group experimental designs and results to create a structured, evidence-informed design process.
  • Design process: The proposed process begins with the application task, selects an appropriate social role, chooses design inspiration and behaviors, and implements the design at a suitable abstraction level.Figure 6.1 presents this sequence as an application-specific design approach.
  • Embodiment selection: A given task may support multiple social roles, while the same role may be implemented through different design metaphors and abstraction levels.This flexibility allows designers constrained by one design dimension to use existing data when selecting the other.
  • Implications: The proposed characterization aims to support robot selection, new robot development, experimental design, novel applications, and iterative design.The scope extends beyond embodiment comparison to broader future design and research activities.
  • Implications: For one reviewed context, results indicated with relatively high confidence that a social role between peer and superior was the best choice.The selected role was then implemented in a robot embodiment following the proposed design process.

A Reviewed Studies

The reviewed studies are organized through tables covering measures, robot embodiments, task categories, social roles, platforms, participant demographics, and reported outcomes. These materials document how the review’s studies were labeled and evaluated.

  • Measures: Table 4.1 lists measures and instruments used to capture participant perceptions of robots.One listed instrument is the Networked Minds Questionnaire for social presence.
  • Measures: Tables 4.2 and 4.3 cover individual-behavior measures and observed interaction measures used in the reviewed studies.The tables identify measures and methods for capturing or computing behavioral and interaction outcomes.
  • Embodiment classification: Table 6.1 labels robots used in reviewed studies by design metaphor and level of abstraction.The same embodiment dimensions are used to characterize the reviewed robot systems.
  • Study classification: Tables A.1 and A.2 label studies by task category, social role, and physical or virtual robot platform.Social roles in Table A.1 use a numeric scale from 1, subordinate, to 9, superior.
  • Results: Table A.3 reports reviewed-study results broken down by task-performance and interaction-performance differences.The table provides the review’s organized outcome comparison across these performance categories.
  • Study characterization: Tables A.4 and A.5 document participant demographics and whether studies used observed measures or self-reported measures.Table A.5 distinguishes task performance, interaction performance, and individual behavior measures.
Loading 1912.00312v1…