PaperlayerView exact revision on arXiv

Source-linked AI summary

Perceived Safety in Physical Human Robot Interaction -- A Survey

Matteo Rubagotti, Inara Tusseyeva, Sara Baltabayeva, Danna Summers, Anara Sandygulova

arXiv:2105.14499v1cs.RO

TL;DR

As autonomous systems increasingly share physical space with people, perceived safety matters alongside actual safety for human–robot acceptance. This survey synthesizes terminology, assessment methods, and evidence across six autonomous-system categories, finding recurring relationships between perceived safety and system motion or characteristics. It also identifies a scope limitation: some factors are absent from studies because experimental conditions made them irrelevant.

  • Problem

    Autonomous systems increasingly share physical space with people, creating a need to understand perceived safety alongside actual safety.

  • Method

    The survey reviews 114 papers using prior surveys, citation expansion, direct database searches, and criteria requiring physical interaction, perceived-safety assessment, and analysis of robot behavior.

  • Results

    Across six autonomous-system categories, the survey synthesizes terminology, assessment methods, experimental conditions, and relationships between perceived safety and robot motion or characteristics.

  • Takeaways & Limitations

    Motion fluency and predictability recur across robot types, while other perceived-safety factors depend on the robot category and interaction scenario.

  • Takeaways & Limitations

    Some factors are neglected in parts of the literature because experimental conditions make them irrelevant, such as robot speed when mobile robots always move slowly.

Abstract

from arXiv · show

This review paper focuses on different aspects of perceived safety for a number of autonomous physical systems. This is a major aspect of robotics research, as more and more applications allow human and autonomous systems to share their space, with crucial implications both on safety and on its perception. The alternative terms used to express related concepts (e.g., psychological safety, trust, comfort, stress, fear, and anxiety) are listed and explained. Then, the available methods to assess perceived safety (i.e., questionnaires, physiological measurements, behavioral assessment, and direct input devices) are described. Six categories of autonomous systems are considered (industrial manipulators, mobile robots, mobile manipulators, humanoid robots, drones, and autonomous vehicles), providing an overview of the main themes related to perceived safety in the specific domain, a description of selected works, and an analysis of how motion and characteristics of the system influence the perception of safety. The survey also discusses experimental duration and location of the reviewed papers as well as identified trends over time.

1. Introduction

The survey examines perceived safety as a requirement for human acceptance of autonomous physical systems sharing space with people. It reviews terminology, assessment methods, robot categories, and relationships between robot behavior, characteristics, and perceived safety.

  • Motivation: Perceived safety complements physical safety because robots must be perceived as safe to be accepted as partners and co-workers.
  • Related work: Earlier reviews addressed assessment methods or broader pHRI safety, while drone and autonomous-vehicle surveys had focused on actual safety rather than its perception.
  • Research questions: The survey asks which terms express perceived safety, how it is measured, and how robot types, motion, characteristics, and experimental conditions relate to it.
  • Selection criteria and assessment: Assessment methods include behavioral observation, physiological measurements, questionnaires, and direct input devices in physical interaction studies.
  • Method: The review identifies papers through prior surveys, citation expansion, and direct database searches, a challenge because perceived-safety terminology is not unified.
  • Contribution and organization: The survey reviews 114 papers focused entirely on perceived safety and extends coverage to drones and autonomous vehicles alongside four robot categories.

2. Defining perceived safety

Perceived safety is described through overlapping synonyms and related concepts, including trust and comfort, while stress, fear, anxiety, and surprise express its absence or erosion. The survey organizes these terms and distinguishes valence/arousal as an alternative emotional representation.

  • Terminology: The survey first defines perceived-safety terminology broadly in psychology and then narrows those definitions to physical human–robot interaction.
  • Synonyms and related concepts: Psychological, mental, and subjective safety are presented as synonymous expressions of perceived safety, whereas trust and comfort capture related aspects.
  • Trust: In this survey, trust concerns how much people believe a robot will not harm them, distinct from trusting it to complete an assigned task.
  • Comfort: Comfort is treated as a related concept involving well-being and socially acceptable robotic assistance, rather than an exact synonym of perceived safety.
  • Lack of perceived safety: Stress, fear, anxiety, and surprise are affective responses considered when they derive from a lack of perceived safety during physical interaction.
  • Valence and arousal: Valence measures whether emotion is positive or negative, while arousal measures its strength, providing an alternative to discrete emotion categories.

3. Assessment methods

Perceived safety in pHRI is assessed through questionnaires, physiological measurements, behavioral assessment, and direct input devices. The survey describes how these methods are applied across robot types and notes that combining methods can improve reliability.

  • Assessment categories: Four assessment categories are used: questionnaires, physiological measurements, behavioral assessment, and direct input devices.The survey presents these categories in a taxonomy and discusses their combinations across robot types.
  • Physiological signals: Physiological assessment was mainly used with industrial manipulators and was absent for mobile robots and drones in the reviewed table.The survey attributes this distribution partly to the difficulty of collecting physiological measurements outdoors or while participants move.
  • Questionnaires: Questionnaires gather self-reports about participants’ knowledge, attitudes, behavior, and human–robot interaction experiences.Common instruments include GSQ, NH-33, NARS, RAS, and RoSAS, which measure constructs such as perceived safety, anxiety, negative attitudes, and discomfort.
  • Physiological signals: Physiological methods measure signals such as heart rate, galvanic skin response, and eye gaze to assess stress, arousal, fear, anxiety, or situation awareness.Physiological signals can complement subjective measures because participants cannot consciously manipulate autonomic nervous-system activity.
  • Behavioral assessment: Behavioral assessment commonly uses human–robot distance as an indicator of perceived safety, drawing on proxemics and recordings of interaction behavior.The relative distance people maintain from a robot is treated as intuitively inversely proportional to their sense of safety.

4. Industrial manipulators [17–45]

Studies of industrial manipulators relate perceived safety to motion, distance, predictability, communication, and user experience. Across the reviewed works, greater distance, lower speed, fluent motion, and advance information generally supported safer perceptions, while responses could vary between participants.

  • Selected works: A safe motion planner reduced surprise and anxiety relative to a potential-field planner, especially when the robot moved at high speeds.Higher robot speeds increased estimated human arousal, while the planner’s danger criterion improved responses on surprise and anxiety.
  • Safety systems: Robot safety systems based on monitored stop and power-and-force limiting were generally perceived as safe in two experimental studies.The finding came from questionnaire-based studies with a wide range of participants.
  • Motion and distance: Perceived safety generally increased with larger human–robot distance and lower robot speed, with reported thresholds of 2 m and 0.5 m/s in one study.Subjects could accept higher speed or acceleration when the robot was farther away.
  • Predictability and communication: Fluent and predictable motion, advance communication, and user control over when motion starts were associated with greater comfort and perceived safety.During handovers, force control and avoidance of abrupt motion were also reported as safety-supporting features.
  • User factors: Perceived safety improved with prior task experience or information about robot safety features, but motion preferences were not fully generalizable across participants.For example, temporal scaling was perceived as safe by one participant group and unsafe by another.

5. Mobile robots [46–58]

Mobile-robot studies examine comfort and perceived safety through approach direction, interpersonal distance, positioning, predictability, and user experience. Preferences depend on the interaction context, including whether people cooperate with the robot or perform an independent task.

  • User experience: Prior experience with the robot was associated with closer approach distances, including averages of 51 cm for experienced participants and 73 cm for inexperienced participants.Both distances fell within the reported 40–80 cm range used for some human–human interactions.
  • Predictability and robot characteristics: More socially interactive behavior was not necessarily more comfortable because lower motion predictability could reduce comfort.Participants were uncomfortable when robot motion was unpredictable, and larger robot size was also associated with discomfort.
  • Approach direction: Front-left and front-right approaches were generally most comfortable, whereas approaches from behind were most uncomfortable.Direct frontal approaches were context-dependent, becoming uncomfortable when participants sat on a chair or leaned against a wall.
  • Distance and positioning: In cooperative tasks, participants preferred personal-zone distances of approximately 46–80 cm and face-to-face positioning.The preferred distance estimate is based on the intersection of findings from two reviewed studies.
  • Distance and positioning: During independent tasks, participants could feel discomfort when the robot approached within 3 m, whereas cooperative tasks supported closer personal-zone distances.Cultural background also affected the distance at which participants felt comfortable.

6. Mobile manipulators [59–65]

Mobile-manipulator studies show that perceived safety depends on motion speed, distance, force, predictability, and the robot’s apparent intent during close physical collaboration.

  • Scope and methods: Mobile manipulators were evaluated in handover and collaboration tasks using both real-world robots and virtual-reality tools.The reviewed platforms included Jido, HERB, and Care-O-bot 3.
  • Intent and communication: Robot-initiated touch was more comfortable when participants believed it was instrumental rather than affective.Participants also reported higher comfort without a verbal warning, suggesting that warnings require careful design.
  • Motion and interaction factors: Participants felt less safe when robots moved quickly near their hands, applied high force, or behaved unpredictably.These effects were reported particularly during object handovers and close physical interaction.
  • Motion and interaction factors: Predictable and legible motions were perceived as safer than functional-only motions during a cup-preparation task.Legibility conveyed intent, but did not increase perceived safety or comfort beyond predictability alone.
  • Motion and interaction factors: Perceived safety improved with greater human-robot distance and slower motion, especially near the face or hands.In one study, participants allowed approach to 57 cm when the robot moved slowly or they had prior task experience.

7. Humanoid robots [66–99]

Humanoid-robot studies examine proxemics, appearance, motion, social cues, and prior experience as determinants of perceived safety, comfort, and trust in close interaction.

  • Scope and tasks: Humanoid-robot studies covered close-distance tasks including handovers, approaching, touch, handshaking, games, and pick-and-place.The reviewed platforms included Robovie, HRP 2, Nao, PR2, Baxter, iCub, and others.
  • Proxemics and individual factors: People’s preferred distance varied with robot appearance, height, approach direction, gaze, gender, prior experience, and attitudes toward robots.Men, robot-experienced participants, and pet owners generally allowed closer approaches in several studies.
  • Motion: Slow motion was generally more comfortable, while a humanoid moving at 40 inches/sec was rated most uncomfortable and 10 inches/sec most pleasant.The studies compared mobile and humanoid robots in approach and daily-life behavior scenarios.
  • Appearance and morphology: Participants kept the closest distance from ASIMO compared with Robovie or a human, while Nao was allowed closer than short or tall PR2 robots.Robot appearance and form could matter more than height in some comparisons.
  • Robot behavior and cues: Higher Baxter arm stiffness increased perceived safety, whereas physical reactivity reduced pleasantness, energy, and dominance.Facial reactivity made Baxter seem more pleasant and energetic during a hand-clapping game.

8. Drones [100–115]

Drone studies link perceived safety to proximity, speed, movement predictability, physical design, noise, communication cues, social framing, and cultural context.

  • Scope and methods: Drone studies used real and simulated UAVs for photography, delivery, monitoring, mapping, and human-drone interaction tasks.Platforms included DJI Phantom, Parrot, AscTec, and simulated drones in VR and CAVE environments.
  • Proximity and social design: A social drone was allowed to approach to 1.06 meters on average, 30% closer than a non-social drone without voice.Participants also differed by gender and altitude, and drone noise increased mental stress.
  • Proximity and social design: Participants stopped an AscTec Hummingbird at 65.5 cm on average, compared with 36.5 cm for a Double telepresence robot.The comparison indicates that perceived proximity varied across robot types.
  • Communication: Intent-signaling mechanisms, gestures, and LED feedback were generally effective and increased perceived safety or acknowledgement.Examples included blinkers, beacons, thrusters, gaze, waggle, nod, toss, orientation, and visual feedback.
  • Motion and design: Steady, predictable motion, social gestures, safeguards, face cues, and feedback lights increased safety, whereas noise, wind, sudden speed changes, unstable maneuvers, and unusual shapes reduced comfort.Medium-sized drones were generally preferred, regardless of autonomy level.
  • Design and culture: Safe-to-touch and social drone designs increased perceived safety, with social drones entering people’s space 30% closer than non-social drones.Cultural differences also appeared: Chinese participants were less likely than US participants to compare drones to pets.

9. Autonomous vehicles [116–130]

Autonomous-vehicle studies assess perceived safety through trust, anthropomorphic and explanatory interfaces, vehicle behavior, communication displays, and real riding or crossing experiences.

  • Scope and methods: Studies used real autonomous vehicles, driving simulators, videos, virtual reality, and Wizard-of-Oz experiments with drivers and pedestrians.Real vehicles were considered more ecologically valid, while simulations and videos were also widely used.
  • Interfaces and trust: Semantic speech output increased trust, anthropomorphism, and acceptance of autonomous vehicles.Other studies similarly examined names, gender, voice, and interfaces visualizing the vehicle’s interpretations and actions.
  • Pedestrian communication: Pedestrians prioritized distance to the vehicle (56%), speed (46%), and traffic density (24%) when deciding whether to cross.Nearly half of participants, 46%, said vehicle displays would be helpful when autonomous vehicles become available.
  • Vehicle behavior: Pedestrians based crossing decisions mainly on vehicle behavior, such as maintaining speed or slowing down, regardless of driving mode or vehicle appearance.When vehicle intent was unclear, differences between vehicles emerged at certain distances.
  • Individual differences: Older participants reported more positive ratings and higher trust than participants below 30 years of age.The survey also found that human drivers felt comfortable following autonomous vehicles closely and placed less weight on crash risk.
  • Passenger experience: Passengers associated perceived safety with low speeds of 8, 16, and 20 km/h, smooth control, environmental awareness, emergency buttons, and trust in technology.Concerns could return for large unattended buses operating at regular route speeds.

10. Discussion

Across robot types, perceived safety is shaped by motion, spatial relation, appearance, predictability, communication, and contact quality, while studies generally favor short laboratory experiments. Publication activity has expanded, especially for several robot categories in recent years.

  • Factors determining perceived safety: Greater distance and lower robot speed are generally associated with higher perceived safety.
  • Factors determining perceived safety: Higher speeds can remain acceptable when the robot is farther from the human, while approach direction, size, and appearance also influence perceived safety.
  • Factors determining perceived safety: Fluent, predictable motion, advance communication cues, and smooth contacts improve perceived safety across applicable interaction contexts.
  • Factors determining perceived safety: Some factors are ubiquitous, whereas others depend on robot type or experimental conditions; for example, speed was not studied for mobile robots whose experimental speeds were consistently low.
  • Experiment duration and location: Short-term experiments lasting minutes to hours were generally preferred, while studies lasting up to two months examined long-term habituation.
  • Experiment duration and location: Most experiments occurred in laboratories, with additional studies in factories, living environments, public events, virtual environments, and driving simulators.
  • Publication trends: Industrial manipulators were studied first; mobile-robot research surged in 2005–2008, while several other robot categories increased sharply in recent years.

11. Conclusions

The survey identifies inconsistent terminology and synthesizes perceived-safety assessment methods across six robot types, revealing common trends and themes across more than one hundred papers.

  • The survey highlights a lack of uniform terminology and categorizes assessment methods across six robot types.
  • Analysis of more than one hundred papers reveals common trends and themes in perceived safety research.
Loading 2105.14499v1…