PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey on Socially Aware Robot Navigation: Taxonomy and Future Challenges

Phani Teja Singamaneni, Pilar Bachiller-Burgos, Luis J. Manso, Anaís Garrell, Alberto Sanfeliu, Anne Spalanzani, Rachid Alami

arXiv:2311.06922v4cs.RO

TL;DR

Existing surveys leave a need for broader robot coverage and a more inclusive definition of socially aware navigation. The authors classify 193 articles across four taxonomies and report that situation-awareness work most often addresses obstacles, agent-trajectory prediction, and proxemics, while none considers all three branches.

  • Problem

    Existing surveys leave a need for a comprehensive account spanning broader robot types and an inclusive definition of socially aware robot navigation.

    Source: §1 · p. 1

  • Method

    The authors reviewed collected articles through a multi-pass, tag-based process and classified 193 articles across four faceted taxonomies.

    Source: §2.1 · p. 4, §9 · p. 25

  • Results

    Obstacles, trajectory prediction of agents, and proxemics are the most frequent situation-awareness aspects, and none of the reviewed proposals considers all three branches.

    Source: §5 · p. 15

  • Takeaways & Limitations

    The survey's classification identifies areas needing more attention, while mobile robots remain the main domain among the papers reviewed.

    Source: §9 · p. 25

    Research on implicit and explicit communication in socially aware navigation is limited, and many studies focus on one form rather than their interaction.

    Source: §4.1 · p. 10

Abstract

from arXiv · show

Socially aware robot navigation is gaining popularity with the increase in delivery and assistive robots. The research is further fueled by a need for socially aware navigation skills in autonomous vehicles to move safely and appropriately in spaces shared with humans. Although most of these are ground robots, drones are also entering the field. In this paper, we present a literature survey of the works on socially aware robot navigation in the past 10 years. We propose four different faceted taxonomies to navigate the literature and examine the field from four different perspectives. Through the taxonomic review, we discuss the current research directions and the extending scope of applications in various domains. Further, we put forward a list of current research opportunities and present a discussion on possible future challenges that are likely to emerge in the field.

1 Introduction

The survey addresses socially appropriate navigation as robots increasingly share spaces with people, and offers a multi-faceted review to help readers navigate research across robot types, methods, and evaluation. It defines the scope of socially aware navigation, reviews a representative literature set, and identifies how its classifications extend earlier surveys.

1 Introduction

  • The survey reviews socially aware navigation across robots, emphasizing how implementation differences affect human perceptions of comfort and safety.Its proposed classifications aim to help developers, application designers, and evaluators find relevant work and support interdisciplinary cooperation.Source: §1 · p. 1

1.1 Socially Aware Robot Navigation

  • A socially aware robot detects people and prioritizes their safety, minimizes disturbance and confusion, signals its intentions, and handles conflicts socially.The authors consider the first two properties minimal requirements; predicting context-dependent human intentions and negotiating conflicts remain difficult, so most research focuses on the first three.Source: §1.1 · p. 2

1.2 Article Collection and Preliminary Analysis

  • The survey selects representative work from the past 10 years, drawing on more than 200 articles while noting fast-growing publication interest over the past 6 years.The collection used IEEE Xplore, ACM Digital Library, and Google Scholar; mobile robots dominate the surveyed literature, alongside work on autonomous vehicles and drones.Source: §1.2 · p. 2

1.3 Previous Surveys

  • The survey extends earlier reviews with a multi-dimensional taxonomy spanning robot types, planning and decision-making, situation awareness and assessment, and evaluation methods and tools.It considers how robot-specific features influence navigation and presents an inclusive classification intended to give a holistic view of the field.Source: §1.3 · pp. 3–4
  • The review adds deeper analysis of environmental and task context, human-intention detection, robot-intention communication, and qualitative versus quantitative evaluation.It focuses on socially aware navigation and classifies evaluation methods to illuminate limitations and challenges in existing evaluation approaches.Source: §1.3 · pp. 3–4

2 Proposed Taxonomies

The survey organizes socially aware navigation literature through four separate taxonomies: robot type, planning and decision-making, situation awareness and assessment, and evaluation and tools. The authors use deductive thematic analysis to classify the reviewed articles.

2.1 Review Process

  • The review iteratively tagged and revised articles, excluding papers where socially aware navigation was not central or contributions were incremental.Taxonomy revisions prompted revisiting the literature and reassigning tags.Source: §2.1 · p. 4
  • The final review comprised 193 articles, each summarized and described in terms of the aspects represented by its assigned tags.Source: §2.1 · p. 4

2.2.1 Taxonomy for Robot type:

  • Robot-type classification distinguishes ground, aerial, and aquatic robots; leaf facets are mutually exclusive for most works, and aquatic navigation had no included works.Robot-specific norms affect socially aware navigation strategy design.Source: §2.2.1 · p. 4, Definitions · p. 5

2.2.2 Taxonomy for Planning and Decision-Making:

  • Planning and decision-making covers communication, navigation-task roles, negotiation, local motion generation, and global motion decision-making, with facets that are not mutually exclusive.Task roles are independent, assistive, or collaborative; communication may be implicit, explicit, or both.Source: §2.2.2 · p. 5

2.2.3 Taxonomy for Situation Awareness and Assess-

  • Situation-awareness classification focuses on representations and predictions of agents and other modeled elements, including environment, social norms, and factors affecting navigation decisions.Its categories include environment, agents, and social norms such as proxemics, comfort, safety, and human preferences.Source: §2.2.3 · pp. 5–6

2.2.4 Taxonomy for Evaluation and Tools:

  • Evaluation-and-tools classification distinguishes qualitative and quantitative evaluation and tools including simulators, datasets, studies, and metrics or benchmarks.Source: §2.2.4 · p. 6

3 Types of robots

The survey classifies socially aware navigation across mobile robots, autonomous vehicles, and aerial robots, with mobile robots accounting for most of the reviewed papers. It describes varied ground platforms and emerging aerial-robot work, including distinct communication and navigation challenges.

3 Types of robots

  • Of 156 papers, 116 concern mobile robots, 22 autonomous vehicles, and 18 aerial robots; over 80% use robots physically or in simulation.Source: §3 · p. 6, Figure 7 · p. 7

3.1 Ground Robots

  • Ground platforms span wheeled humanoids and simple bases to wheelchairs, autonomous cars, and legged robots; early work often used humanoids for guidance or assistance tasks.Source: §3.1 · pp. 7–8
  • Autonomous-vehicle work is newer and addresses pedestrian modeling and cooperation with human drivers; mobile robots also use screens, lights, head motion, or haptics for explicit signaling.Source: §3.1 · p. 8

3.2 Aerial Robots

  • Multi-rotor drones dominate aerial work, while winged drones remain little explored; noise, wind, and unfamiliarity can undermine perceived safety, comfort, and reliability.Source: §3.2 · p. 9
  • Aerial approaches transfer mobile-robot costs (Truc et al. 2022), adapt 3D social forces (Garrell et al. 2017; Carretero 2017; Garrell et al. 2019), or learn trajectories (Yoon et al. 2019).Yao et al. (2017, 2019) used blimps to follow people indoors while reacting to gestures.Source: §3.2 · p. 9

4 Planning and Decision-Making

This section surveys socially aware navigation planning and decision-making through communication, negotiation, navigation-task type, and local and global motion methods. It distinguishes sensor- and surroundings-based local decisions from coarse global trajectory planning, covering learning and other low-level local methods as well as adaptation strategies and task-dependent global planning.

4 Planning and Decision-Making

  • All 149 methodological papers matched at least one planning and decision-making criterion; Figure 11 presents their distribution.The survey frames planning broadly, including classical techniques, local motion generation, global motion, communication, negotiation, and navigation-task type.Source: §4 · p. 9, Figure 11 · p. 10

4.1 Communication

  • Bidirectional communication helps robots and people share intentions to support safety, efficiency, and comfort, but explicit and implicit strategies remain limited and can be difficult to integrate.Explicit communication includes speech, written instructions, displays, and gestures; implicit cues can be challenging to interpret, especially in complex interactions, and robots may not convey human nonverbal signals equivalently.Source: §4.1 · pp. 9–10

4.2 Types of Navigation Task

  • Navigation tasks are independent, assistive, or collaborative and may shift with context; collaboration requires shared goals, agent modeling, intention awareness, and behavioral adaptation.Collaborative methods include rule-based approaches, which are predictable but less adaptable in changing settings, and learning methods, which adapt but can require substantial data and computation.Source: §4.2 · p. 11
  • Assistive navigation uses AI and machine learning for context-aware support, while independent navigation moves autonomously through crowds but still faces social-context, multimodal-sensing, safety, and privacy challenges.Assistive approaches include deep learning for human-friendly, location-specific instructions and reinforcement learning to adapt to people and changing surroundings.Source: §4.2 · pp. 11–12

4.3 Negotiation

  • Negotiation ranges from explicit requests or stated intentions to implicit adaptation based on detected intentions, and robots may need to adjust strategies as circumstances change.It involves interpreting pedestrian needs, communicating robot goals and constraints, and coordinating actions or resolving conflicts to reach workable outcomes.Source: §4.3 · p. 12

4.4 Local Motion Generation

  • Planning-based local motion methods generate trajectories or velocity commands using techniques such as MPC, dynamic windows, elastic bands, and dynamic-obstacle avoidance.MPC optimizes trajectories over a finite horizon; dynamic-window methods select among reachable velocities, while timed elastic bands can account for kinodynamic constraints.Source: §4.4 · pp. 12–13
  • Force-based methods derive commands from attractive and repulsive forces, whereas learning-based methods often generate velocities directly without constructing a precise trajectory.Social-force methods model interactions among people and robots; learning approaches include reinforcement, deep, imitation, and inverse reinforcement learning, and some combine learning with social-force models.Source: §4.4 · pp. 13–14
  • Local motion generation uses learning approaches, including deep reinforcement learning, imitation learning, and inverse reinforcement learning, for navigation and social behaviors.Examples include DRL combined with SFM to study rewards and human-motion prediction (Gil and Sanfeliu 2019, 2022), warehouse human avoidance (Guldenring et al. 2020), pedestrian-aware AV decisions (Deshpande et al. 2020), and traffic coordination and altruism (Toghi et al. 2021, 2022). Imitation learning supports robot navigation and drone flight (Fahad et al. 2020; Liu et al. 2020; Garrell et al. 2019), while inverse RL is used for human approach and public-space navigation (Ramirez et al. 2016; P´erez-Higueras et al. 2014); learned models can capture motion dynamics, social norms, and human feedback.Source: §4.4 · p. 14
  • Other low-level methods address social safety through human-aware flying-robot planning, behavior-based distance and trajectory prediction, contact response, and pedestrian-density speed control.The approaches are described by Yoon et al. (2019), Bera et al. (2017), Paez-Granados et al. (2022), and Jiang et al. (2022), respectively; Bera et al. use video-observed personality traits in a Bayesian model.Source: §4.4 · p. 14

4.5 Global Motion Decision-Making

  • Global motion decision-making computes coarse trajectories that account for the environment, collision avoidance, and bystander needs, using planning-based or learning-based approaches.Unlike local decision-making, it does not rely only on sensors and immediate surroundings.Source: §4.5 · p. 14
  • Global planning commonly uses search or sampling methods, while changing environments motivate sensory feedback, combined global and local maps, and continuous replanning.Examples in the survey include A*, D*, diffusion maps, Dijkstra, PRM, RRT, Risk-RRT, PRM-RRT, and Fast Marching methods (Luo et al. 2018; Charalampous et al. 2016; Chen et al. 2017; P´erez-Higueras et al. 2014; Truong and Ngo 2018; Korkmaz 2021; Becerra et al. 2020; Shrestha et al. 2015; Narayanan et al. 2018; Vega-Magro et al. 2017; Talebpour et al. 2016).Source: §4.5 · pp. 14–15
  • Global learning methods can estimate future states or decisions, guide initial plans or local-planner sub-goals, and support high-level decisions, but may need substantial training and generalize poorly.The survey covers DRL, CNN architectures, and inverse RL (Brito et al. 2021; Valiente et al. 2022; P´erez-Higueras et al. 2018; Vasquez et al. 2014).Source: §4.5 · p. 15
  • Other global-planning considerations include environment representation, social cues, multimodal sensing and communication, and Wizard-of-Oz studies; strategy choice depends on task demands.Source: §4.5 · p. 15

5 Situation Awareness and Assessment

Reviewed proposals draw on elements of the three main situation-awareness branches, but none considers all three; obstacles, trajectory prediction, and proxemics are the most frequent aspects.

5.1 Environment

  • Environment analysis finds limited attention to semantics, three obstacle-representation types—dense, sparse, and hybrid—and only four papers considering other environmental factors.Those other factors are people’s density and gateways.Source: §5.1 · pp. 15–16

5.2 Agents

  • Agent assessment centers on trajectory prediction and also classifies interactions and actions, intentions, and attributes beyond basic pose.The review says detected intentions in its taxonomy are expected rather than unexpected, and other attributes include personality, emotion, altruism, sympathy, and cooperation.Source: §5.2 · pp. 17–18

5.3 Social norms

  • Proxemics is the dominant social norm, commonly modeled as personal-space constraints, while some systems modulate speed and others apply additional movement or interaction rules.Personal-space models can account for factors such as relative velocity, a person’s status or emotion, and the surrounding area.Source: §5.3 · p. 19

6 Evaluation

The survey describes evaluation as a mix of qualitative and quantitative methods, with quantitative evaluation appearing in more than 100 papers. It also reviews studies, simulators, datasets, benchmarks, and metrics that support socially aware navigation research.

6.1 Methods

  • Qualitative evaluation examines robot and human motion and uses user studies to assess people’s experiences, expectations, and perceptions.About half of the surveyed papers use some qualitative methodology, and around 33 conduct user studies; these studies can be difficult to replicate and organize.Source: §6.1 · pp. 19–20
  • Quantitative evaluation covers navigation performance, human discomfort, and naturalness, with discomfort measures often drawing on proxemics theory.The survey highlights Social Individual and Group Indices, Relative Motion Index, and Social Direction Index; desired Social Direction Index values depend on the situation and number of humans.Source: §6.1 · p. 21
  • More than 100 surveyed articles use quantitative evaluation, whose analytical metrics support large-scale assessment and comparisons between navigation systems.Metrics may be calculated across trajectories or goals, and some track changes over time; some studies combine qualitative and quantitative analysis.Source: §6.1 · pp. 21–22

6.2 Tools

  • Studies inform socially appropriate behaviors, while simulators support rapid testing across human densities, although few robot simulators support human-robot interaction.Studies cover communication, pedestrian-vehicle interactions, and legibility; newer simulators include crowded scenarios, multimodal interaction, or intelligent human agents.Source: §6.2 · p. 22
  • Robot-navigation datasets capture human responses to robots, and benchmarks provide metrics and protocols for comparing socially aware navigation systems.THÖR records controlled indoor data and SCAND includes indoor and outdoor data; Socnavbench, SocialGym, and SEAN provide benchmarking tools, while Wang et al. (2022) propose crowd-navigation protocols.Source: §6.2 · pp. 22–23

7 Proposals

The survey proposes eight priorities for socially aware navigation, spanning richer human and robot modeling, adaptive behavior, shared benchmarks, communication, evaluation, comfort and trust, and dynamic-agent planning.

Proposal 1: Enhance human models

  • Human models should go beyond position, speed, and motion intent to represent reactions and mental states using action-reaction evidence and psychological models.The survey says limited human models can contribute to unexpected robot behavior.Source: §7 · p. 23

Proposal 2: Design user-adaptive robot behaviors

  • Improve human perception to identify people and use characteristics such as age, gender, and height to filter or shift robot behavior adaptively.The survey anticipates that online adaptive behaviors may become feasible as computer-vision AI advances.Source: §7 · p. 23

Proposal 3: Define context-based benchmarks and universal standards

  • Define standard contexts and human actions for benchmarks, since interactions vary with both and the lack of universal standards makes algorithms difficult to compare.Even within one environment, behaviors may need to adapt to whether a person is walking leisurely, rushing, running, or approaching a place.Source: §7 · p. 23

Proposal 4: Focus on robot-specific parameters

  • Account for robot-specific characteristics such as shape and size, which affect interaction preferences but are disregarded by most socially aware navigation algorithms.Develop strategies that adapt to different robot characteristics.Source: §7 · p. 23

Proposal 5: Establish good communication protocols

  • Develop communication protocols that convey robot intent, and use user studies to address understudied timing and communication means.Both implicit and explicit communication strategies have been studied, but research on when and how to communicate remains limited.Source: §7 · p. 23

Proposal 6: Explore alternatives to proxemics-based metrics

  • Develop alternatives to proxemics metrics through HRI studies: discomfort measures and thresholds vary by situation, human-human data can mislead, and current benchmark metrics do not fit every setting.These limitations make it difficult to compare the social awareness of different navigation algorithms.Source: §7 · pp. 23–24

Proposal 7: Prioritize comfort and trust

  • Design for comfort and trust alongside safety, including prediction of robot motion, transparent decision-making, and reliable behavior across different robot types.The survey identifies comfort and trust as receiving comparatively limited attention despite significant research on safety.Source: §7 · p. 24

Proposal 8: Include dynamic models of other agents

  • Include other agents’ dynamic models in planning, since slow-motion assumptions become problematic for fast-moving bicycles, motorcycles, or cars and can increase urban risks.Higher speeds make accounting for these dynamics crucial when robots generate plans and actions.Source: §7 · p. 24

8 Future Challenges

The section identifies long-term challenges spanning human-robot interaction, urban deployment and regulation, AI and decision-making, and evaluation. It calls for navigation systems that anticipate situations and conflicts while meeting human expectations and operating safely in shared spaces.

Human-robot relationship

  • Robots must account for cultural norms and individual preferences while ensuring human acceptability, safety, trust, and privacy in interactions.They also need to estimate the intentions of people they accompany and adapt navigation to abnormal behaviors such as vandalism or obstruction.Source: §8 · p. 24

Urban design

  • Urban deployment must address city-specific street typologies, designated robot zones, and interactions with mobility devices, bicycles, motorcycles, and cars.Cities and buildings are not yet prepared for robots in open areas, and segregation can help prevent conflicts between robots and people.Source: §8 · p. 24

Regulation

  • Because current urban regulations generally prohibit autonomous robots, specific rules for goods-delivery robots are needed to support safe and socially accepted deployment.The paper suggests regulations similar to those for autonomous cars.Source: §8 · p. 24

AI and decision-making

  • Socially aware navigation still lacks sufficient work on human preferences, intentions, and social norms, as well as situation understanding and short-term conflict prediction for proactive planning.Anticipating how a situation may evolve over the next few seconds can enable robots to generate new plans and actions.Source: §8 · pp. 24–25
  • Cooperation, negotiation, and trajectory planning require predictions of task dynamics, agent actions, environmental conditions, and expected human behavior.Such predictions can support behavior adaptation, coordination, and well-informed decisions.Source: §8 · p. 25

Evaluation

  • Evaluation must go beyond path length and collision avoidance to assess user experience, social acceptance, and perceived trustworthiness against human expectations.The paper calls for robust assessment procedures that include these social characteristics.Source: §8 · p. 25

9 Conclusions

The survey analyzes 193 articles across four faceted taxonomies, finding that mobile robots remain the main domain of interest. It identifies future needs for reliable adaptation to human behavior, multimodal communication, and attention to ethical and legal issues as robot use expands.

9 Conclusions

  • The analysis classifies 193 articles across four faceted taxonomies and finds that mobile robots remain the main domain of interest.The survey also includes socially aware navigation for autonomous cars and drones.Source: §9 · p. 25
  • Taxonomy-based classifications contribute to prior work on human-aware social robot navigation, including proxemics, planning, perception, mapping, and evaluation.The survey references most prior surveys and studies in the field.Source: §9 · p. 25
  • Future research needs more reliable techniques to predict and adapt to human behavior in dynamic contexts, communicate robot goals through multiple modalities, and address ethical and legal issues.These issues grow as robots are used in healthcare, education, and other social domains.Source: §9 · p. 25
Loading 2311.06922v4…