PaperlayerView exact revision on arXiv

Source-linked AI summary

Proximity Perception in Human-Centered Robotics: A Survey on Sensing Systems and Applications

Stefan Escaida Navarro, Stephan Mühlbacher-Karrer, Hosam Alagi, Hubert Zangl, Keisuke Koyama, Björn Hein, Christian Duriez, Joshua R. Smith

arXiv:2108.07206v2cs.RO

TL;DR

Proximity sensing in human-centered robotics has lacked significant formalization. This survey categorizes applications and behaviors, reviews sensing principles and technologies, and reports sustained attention to sensorimotor integration for reactive behavior.

  • Problem

    The field has lacked significant formalization, making a basic categorization of proximity-sensing applications and behaviors useful for human-centered robotics.

    Source: B. Application and Behavior Types · p. 3, §VI · p. 18

  • Method

    The survey categorizes literature by application and behavior complexity, reviews measurement principles, and organizes technology characteristics in a summary table.

    Source: §I · p. 2, §III · p. 4, §VI · p. 19

  • Results

    The survey reports sustained attention to integrating proximity sensing into sensorimotor functionality to realize highly reactive robotic behavior.

    Source: §VI · p. 19

  • Takeaways & Limitations

    Medical and service robotics are promising areas for proximity perception, but more research is needed to establish use cases and methods before commercialization.

    Source: A. Human-robot interaction in the… · p. 17

    Capacitive sensing typically reports 20–125 Hz rates, compared with >1 kHz for recent optical approaches, because electric-field effects are often weak and yield low signal-to-noise ratios.

    Source: 4) Further Aspects of Capacitive… · p. 6

Abstract

from arXiv · show

Proximity perception is a technology that has the potential to play an essential role in the future of robotics. It can fulfill the promise of safe, robust, and autonomous systems in industry and everyday life, alongside humans, as well as in remote locations in space and underwater. In this survey paper, we cover the developments of this field from the early days up to the present, with a focus on human-centered robotics. Here, proximity sensors are typically deployed in two scenarios: first, on the exterior of manipulator arms to support safety and interaction functionality, and second, on the inside of grippers or hands to support grasping and exploration. Starting from this observation, we propose a categorization for the approaches found in the literature. To provide a basis for understanding these approaches, we devote effort to present the technologies and different measuring principles that were developed over the years, also providing a summary in form of a table. Then, we show the diversity of applications that have been presented in the literature. Finally, we give an overview of the most important trends that will shape the future of this domain.

I. INTRODUCTION · II. PROXIMITY SENSOR: CHARACTERIZATION, APPLICATIONS AND SAFETY CONSIDERATIONS · A. Characterization

The survey frames proximity perception as a complement to vision and touch for safe, responsive human-centered robotics, organizing applications by robot component and behavior. It reviews sensor principles, applications, safety considerations, and the attributes and ranges used to characterize proximity sensing.

I. INTRODUCTION

  • Proximity perception complements vision and touch by addressing occlusions, blind spots, and robot pose uncertainty while supporting safe, uninterrupted interaction in unstructured environments.Sensor designs must coexist with the existing perceptive modalities.Source: §I · p. 1
  • Human-centered applications divide into manipulator-arm skins for safety and interaction (AT-I) and gripper or hand sensors for grasping and exploration (AT-II).Proximity sensing can detect people before contact where tactile sensing alone would require very low robot velocities, and can support pre-touch closed-loop hand preshaping.Source: §I · pp. 1–2
  • In human-centered robotics, proximity perception concerns short human–robot distances and aims to improve interaction and safety; sensing principles may also apply to automotive and UAV domains.The survey distinguishes this focus from automotive and UAV work targeting larger distances and speeds, autonomous travel, and contact avoidance.Source: §I · p. 2
  • The survey organizes literature by two application types and behavior complexity, reviews sensor principles and applications, compares technologies, and outlines future trends.Its coverage includes capacitive, optical, and radar designs, applications in robotics and industry, and concluding remarks.Source: §I · p. 2

A. Characterization

  • Proximity sensors detect objects without contact and are typically active, short-range, skinlike, low-latency, and integrated into robot sensory-motor functions.They can form sensing networks on robot-arm segments and fingers, enable reflex-like behavior, address vision occlusions, and supervise approaching objects before tactile contact.Source: A. Characterization · pp. 2–3
  • Detection distances below 50 cm are proposed as proximity range, approximately where visual occlusions become problematic in human–robot interaction and collaboration.The boundary is not strict; it is also near the roughly 45 cm intimate-space distance described by Edward T. Hall.Source: A. Characterization · p. 3
  • Below 10 cm, pre-touch sensing can support closed-loop adjustment of finger and hand posture until contact, provided sensing remains uninterrupted.Even sensors with limited detection range can be useful for this control.Source: A. Characterization · p. 3

B. Application and Behavior Types · C. Safety Considerations and Norm Compliance · III. MEASUREMENT PRINCIPLES FOR PROXIMITY SENSING

The sections classify proximity-sensing applications and behaviors, outline a standards-based safety separation distance, and introduce the measurement principles and representative systems reviewed in the paper. The measurement-principles section connects to the later discussion of implemented applications.

B. Application and Behavior Types

  • Proximity-sensing applications are grouped as safety and HRI/HRC (AT-I) or preshaping and grasping (AT-II).The paper presents these as broad categories, with examples in Figs. 1, 2, and 4.Source: B. Application and Behavior Types · p. 3
  • Reactive or reflex-like behaviors (BT-I) are low-complexity, while cognitive or model-based behaviors (BT-II) are high-complexity and may include object exploration.A safety stop is a BT-I example tied to sensor and controller update rates; object exploration uses an object model and planner across multiple phases. BT-I behaviors generally appear as subsystems of BT-II.Source: B. Application and Behavior Types · p. 3

C. Safety Considerations and Norm Compliance

  • Sp(t = t0) = vh(Tr + Ts) + vrTr + Ss + Ci + Zd + Zr specifies the protective separation distance for speed and separation monitoring.The terms represent human speed, robot reaction and stopping times, robot speed and stopping distance, intrusion distance, and human and robot position uncertainties; the passage gives vh = 1.6 m s when human speed is not monitored.Source: C. Safety Considerations and Norm… · p. 4
  • Sp = 0.236 m for the stated UR10e configuration, excluding human-position uncertainty and intrusion distance.The example uses a 500 Hz control loop, vr = 5 m s, Tr = 4 ms, Ts = 100 ms, Ss = 50 mm, and Zr = 0.05 mm; the excluded terms depend on sensor parameters monitoring the area.Source: C. Safety Considerations and Norm… · p. 4

III. MEASUREMENT PRINCIPLES FOR PROXIMITY SENSING

  • The measurement-principles section introduces physical principles for proximity sensing, explains basics to support literature review, and systematizes them in Table I.It also reviews representative implementations of proximity-sensing technologies.Source: §III · p. 4
  • The measurement-principles discussion links to Section IV on implemented applications, with cross-references to relevant relationships that are not exhaustive..Source: §III · p. 4

A. Capacitive Sensing … 3) Capacitive Sensing in AT-II:

The section surveys capacitive proximity sensing principles and measurement approaches, then traces their use in robotics from early obstacle detection to sensors on robot links and grippers. It describes how electrode configuration and object properties shape sensing behavior and application limits.

A. Capacitive Sensing

  • Capacitive sensing estimates distance and object properties from changes in electric fields, but signals depend nonlinearly on object geometry, pose, grounding, and material.Humans are reliably detected, while small or non-conductive objects can be more difficult to sense.Source: A. Capacitive Sensing · pp. 4–5
  • Capacitance is commonly measured through displacement currents or oscillator frequency, usually at frequencies not much above 1 MHz where quasi-static assumptions apply.Self-capacitance measures matrix diagonal elements and mutual capacitance measures off-diagonal elements; linearly independent excitation patterns are needed to determine the full matrix.Source: A. Capacitive Sensing · pp. 4–5
  • Single-ended sensing typically yields higher capacitance and signal-to-noise ratio, while mutual-capacitance sensing provides more independent electrode-pair measurements for applications such as electrical capacitance tomography.Grounded conductive objects generally increase measured capacitance in single-ended mode but can shield the field and reduce mutual capacitance; low-ground coupling can instead increase mutual capacitance through polarization.Source: A. Capacitive Sensing · pp. 4–5

1) Early Capacitive Technologies in Robotics:

  • Early robotic work demonstrated mutual-capacitance obstacle detection on manipulator links in 1988, with conductive obstacles detected more reliably than insulating ones.Later work explored electrode designs, evaluated collision avoidance before favoring infrared sensing, and developed large arm-mounted arrays [14].Source: 1) Early Capacitive Technologies in… · p. 5

2) Capacitive Sensing for AT-I:

  • For AT-I, mutual-capacitance sensors and modular single-ended skins were developed to cover robot links for obstacle detection, collision avoidance, and human-robot interaction.Single-ended sensors may not detect insulating materials adequately, so combinations with time-of-flight sensing were proposed; companies also adopted capacitive sensing for safety features.Source: 2) Capacitive Sensing for AT-I · p. 5

3) Capacitive Sensing in AT-II:

  • For AT-II, multi-transmitter and receiver gripper electrodes support short-, mid-, and long-range sensing for reactive preshaping and can detect changes in an object's grounding state when a person holds it.A tactile proximity sensor was also integrated into fingertips.Source: 3) Capacitive Sensing in AT-II · p. 5
  • AT-II integrations include gripper-based reactive preshaping and telemanipulation, humanoid-hand fill-state sensing, and capacitive end-effectors for detecting human limbs in dressing and washing tasks.Small electrodes aid integration and spatial resolution but reduce sensing range and sensitivity; mutual-capacitance sensing with flexible spatial resolution addresses some self-capacitance material limitations. Capacitive sensors' mechanical robustness also suits harsh industrial settings.Source: 3) Capacitive Sensing in AT-II · p. 6

4) Further Aspects of Capacitive Proximity Sensing Technologies in Robotics: · 5) Spatial Resolution and Sensing Range for Capacitive Sensors: · B. Optical Sensing

The sections describe capacitive sensing’s design tradeoffs and adaptable resolution and range, then outline optical sensing principles and their reflectance-related tradeoffs.

4) Further Aspects of Capacitive Proximity Sensing Technologies in Robotics:

  • Capacitive designs combine tactile and proximity sensing,,,,, recognize materials with multi-frequency excitation,,, and explore tomography, and flexible resolution,,.Source: 4) Further Aspects of Capacitive… · p. 6
  • Capacitive sensing typically updates at 20–125 Hz, with exceptions reaching several kilohertz, versus >1 kHz in recent optical approaches.Weak electric-field effects, low signal-to-noise ratios, low-frequency carrier circuits, and time-multiplexed sensing elements constrain update rates; stronger excitation raises cost and power consumption.Source: 4) Further Aspects of Capacitive… · p. 6
  • Bio-inspired underwater robots mimic weakly electric fish, whose electric organs and skin receptors detect field disturbances through mutual impedance in low-visibility, cluttered environments,.The fish’s electrosense is used for navigation and prey detection; the figure illustrates the freshwater elephantfish and its electric field [47].Source: 4) Further Aspects of Capacitive… · p. 6, Figure 7 · p. 6

5) Spatial Resolution and Sensing Range for Capacitive Sensors:

  • Capacitive resolution is adaptable, but smaller electrodes challenge sensitivity: 1×1 mm² tactile cells contrast with 15×25 mm² electrodes detecting conductive objects at ≈40 mm, while 50×50–100×100 mm² electrodes reach ≈30 cm.Sensing range also depends on electrode spacing in mutual-capacitive mode and circuit design; increasing range makes self-influence an issue.Source: 5) Spatial Resolution and Sensing… · p. 6

B. Optical Sensing

  • Optical proximity sensing uses O-RLI, O-ToF, and O-Tri with an emitter and receiver on one surface, while O-BB detects a broken beam across distinct surfaces.O-RLI measures reflected intensity, O-ToF measures return time, and O-Tri uses the incident light position or angle; modulated light can suppress extraneous influences in O-RLI.Source: B. Optical Sensing · pp. 6–7, Figure 8 · p. 7
  • O-RLI proximity depends on object reflectance, whereas O-ToF and O-Tri measure distance without direct reflectance dependence but can struggle with specular surfaces; O-BB can detect shiny objects.O-RLI is widely used for manipulators and grippers because its sensor structure and processing are simple and integrate readily.Source: B. Optical Sensing · p. 7

1) Reflected Light Intensity-Type Sensors (O-RLI) at an Early Stage: · 3) Reflected Light Intensity-Type Sensors (O-RLI) Using Sensing Modules): · 4) Time-of-Flight-Type (O-ToF) Sensors:

The passages trace optical proximity sensing from early O-RLI skins and gripper arrays to compact modules and O-ToF systems, with applications including collision avoidance, grasping, force sensing, and soft robotics.

1) Reflected Light Intensity-Type Sensors (O-RLI) at an Early Stage:

  • Lewis et al. proposed an O-RLI gripper in 1973, but large components and limited CPU performance then made arrays on manipulator links or grippers technically difficult.The program aimed to integrate sensory and motor functions for autonomous manipulation and locomotion. By the late 1980s and early 1990s, O-RLI proximity skins for collision avoidance had been demonstrated,.Source: 1) Reflected Light Intensity-Type… · p. 7
  • O-RLI skins used flexible circuit boards to cover manipulator arms, with modulated light mitigating cross-talk; related work placed distributed sensors on a two-jaw gripper,,.These designs required custom analog front-end electronics for the skin as a whole.Source: 1) Reflected Light Intensity-Type… · p. 7, Figure 9 · p. 7
  • Optical fibers enabled O-RLI sensing in confined spaces, including a disc-like end-effector with 32 fibers routing reflected light around 360° to a display –.The display’s intensity values were recorded by a camera; other implementations routed fibers from a fingertip to a KEYENCE signal-processing module –.Source: 1) Reflected Light Intensity-Type… · p. 8
  • A fingertip O-RLI sensor with twelve photoreflectors localized proximity events at < 1 ms sampling, while simple calibration methods remained limited by millimeter or sub-millimeter errors –.The same work included a linear sensor array and calibration using fingertip positioning and reflected-light changes; position errors or circuit noise limited accuracy for reactive preshaping. The figure reports < 1 ms measurement time and less than 51 µm distance resolution for the high-precision sensor [70], [71].Source: 1) Reflected Light Intensity-Type… · p. 8, Figure 11 · p. 8

3) Reflected Light Intensity-Type Sensors (O-RLI) Using Sensing Modules):

  • Compact, low-noise O-RLI modules with I2C connectivity enabled multi-sensor fingertips and robot skins; the VCNL4010 measures 1–200 mm within 4 ms.The VCNL4010 footprint is 3.95× 3.95 mm2, though readings are affected by surface reflectance. Patel et al. used it to detect distance, contact, and force; Hughes et al. used the same sensor structure for flexible skin and gesture recognition. Other examples include an 8×8 sensor matrix [57], Barrett Hand finger sensors, and a 3 × 4 mm2 modular sensor.Source: 1) Reflected Light Intensity-Type… · p. 7, 3) Reflected Light Intensity-Type… · p. 8

4) Time-of-Flight-Type (O-ToF) Sensors:

  • The VL6180X measures 10–100 mm with 1 mm resolution, and fingertip-sized O-ToF sensors enabled robust Rubik’s Cube manipulation,.One sensor takes 7 ms to several tens of ms per measurement, with repeat measurement error of ±1–2 mm. Lancaster et al. also used reflected light intensity through transparent rubber to measure force; a rounded rubber configuration improved force sensitivity.Source: 4) Time-of-Flight-Type (O-ToF) Sensors · pp. 8–9
  • O-ToF and O-RLI combinations supported simultaneous fingertip and hand-base adjustment, while O-ToF skins and capacitive–O-ToF cuffs targeted collaborative-robot collision avoidance,,,.In the multimodal sensor, O-RLI sensed object distance and posture on a table, while O-ToF measured distance from the table surface; the cuffs also supported material recognition.Source: 4) Time-of-Flight-Type (O-ToF) Sensors · p. 9
  • O-ToF modules were integrated into soft grippers using a thin PDMS sheet with EGaIn copper traces and additional IMU, pressure, and temperature sensors,.The VL6180X module itself was not deformable.Source: 4) Time-of-Flight-Type (O-ToF) Sensors · p. 9

5) Triangulation-Type Sensors (O-Tri): … D. Other Sensing Principles

The surveyed approaches span optical triangulation and break-beam sensing, optical sensor trade-offs, radar applications, and less-mainstream acoustic, inductive, and whisker principles. Together, the passages describe sensing capabilities and limitations relevant to human-centered robotics.

5) Triangulation-Type Sensors (O-Tri):

  • Koyama et al., developed a fingertip-sized triangulation sensor that measures distance with error below 31 µm in under 1 ms and detects object posture.Its dimensions are 18×28.5×38.5 mm3.Source: 5) Triangulation-Type Sensors (O-Tri) · p. 9

6) Break-Beam-Type Sensors (O-BB):

  • Break-beam sensing in grippers supports reactive motion and preshaping, including grasping challenging objects such as semi-transparent tissues.Teichmann et al. used light blocking to trigger reactive motions in a parallel-jaw gripper and also described a three-fingered hand.Source: 6) Break-Beam-Type Sensors (O-BB) · p. 9

7) Discussion on Optical Sensors:

  • Optical devices are small and fast, but can struggle with transparent, black, and shiny objects; combining optical and capacitance sensing can address these detection difficulties.Source: 7) Discussion on Optical Sensors · p. 9
  • Optical sensors span compact elements and varied ranges: O-RLI reports about 300 mm, O-ToF up to 4 m, and O-Tri 1.5 m, each with a near-sensor dead spot.The reported dead spots are small for O-RLI, 10 cm for O-ToF, and 20 cm for O-Tri; high-density elements on large-area skin can impose substantial wiring and signaling effort.Source: 7) Discussion on Optical Sensors · p. 9

C. Radar

  • Smaller system-on-chip radar devices have increased radar’s appeal for robotic platforms, while automotive safety expertise may support human-robot applications.Radar can withstand harsh weather and complement optical technologies; a survey [90] covers automotive and HCI uses including tracking and gesture recognition.Source: C. Radar · p. 9
  • FMCW radar measures distance and velocity simultaneously with high close-range resolution, supporting proximity perception, collision avoidance, and robotic HRI such as gesture control.A 60 GHz chip with integrated transmit and receive antennas was introduced for fine gesture interaction; the illustrated principle derives range from signal delay and velocity from Doppler-related phase changes across chirps.Source: C. Radar · p. 9, Figure 13 · p. 10
  • Radar robotics work includes FMCW simulation for signal-processing and machine-learning development before deployment, multimodal grasping and classification, SAR imaging and navigation, and subsurface mapping,,.A 160 GHz multi-antenna radar with flexible waveguides was also studied for collaborative robots.Source: C. Radar · p. 10

D. Other Sensing Principles

  • Acoustic, inductive, and whisker sensing remain less mainstream than capacitive, optical, or radar approaches as of 2021, but can offer alternatives that outperform them in some scenarios.Source: D. Other Sensing Principles · p. 10

1) Acoustic: … IV. APPLICATIONS AND METHODS IN THE RESEARCH AND INDUSTRY DOMAINS

The surveyed approaches span acoustic, inductive, and contact-based whisker sensing, as well as modular designs that combine proximity with other modalities. Their applications range from close-range exploration to longer-range sensing and robot navigation or obstacle avoidance.

1) Acoustic:

  • Ultrasound is widely available and used mainly for mid- and long-range sensing, while sensor dead spots and difficult integration limit its practicality for pre-touch applications.A tilted, offset sensor with a parabolic mirror addresses some limitations, but the reported design still uses one acoustic sensing element.Source: 1) Acoustic · p. 10
  • The seashell-effect approach estimates distance from cavity-resonance changes and works at very close range, up to approximately 4 mm,.The authors explore it because it avoids optical sensing difficulties with transparency or reflections and capacitive sensing difficulties with low dielectric contrast.Source: 1) Acoustic · pp. 10–11

2) Inductive:

  • Inductive-capacitive systems report ranges up to 150 mm for conductive materials and 300 mm; combined signals distinguish humans from grounded laptops, and multiple coils estimate full 6 DoF.Inductive sensing detects objects through their disturbance of an alternating magnetic field; highly conductive materials have a strong effect.Source: 2) Inductive · p. 11

3) Whiskers:

  • Artificial whiskers bend under external forces, with deformation typically measured at the base; although contact-based, they probe nearby environments like proximity sensors,.Source: 3) Whiskers · p. 11

E. Multi-modal and Modular Sensors

  • Coexisting proximity, tactile, inertial, and temperature modalities can fill perception gaps; HEX-o-Skin exemplifies modular units suited to covering large robot areas.The survey identifies multimodal sensing alongside other robot modalities as important to proximity perception.Source: E. Multi-modal and Modular Sensors · p. 11
  • Multimodal designs use specialized principles for each modality, while other designs combine redundant proximity measurements, such as O-ToF with C-SE,, or O-ToF with O-RLI.Examples also pair proximity and tactile sensing using optical, capacitive, or inductive principles,,,,.Source: E. Multi-modal and Modular Sensors · p. 11

IV. APPLICATIONS AND METHODS IN THE RESEARCH AND INDUSTRY DOMAINS

  • The applications-and-methods section is organized by AT-I and AT-II and by behavior complexity, from BT-I to BT-II.Fig. 14 illustrates a redundant robot using link-mounted proximity sensors to move toward a target end-effector configuration while moving links away from an obstacle [115].Source: §IV · p. 11, Figure 14 · p. 12

A. Reactive collision-avoidance and Contour Following (AT-I, BT-I) · 1) Early Jacobian-type Approaches: · 2) Early Geometric Approaches:

The section surveys reactive collision avoidance through Jacobian-based motion control and geometric contour following. Early work used proximity sensing to guide robot motion around obstacles while preserving progress toward a goal.

A. Reactive collision-avoidance and Contour Following (AT-I, BT-I)

  • Early collision-avoidance work used proximity sensing as a low-latency alternative to cameras and sought spatially resolved obstacle distance, location, and surface-normal data,.The section frames collision avoidance as fundamental to autonomous robots and safe human-robot interaction.Source: A. Reactive collision-avoidance and… · p. 12

1) Early Jacobian-type Approaches:

  • Maciejewski and Klein introduced an obstacle-point Jacobian relating joint velocities to the velocity of a point on the robot.The approach adapts end-effector Jacobian reasoning to obstacle avoidance.Source: 1) Early Jacobian-type Approaches · p. 12
  • The obstacle Jacobian maps desired motion away from an obstacle into joint motion, which can be projected into a higher-priority task’s null space or assigned priority itself.This supports continued end-effector motion while avoiding obstacles, or collision avoidance can take precedence over the desired trajectory.Source: 1) Early Jacobian-type Approaches · p. 12
  • Wegerif et al., tested collision avoidance on a sensor-equipped PUMA 600 with static and dynamic obstacles, while Tamasy presented a smart sensor network and PIPR control architecture.Wegerif’s system used about 120 infrared sender-receiver pairs across three links and prioritized collision avoidance; Tamasy’s architecture included quadratic programming for joint velocities.Source: 1) Early Jacobian-type Approaches · p. 12
  • The WHAP capacitive skin enabled a 2-link robot to circumvent concrete and metallic obstacles; a sensor Jacobian later slowed affected PUMA 560 degrees of freedom during teleoperation,.The teleoperation system did not automatically move away from obstacles because the authors considered that behavior undesirable in that scenario.Source: 1) Early Jacobian-type Approaches · p. 13

2) Early Geometric Approaches:

  • Lumelsky and Cheung,,, proposed following an obstacle’s tangent plane in configuration space using infrared sensors integrated onto a manipulator.Their work also demonstrated the methods and technology in teleoperation with one or more dynamic obstacles.Source: 2) Early Geometric Approaches · p. 13
  • Nunes et al. described end-effector-level contour following: the robot moves parallel to the obstacle’s tangent plane toward the goal until the obstacle no longer blocks the direct path.The illustrated path proceeds from obstacle detection through contour following to the point where direct motion to the goal is clear.Source: 2) Early Geometric Approaches · p. 13

3) Recent Jacobian-type Approaches: … 1) Reactive Preshaping Control:

The survey describes proximity-based collision avoidance and reactive preshaping, where sensor feedback guides robot motion and hand adjustment before contact. It covers geometric contour following and applications to grasping, handovers, and caregiving.

3) Recent Jacobian-type Approaches:

  • Prior work describes Danger Field-based proximity-sensor placement, priority-based task deactivation for evasive motion, capacitive Virtual Whiskers for reactive collision avoidance, and an optimized Jacobian-type collision-avoidance scheme,.The Virtual Whiskers sensor uses 7 electrodes near a KUKA LWR 4 wrist and samples at up to 1 kHz.Source: 3) Recent Jacobian-type Approaches · pp. 13–14

4) Recent Geometric Approaches:

  • Geometric approaches estimate obstacle surfaces to derive contour-following motion, with predictive curvature estimates extending contour following to 3D,,.Curvature can be estimated from current sensor values when spatial resolution permits; the 3D approach detects 2D obstacle-surface curvature.Source: Figure 15 · p. 13, 4) Recent Geometric Approaches · p. 14

B. Reactive Preshaping and Grasping (AT-II, BT-I)

  • Reactive preshaping closes proximity feedback around preliminary finger and end-effector adjustment before grasping, enabling reflex-like adaptation to moving objects such as during handovers.It can align object and fingertip normals for larger, stable contact, synchronize finger contact to limit object motion, permit fast noncontact movement, and adjust poses despite visual occlusion.Source: B. Reactive Preshaping and Grasping… · p. 14

1) Reactive Preshaping Control:

  • Mayton et al. independently controlled Barrett Hand fingers using capacitive sensor targets, demonstrating grasping, robust human handover, and co-manipulation-state detection.The work followed an earlier closed-loop two-jaw gripper demonstration by Espiau and Catros.Source: 1) Reactive Preshaping Control · p. 14
  • Hsiao et al. reported 4 mm average distance error, 5.3° pitch-rotation posture error, and 17.7° roll-rotation posture error in optical-sensor reactive grasping.Their controller initially used raw sensor output as its target, then switched to estimated distance and posture after those estimates became available.Source: 1) Reactive Preshaping Control · p. 14
  • Capacitive feedback enabled simultaneous six-DoF control of a two-jaw gripper, while related 2×2 sensor arrangements supported gripper alignment to a climbing pole and reactive preshaping.The six-DoF method compared sensor intensities across four areas per finger to detect object posture and position simultaneously.Source: 1) Reactive Preshaping Control · p. 15
  • Koyama et al.,, updated an 8-DoF hand every 5 ms; TTC control avoids prior reflectance calibration, while compliant spacers enabled low-force catching of soft objects,.The catching demonstrations included a marshmallow and paper balloon with negligible deformation; 3 mm elastic pads indicated contact from measured distance.Source: 1) Reactive Preshaping Control · p. 15
  • Reactive preshaping and contour following were applied to patient dressing and washing, aligning end-effector distance and orientation to limb contours while mitigating visual occlusion,.A 2×3 end-effector sensor array supported the alignment and contour following.Source: 1) Reactive Preshaping Control · p. 15

C. Higher Complexity Methods and Behaviors (Towards BT-II) … 3) Teleoperation and VR (AT-I, AT-II):

The section surveys higher-complexity cognitive and model-based behaviors, including pre-touch exploration, bio-inspired sensing, and teleoperation. These approaches use proximity information to build object models, guide sensing behavior, and support human-robot interaction.

1) Pre-touch Exploration (AT-II):

  • Pre-touch exploration completes incomplete object models for grasp planning, with deep learning improving object-model and pose estimates,,,.Jiang et al. explore unknown regions in RGB-D point clouds; Maldonado et al. also classify surface textures, while Lancaster et al. use deep learning to guide exploration.Source: 1) Pre-touch Exploration (AT-II) · p. 15
  • Proximity-guided scans aggregate object geometry into point clouds, including a robot-hand scan of a model space shuttle that captured its major features [72],.A two-jaw gripper scans objects by moving around them, while robot kinematics supports point-cloud aggregation.Source: 1) Pre-touch Exploration (AT-II) · p. 16
  • Pre-touch methods also refine edges and corners, classify objects by haptic properties, locate graspable objects, and estimate bottle fill state,,,,.Corner detection is complemented by iterative tactile samples; bottle fill state is explored by tilting and decided in a Bayesian framework. Proximity sensing also improves robustness in contact-based crack detection.Source: 1) Pre-touch Exploration (AT-II) · p. 16

2) Bio-inspired Sensing and Behavior (AT-I, AT-II):

  • Biomimetic electrosense supports reflex-like probe control and model-based recognition of object material and geometry, while active alignment and machine learning identify spheroids,,.Bazeille et al. estimate material properties and ellipse parameters, location, and orientation by fitting a forward model to measurement sequences.Source: 2) Bio-inspired Sensing and Behavior… · p. 16

3) Teleoperation and VR (AT-I, AT-II):

  • Teleoperation studies progressed from proximity-based control without force or tactile display,, to virtual fixtures and exploration of moving objects using finger-tip point clouds.Proximity-generated haptic cues provide touch-like perceptual benefits without requiring the robot to contact unknown structures.Source: 3) Teleoperation and VR (AT-I, AT-II) · p. 16
  • A shared-autonomy controller predicts collisions from proximity signals, limits robot velocity, and provides PHANTOM Omni force feedback; experiments found reduced task completion time and user workload.The reported experiments were conducted over three days using a virtual environment.Source: 3) Teleoperation and VR (AT-I, AT-II) · p. 16
  • Proximity-based 6D force feedback and spatially resolved tactile feedback help teleoperators explore and detect shape cues despite visual distortions or occlusions,.These systems extend end-effector visual perception with capacitive proximity sensors.Source: 3) Teleoperation and VR (AT-I, AT-II) · p. 16
  • Capacitive gesture control enables remote operation of a mobile manipulator, while augmented and virtual reality can present proximity through visual, force, tactile, audio, or electrical cues,,.The capacitive interface acts as a virtual 3D mouse and is robust to water, occlusions, and objects covering parts of the interface; it does not provide force feedback.Source: 3) Teleoperation and VR (AT-I, AT-II) · p. 16

4) Material Classification (AT-II): … A. Human-robot interaction in the industry and service domains

The survey reviews proximity sensing for material classification, tracking, and industrial human-robot interaction, and identifies certification, research, and commercialization challenges for future applications.

4) Material Classification (AT-II):

  • Three-frequency capacitive sensing classified seven materials, while capacitive spectroscopy used 290 frequencies from 10 kHz to 300 kHz and analyzed signal amplitude and phase.These methods use frequency-dependent relative permittivity to identify materials.Source: 4) Material Classification (AT-II) · p. 16
  • Other approaches used capacitive-array configurations, GPR soil-layer mapping, and fingertip optical-acoustic sensing, which classified three materials with over 87% accuracy,,.Estimating material properties also features in the survey’s previously discussed bio-inspired approaches.Source: 4) Material Classification (AT-II) · p. 17

5) Tracking (AT-I, AT-II):

  • Proximity streams supported tracking a cylinder’s 2D position and reflectance with an extended Kalman filter, and tracking two hands through occlusions using capacitive-array data,,.The cylinder-tracking work discusses potential for AT-II; the two-hand tracking approach targets HRI tasks (AT-I).Source: 5) Tracking (AT-I, AT-II) · p. 17

D. Industrial Technologies and Solutions

  • Commercial industrial systems use capacitive proximity skins, including BOSCH APAS,, FOGALE’s tactile-proximity skin (up to 300 mm),, and MRK-Systeme’s KUKA skin (up to 350 mm).FOGALE’s skin was used with a control framework for obstacle avoidance in HRI.Source: D. Industrial Technologies and Solutions · p. 17

V. FUTURE PERSPECTIVES: GRAND CHALLENGES

  • The future-perspectives section frames its outlook around grand challenges identified at the ends of its three subsections.Source: §V · p. 17

A. Human-robot interaction in the industry and service domains

  • Safety certification remains a barrier to widespread industrial proximity sensing, while fenceless automation may offer near-term value and modular radar may ease certification.Current technologies require solution-level certification; the survey calls for an effective procedure and notes that modular radar could use safety-rated components.Source: A. Human-robot interaction in the… · p. 17
  • Proximity sensing could enable service and medical interactions when people are very close to robots, but costly certification and further research on use cases hinder commercial viability.The survey describes this area as promising while noting that commercially viable applications have yet to be established.Source: A. Human-robot interaction in the… · pp. 17–18
  • Further work should test proximity-driven grasping robustness for moving-object handovers and teleoperation, and clarify how highly redundant robots benefit from sensing.The survey notes that research on proximity sensing in highly redundant systems is scarce.Source: A. Human-robot interaction in the… · pp. 17–18

B. Cognitive Robotics · C. Soft Robotics · VI. SUMMARY AND CONCLUSIONS

The survey categorizes proximity-sensing technologies and robotic applications, reviews their development and measurement principles, and identifies challenges and future directions in cognitive and soft robotics. It projects broader adoption as proximity sensing matures alongside tactile and visual perception, especially for human interaction.

B. Cognitive Robotics

  • Active perception uses purposeful actions to enable perception; the authors identify combined vision and touch as a trend and propose adding proximity sensing for multimodal exploration of occluded regions,,,,.They also identify multimodal human modeling for safe interaction and collaboration as an important domain.Source: B. Cognitive Robotics · p. 18
  • Cognitive approaches need proximity sensors that combine readily with vision and touch, while realistic simulation models for measurement principles remain a temporal-performance challenge for sim-to-real learning.Source: B. Cognitive Robotics · p. 18

C. Soft Robotics

  • Soft robots could use proximity sensing for task-hierarchy-aware collision avoidance, reactive preshaping, grasping, exploration, and reaching desired contact states; simulation frameworks such as SOFA are also relevant.The authors emphasize purposeful contact as a particular interest in soft robotics.Source: C. Soft Robotics · p. 18
  • Soft robots need control methods that use proximity information to adapt actuation, while sensor integration in deformable structures must address deformation-related cross-talk and support signal interpretation or sim-to-real learning,.Source: C. Soft Robotics · p. 18
  • Except for capacitive sensing, deformable or stretchable proximity-sensor elements are challenging to realize, though shrinking rigid components such as ICs within stretchable substrates may enable deformable circuits,.Source: C. Soft Robotics · p. 18

VI. SUMMARY AND CONCLUSIONS

  • The survey proposes a categorization of proximity-sensing applications and technologies, characterizes proximity sensors, and organizes reported technologies and measurement principles in Table I.The table includes sensing range, update rate, and sensing-element size; the review covers developments from the early 1970s through 2021.Source: §VI · p. 19
  • The survey distinguishes collision avoidance and human-robot interaction (AT-I) from preshaping and grasping (AT-II), and reports both reactive sensory-motor and increasingly cognitive approaches.Cognitive approaches include teleoperation, autonomous object exploration, and bio-inspired approaches mimicking weakly electric fish.Source: §VI · p. 19
  • The authors project that maturing integration, costs, and norm conformity will enable proximity sensors to coexist with tactile and visual perception and support diverse solutions, especially human interaction.Source: §VI · p. 19

Appendix

The appendix provides a year-sorted overview of proximity-sensing works and the attributes used to compare them.

Appendix

  • Table I organizes works by year and records their sensing principles, reported ranges and fields of view, rates, dimensions, obstacle handling, components, and categorization.The listed columns also include references, commercial availability, and the basis for categorization.Source: Appendix · p. 25, Table I · p. 27
Loading 2108.07206v2…