PaperlayerView exact revision on arXiv

Source-linked AI summary

Object Handovers: a Review for Robotics

Valerio Ortenzi, Akansel Cosgun, Tommaso Pardi, Wesley Chan, Elizabeth Croft, Dana Kulic

arXiv:2007.12952v2cs.ROeess.SY

TL;DR

Human-robot object handovers require coordinated cognitive and physical behaviour, yet robotic exchanges remain less fluent and less comparable than human interactions. This paper reviews handovers by phase and actor, synthesizes methods, behaviours, metrics, safety, and ergonomics, and identifies communication, cognitive-physical integration, fluency, and standardized evaluation as central improvement areas.

  • Problem

    Robots must perform efficient, fluent object handovers in unstructured environments, but current human-robot exchanges do not yet match human coordination and lack consistent evaluation protocols.

  • Method

    The paper reviews human-human and human-robot handovers across pre-handover and physical-exchange phases, examining both actors, cognitive and physical processes, safety, and evaluation metrics.

  • Results

    The review finds that robot-to-human handovers are studied more often than human-to-robot handovers, physical exchange is less studied than preparation, and experimental protocols lack uniformity.

  • Takeaways & Limitations

    More fluent communication and integration of cognitive and physical reasoning, together with standardized protocols and metrics, are identified as priorities for improving robotic handovers.

Abstract

from arXiv · show

This article surveys the literature on human-robot object handovers. A handover is a collaborative joint action where an agent, the giver, gives an object to another agent, the receiver. The physical exchange starts when the receiver first contacts the object held by the giver and ends when the giver fully releases the object to the receiver. However, important cognitive and physical processes begin before the physical exchange, including initiating implicit agreement with respect to the location and timing of the exchange. From this perspective, we structure our review into the two main phases delimited by the aforementioned events: 1) a pre-handover phase, and 2) the physical exchange. We focus our analysis on the two actors (giver and receiver) and report the state of the art of robotic givers (robot-to-human handovers) and the robotic receivers (human-to-robot handovers). We report a comprehensive list of qualitative and quantitative metrics commonly used to assess the interaction. While focusing our review on the cognitive level (e.g., prediction, perception, motion planning, learning) and the physical level (e.g., motion, grasping, grip release) of the handover, we briefly discuss also the concepts of safety, social context, and ergonomics. We compare the behaviours displayed during human-to-human handovers to the state of the art of robotic assistants, and identify the major areas of improvement for robotic assistants to reach performance comparable to human interactions. Finally, we propose a minimal set of metrics that should be used in order to enable a fair comparison among the approaches.

I. INTRODUCTION

Object handover is a joint action in which a giver transfers an object to a receiver through coordinated cognitive and physical processes. The review divides handovers into pre-handover preparation and physical exchange, surveys human and robotic behaviour, and proposes common evaluation metrics.

  • I. INTRODUCTION: The review addresses handovers in unstructured environments, where robots must perceive dynamic surroundings and adapt their planning and actions.Traditional work cells are structured, whereas households, hospitals, and other unstructured settings demand more adaptive capabilities.
  • I. INTRODUCTION: Object handover is a collaborative joint action requiring the giver and receiver to coordinate the transfer of an object.The interaction draws on prediction, perception, action, learning, and adjustment by both participants.
  • I. INTRODUCTION: The review analyses giver and receiver reasoning, communication, grasping, motion planning, control, and grip modulation across both handover phases.It also considers safety and proposes metrics for evaluating handovers.
  • I. INTRODUCTION: Human joint actions rely on shared representations, action prediction, and integration of predicted effects to coordinate behaviour in space and time.Coordination may be planned from shared goals or emerge from perception-action couplings.
  • I. INTRODUCTION: A handover has a pre-handover phase followed by physical exchange, beginning with the receiver’s first contact and ending when the giver releases the object.The pre-handover phase includes communication, giver grasping, and transport; the physical phase ends when the receiver fully controls the object.
  • I. INTRODUCTION: Direct handovers transfer an object from the giver’s hand to the receiver’s hand, whereas indirect handovers place the object on an intermediate surface.Indirect exchange can provide greater flexibility in how the object is passed.

III. PRE-HANDOVER PHASE

The pre-handover phase prepares the physical exchange through communication, prediction, perception, grasping, and motion planning. Both agents contribute, although the giver performs more overt preparation while the receiver’s readiness and actions influence the giver.

  • III. PRE-HANDOVER PHASE: The pre-handover phase begins after initiation and prepares the object, agents, and exchange location for physical contact.It includes communication, grasping, transport, and planning before the receiver touches the object.
  • III. PRE-HANDOVER PHASE: The giver predicts the receiver’s intended task and grasp, then plans to grasp or re-grasp and offer the object accordingly.Visual and tactile feedback support tracking the object and the interaction state during preparation.
  • III. PRE-HANDOVER PHASE: Handover preparation therefore couples giver planning with receiver prediction, communication, and anticipatory movement.The two agents share the preparation process even though their observable activity differs.
  • III. PRE-HANDOVER PHASE: The receiver’s attention, preparedness, actions, and communication signal readiness and influence the giver’s behaviour before contact.The receiver may also predict the giver’s behaviour and move toward the anticipated handover location.

A. Communication

Communication coordinates what, when, and where a handover will occur through speech, gaze, posture, gestures, motion, and object presentation. Robotic studies indicate that visual intent cues and gaze can improve human experience and coordination.

  • A. Communication: Communication initiates handovers and coordinates them by helping partners predict one another’s actions and reducing uncertainty.Humans communicate the object and action as well as the intended timing and location.
  • A. Communication: Speech can initiate and coordinate a handover, but divided attention across auditory and visual modalities can degrade coordination.Robot-human dialogue may establish roles and coordinate subsequent actions.
  • A. Communication: Robot gaze can produce faster object reaching, more natural interaction perceptions, and faster human response times.A deliberate release delay can also increase attention to the robot’s head and compliance with its suggestions.
  • A. Communication: Humans also communicate handover intent through body posture, arm pose, gestures, motion trajectories, and how the object is presented.Robotic approaches have investigated these cues for intent detection and interaction.
  • A. Communication: Projection methods that show the object pose and intended grasp pose substantially improve the subjective experience of human-to-robot handovers.These projections communicate the robot’s intended action to the human partner.

B. Grasp Planning

Handover grasp planning should account not only for object stability and speed but also for the receiver’s subsequent task, functional constraints, safety, and ergonomic demands. Task-aware grasps can reduce receiver adjustments and improve task performance and interaction perception.

  • B. Grasp Planning: Second-order grasp planning chooses the giver’s grasp and object presentation with the receiver’s subsequent task in mind.The aim is to minimise manipulation the receiver must perform before using the object.
  • B. Grasp Planning: Human givers adapt grasp type, location, and object orientation to the receiver’s task and the object’s functional parts.Grasping is described as a purposive action shaped by task demands and cooperative intent.
  • B. Grasp Planning: Object shape, function, safety, gripper constraints, environmental factors, and receiver requirements all affect handover grasp planning.These factors influence how the object should be oriented and where the giver should grasp it.
  • B. Grasp Planning: Accounting for the receiver’s subsequent task reduced task-completion time by eliminating post-handover object readjustments.Human perceptions also improved when functional-part constraints were more restrictive.
  • B. Grasp Planning: Robotic grasp success is often assessed by stability or speed, while task requirements for force and mobility are frequently overlooked.Joint planning of robotic giver and receiver grasps is difficult to extend to human receivers because their behaviour is harder to model reliably.

C. Perception

Reliable perception in handovers must track the object, both hands, and the partner’s body, while supporting grasp planning, motion prediction, and reachable exchange locations.

  • Perception challenges: Reliable perception must track the object, both hands, and the partner’s full-body motion during handovers.Vision is commonly the main perception channel, but occlusion during grasping makes tracking difficult.
  • Grasp perception: Grasp planning can use object-and-hand tracking or classify human grasps into categories such as “waiting” and “lifting”.Classification simplifies planning but detects only a relatively small subset of the richer grasp behaviours displayed by humans.
  • Motion prediction: Human motion prediction methods estimate handover location and timing using Dynamic Movement Primitives, Extended Kalman Filters, minimum-jerk trajectories, or probabilistic motion models.
  • Handover location: Handover locations must be reachable by both agents and are influenced by interpersonal distance and the exchange height.Human-human handovers occur roughly midway between giver and receiver.
  • Handover location: A task-specific interaction workspace can be formed from the intersection of robot and human reachable spaces, incorporating human effort and biomechanical properties.

E. Motion Planning and Control

Successful handover motion combines concurrent, readable coordination with adaptive planning and feedback control rather than relying solely on fixed trajectories.

  • Motion coordination: Human-human handovers use smooth concurrent movements, with giver and receiver reaching toward each other rather than acting in separate successive phases.
  • Motion readability: Robot motion should be legible so partners can infer its goal and predictable so it matches expectations given that goal.Human-like or natural robot configurations are reported as more readable to people.
  • Adaptive planning: Fully pre-planned motion lacks general adaptability when the environment or partner behaviour changes.Robustness, reactivity, and context awareness are proposed as guiding principles for interaction design.
  • Feedback control: Control architectures use sensory feedback to change robot behaviour, with impedance and admittance control among common physical-interaction strategies.
  • Feedback control: A phaseless controller can avoid switching paradigms across reaching, passing, and retracting by coupling giver and receiver movements, but one implementation assumes known object mass.
  • Hybrid control: Dynamic Movement Primitives combine stronger feedforward influence early in motion with stronger feedback influence near physical object exchange.
  • Physical exchange: During physical exchange, the giver uses vision and force feedback after contact to assess grasping before releasing, while the receiver plans a stable and task-appropriate grasp.

A. Grip Force Modulation

Grip-force modulation coordinates object safety with transfer efficiency: the giver maintains stability and releases in response to the receiver, while safety also depends on disturbance handling, motion planning, and social conventions.

  • Grip-force coordination: Grip-force release combines anticipatory and somatosensory feedback control, with release speed correlated with the receiver’s reaching velocity.The giver is associated with object safety, while the receiver modulates exchange efficiency.
  • Grip-force modulation: The giver typically uses excess grip force to prevent slipping or dropping, and grip force varies linearly with load force except when either actor supports very little load.
  • Grip-force release: Proactive release improves handover fluency and subjective perception compared with a fixed release strategy.
  • Disturbance handling: The giver should recognize unwanted receiver contact as a disturbance and compensate to maintain a stable grasp rather than releasing the object.
  • Failure detection: Object acceleration measured with an optical gripper sensor can indicate handover failure when the object drops.Humans primarily rely on vision to detect object falls during handovers.
  • Safety scope: Handover safety includes human, object-transfer, and robot safety, supported through software and hardware measures.Collaborative-robot behaviour is also regulated by ISO/TS 15066:2016.
  • Safety during transfer: Safe handovers require appropriate pulling force and timing because errors can pull the human with the object or allow the object to drop.
  • Social safety: Social conventions such as offering a knife by its handle can make behaviour safer and more socially acceptable than presenting the blade.

VI. METRICS

Handover evaluation requires standardized qualitative and quantitative measures spanning task performance, human physiological responses, and subjective experience, because single success rates do not capture interaction quality.

  • Metric design: Standardized metrics could make handover techniques easier and fairer to compare, but the breadth of interaction aspects makes universal metric sets difficult.Metrics should assess handovers both qualitatively and quantitatively.
  • Metric categories: The review groups metrics into task performance, psycho-physiological responses, and subjective user questionnaires.
  • Task performance metrics: Success rate is the number of successful handovers divided by the total number of trials.
  • Task performance metrics: Success rate alone gives a statistical view, does not explain how or why errors occurred, and is difficult to compare across different experimental protocols.
  • Fluency metrics: Fluency can be evaluated using concurrent activity, human idle time, robot idle time, and robot functional delay.These measures relate to task effectiveness and interaction effort.
  • Human-response metrics: Psycho-physiological measures such as electromyography can quantify motor activity and help estimate the human partner’s affective state.

C. Subjective Metrics

Subjective metrics capture the human partner’s perception of handover quality, including fluency, trust, cooperation, safety, comfort, and cognitive workload. Reviewed studies commonly used post-study Likert-scale surveys, but protocols varied in metrics and test objects.

  • Subjective measures: Subjective metrics assess perceived task difficulty, cooperation, trust, robot contribution, anthropomorphism, animacy, likeability, intelligence, safety, and discomfort.The RoSAS framework organizes social perception around warmth, competence, and discomfort, while NASA-TLX can assess cognitive workload.
  • Evaluation practice: Post-study Likert-scale surveys most commonly evaluated interaction fluency, safety, comfort, satisfaction, ease of use, and perceived competence.These surveys were the most common vehicle for user studies in the reviewed papers.
  • Test objects: The reviewed experiments predominantly used a single object class, especially cylindrical bottles and rectangular boxes.Some studies instead used sensorized custom objects or application-specific objects such as flyers.
  • Protocol consistency: Experimental protocols lacked uniformity in post-handover tasks, evaluation metrics, and the number of test objects.The review identifies standardization as necessary for fairer comparison and more fluent handovers.

A. Open challenge 1: Adaptability

Adaptability remains an open challenge because human partners differ in how they interact with robots, and repeated accommodation can cause fatigue. The review highlights the need for robots to integrate cognitive and physical reasoning while adapting to partner feedback.

  • Adaptability: Robots should display adaptation and understanding to better match human handover skills of understanding and adaptation.The review treats cognitive coordination as important alongside physical coordination for robots to function as partners rather than only tools.
  • Review scope: The review tables compare R2H and H2R systems across handover phases, sensors, location adaptation, post-handover tasks, metrics, and object counts.These dimensions frame how adaptability and experimental coverage are assessed across real-robot studies.
  • Adaptability: Repeatedly accommodating a robot can fatigue human workers during extended interaction.Different users also interact with a robot in different ways during handovers, making partner feedback important.

2) Communication:

Communication is central to coordination, but reviewed robots generally provide fewer communication cues than humans. The review therefore identifies communication improvements as important for more natural and fluent handovers.

  • Communication: Robots generally lack communication skills for object handovers, while most prior effort emphasizes motion planning, control, grasping, and perception.Only a minority of reviewed papers included communication in their implementation.
  • Communication: Improved robot communication cues could increase the naturalness and fluency of human-robot handovers.Communication supports coordination by helping the human partner predict the robot’s actions.
  • Physical exchange: Grip-force modulation and object-fall handling also remain underexplored, with excessive receiver pulling force potentially endangering the human partner.The review calls for further investigation of grip modulation and broader hardware, grasping, and object choices.

1) Role of the post-handover task:

A post-handover task evaluates whether the receiver can use the object immediately, extending assessment beyond the transfer itself. The review also proposes a reproducible metric set covering objective performance and human experience.

  • Role of the post-handover task: Post-handover tasks test whether the receiver can use the object directly or must re-adjust the temporary grasp.Re-adjustment can increase task time and strain while reducing perceived partnership and increasing cognitive difficulty.
  • Role of the post-handover task: Including a receiver task helps assess the overall dyad and the human partner’s experience after the object transfer.The review argues that handover quality should not be evaluated solely at the moment of physical exchange.
  • Proposed set of metrics: The proposed minimal metric set includes success rate, total handover time, receiver task completion time, fluency, trust in the robot, and working alliance.These metrics cover objective task performance and subjective human experience.
  • Proposed set of metrics: Psycho-physiological measures are excluded from the minimal set because body-worn sensors make them difficult to standardize and deploy broadly.The proposed set prioritizes measures that are clearly defined, reproducible, and easy to measure.

3) Objects:

Object selection is a major source of variation in handover studies: most use one easy-to-grasp object class, limiting behavioral coverage and fair comparison. The review therefore proposes broader object sets alongside standardized metrics, while noting that object choice should follow each study’s focus.

  • Objects: Most handover studies use a single object class, typically cylindrical bottles or rectangular boxes that are easier to grasp.The review argues that this limits generalization across objects and tasks.
  • Objects: Experiments should include power, intermediate, and precision-grasp objects to elicit varied offering, reception, and post-handover behaviors.Different weights and shapes can also support investigation of distinct manipulation tasks.
  • Objects: Object choice should remain dependent on the specific focus of each study.The review presents broader object coverage as a general recommendation rather than a universal experimental requirement.
  • Objects: A minimal set of metrics and objects can improve fair comparison across handover approaches despite their non-uniform protocols.The proposed coverage targets common aspects including communication, planning, and grip release.
Loading 2007.12952v2…