Introduction
Object handover is a short task with several failure points. The robot has to approach without hitting the person, present the object in a reachable pose, detect contact or intent, regulate grip force and release only when the other person has control.
A useful evaluation records success, transfer time, peak interaction force, drops, premature releases, failed intent detections and recoveries across different people and objects.
Key facts
- A hand entering the camera frame is weak evidence that the person has control of the object.
- Vision, force, tactile and object-motion cues can be combined for release decisions.
- Heavy, fragile, flexible and slippery objects need different grip and release settings.
- Handover testing should include aborted reaches, changed intent and partial grasps.
Detect when the person is ready
Vision can estimate hand position, reach direction and object pose. Force or tactile sensing can confirm contact and loading. A release rule based only on proximity can drop the object when the person reaches past it or changes their mind.
Useful systems combine several cues and include a timeout. The robot should know how to retract or wait when readiness is uncertain.
Control grip force during the exchange
The grip must prevent slip without crushing the object or resisting the human pull. Wrist force sensing, fingertip tactile sensing, motor current and impedance control can all contribute depending on the hardware.
Testing should report the object mass and material. A rigid box, paper cup, cable and fabric item create different contact and deformation behavior.
Choose a release rule that can be tested
Define the measurable event that triggers release: sustained external pull, load transfer detected at the wrist, tactile contact pattern, or a combination. Report the threshold and dwell time when those values are public.
Measure premature release and delayed release separately. A robot that never drops an object because it refuses to let go has not solved the task.
Test recovery and changed human intent
Include cases where the person reaches, stops, moves to the other side, grasps weakly or lets go after initial contact. The robot should either keep control or move to a safe recovery state.
For repeated evaluation, use the same set of people, object poses and object masses, then add controlled variation. That makes improvements easier to measure.
What should a published handover result include?
Report number of trials, participants, object set, robot control mode, success definition, transfer time, drops, contact forces when measured and human interventions. State whether the robot knew the object pose in advance or estimated it online.
A short clip can show that a handover occurred. It cannot establish reliability across users and objects without repeated trials.
Limitations and missing information
- ISO/FDIS 13482 remains under development in the cited 2026 catalogue entry and its service-robot scope should not be treated as a universal humanoid handover protocol.
- The cited sources do not establish one safe grip-force or release threshold for every object and user.
- No human-subject experiment is reported by TechniaHQRobot on this page.
Conclusion
A handover result becomes useful when the reader can see how readiness was detected, how the robot knew the person had control, what forces were measured and how often the transfer failed or needed recovery.
Sources and methodology
The guide uses public standards material to frame safety and test questions. It does not infer thresholds or success rates that the cited documents do not publish.
Related TechniaHQRobot guides
Share this article
Share the current TechniaHQRobot article page.
Continue reading
Open the latest robotics reporting, Physical AI analysis and hardware notes.