Engineering guide2026-09-028 min read

Mobile-manipulator docking acceptance: test in workpiece coordinates

An AMR reaching a navigation goal does not mean its arm is aligned to the workpiece. Accept the complete transform chain and the task margin in workpiece coordinates.

By Matrix Dimension Robotics engineering team

Concept rendering of a wheeled humanoid mobile manipulator working at an industrial workstation

The short answer: do not accept a mobile manipulator from an “AMR arrived” flag or one base-repeatability figure. Measure global arrival, local pose relative to the station, workpiece or fixture registration, and arm/tool execution separately, then verify the final task margin in workpiece coordinates. If vision, mechanical guidance or compliant search closes the remaining gap, prove that the compensator works under reachable disturbances, rejects bad data and recovers safely.

Arrival and readiness to manipulate are different states

ISO 19649 defines docking as reaching and/or connecting to a station, facility or another mobile platform to perform an intended task. Its definition of mobile-robot pose can include the poses of both the platform and any attached manipulator relative to the world. That scope is useful: a map goal is only one end of the chain. The frame that matters to the task is usually a fixture, machine, tray, connector or workpiece.

NIST's comparison of mobile-manipulator registration methods treats docking and workpiece registration as a performance problem whose outcome is fast, repeatable and accurate end-effector alignment to a physical target. The later NIST AMS 100-45r1 report uses a reconfigurable workpiece artifact and external optical ground truth to measure a mobile-manipulator-on-a-cart and compare two registration techniques. The engineering implication is clear: base, arm and workpiece specifications do not remain independent once the task begins.

Separate the error layers before choosing a remedy

Layer	What to measure	Typical mechanism	What it does not prove
Global arrival	Success, time and exceptions when reaching a staging region from representative routes and loads	Map localization, path planning, traffic control	Alignment to the workpiece
Local docking	X/Y/heading distribution relative to a station datum, plus retries and failures	Local target, vision/laser servoing, mechanical guidance	Part variation inside the fixture
Workpiece registration	Transform and uncertainty from station, fixture or part to the robot base/sensor	Fixed datum, vision/touch measurement, reteaching	Arm, TCP and tool capability
End task	Approach, grasp, insertion, inspection or process result in workpiece coordinates	Trajectory control, compliance, search, process confirmation	Performance beyond the tested task envelope

ASTM F3499-21 provides a quantitative method for confirming A-UGV positioning or docking maneuverability and repeatability. Its public scope permits arms and other load-transfer mechanisms, and compares repeatability on each measured axis with a defined complete task performance margin. That is a strong base-layer framework; the buyer still has to supply the real workpiece, approach direction, tolerances and application success criteria.

Four precision strategies, selected by the task

Strategy	Use it when	Main variation absorbed	Acceptance focus
Tighter local base docking	A fixed dock serves a stable fixture and workspace or sensor view is narrow	Planar arrival and heading residuals	Multiple approaches, floor/load change, distribution, retry and retreat
Register the workpiece at the station	The fixture or part moves and the task needs part, not map, coordinates	Docking residual, fixture drift, loading variation	Observability, ground truth, occlusion/glare, stale-data rejection
Mechanical datum or guidance	Rigid contact, power/data connection or a very narrow entrance is required	Translation and angle inside a passive capture range	Capture envelope, force, wear, jamming and incomplete-seat detection
Compliant end-effector search	Slow contact exploration is allowed and force/displacement can confirm the feature	Small residual pose errors	Search envelope, force limit, timeout, false contact, exit and reset

These mechanisms may be combined, but each needs a distinct job. Base yaw error becomes a larger lateral displacement as the target moves farther from the rotation centre. A posture near the edge of the arm workspace may also lose orientation or recovery margin. Use the actual station geometry, arm posture and task tolerance; do not add unrelated catalogue “accuracy” figures as though they shared one frame and measurement method.

An executable FAT-to-SAT acceptance workflow

Freeze the task and outcome.Define the part, tool, approach direction, allowed pose error, process force or vision result, cycle window and states that must count as failure. Keep arrival, registration and task success as separate events.
Map the transform and evidence chain.Include map-to-station, station-to-platform, platform-to-arm-base, flange-to-TCP, sensor-to-robot and fixture-to-part. Record who establishes each transform, how it is checked and what invalidates it.
Measure docking on its own.Following the intent of ASTM F3499, repeat from the real set of start locations and approach directions. Keep per-axis distributions, duration, retries and failures—not only a mean. Match payload, wheel/castor state and floor conditions to the application.
Measure local registration.Use independent ground truth or a traceable artifact to challenge the vision, laser, touch or mechanical datum. Introduce reachable fixture movement, occlusion, lighting and target variation, and verify that stale or wrong transforms are rejected.
Measure the end result in workpiece coordinates.With tool, program and speed frozen, span docking residuals and arm postures. Record task result, deviation, contact force, cycle and intervention by condition. ISO 9283 can support separate manipulator performance characterization, but it does not replace the complete mobile-task test.
Add only reachable disturbances.NIST's mobility measurement programme lists slip, uneven floors, dynamic obstacles, vibration, heavy or dynamic loads and forced delocalization among issues observed in prior localization tests. Select the subset that exists at the site and identify whether each affects arrival, registration or manipulation.
Exercise failure and recovery.Occlude a local target, move the fixture, begin outside the capture envelope or withhold task confirmation. The system should stop, retreat, redock, reregister or hand over—not continue in a known-bad frame.
Freeze the delivery baseline.Retain station definitions, targets/datums, transform versions, raw measurements, task records, failure classes and maintenance retest rules. Revalidate the affected links after wheel, sensor, arm, tool or fixture changes.

A software success flag needs a physical meaning

The official Nav2 Docking Server tutorial separates navigation to a staging pose, dock detection and pose refinement, closed-loop approach, and contact or charging confirmation. Its action also reports error codes, retries and elapsed time. This is a useful state-machine pattern, but its example thresholds and plugin behaviour are not manufacturing acceptance limits. At a manipulation station, “docked” should lead to further checks for valid workpiece registration, arm reachability, an updated collision scene and physical task confirmation.

Minimum evidence pack

Task tolerance and definitions of success, failure and human intervention;
complete transform chain, calibration/registration method, ground truth and uncertainty;
raw docking distributions and failures across starts, directions and real payloads;
station, fixture, target, floor and environmental disturbance matrix;
end-effector outcome, cycle, retry and recovery evidence in workpiece coordinates;
configuration versions, maintenance triggers, retest scope and safe exit path.

Safety boundary: ISO 10218-2:2025 addresses integration safety for industrial robot applications and cells. A system with a mobile platform also needs a standards-scope review against its actual use, vehicle, machinery and jurisdiction. This article describes performance evidence; it does not replace risk assessment, functional-safety design, safeguarding validation or conformity work.

Frequently asked questions

If AMR repeatability passes, will the arm task pass too?

Not necessarily. Fixture loading, station datum, platform-to-arm mounting, TCP, sensor registration and arm posture all enter the final chain. Test the complete outcome in workpiece coordinates.

Can vision registration replace precision docking?

It can absorb part of the docking residual when the target is observable, the arm retains reach and orientation margin, and bad registration is detectable. It cannot recover a target outside view or the mechanical task envelope.

Why repeat docking from different starts and directions?

Route, approach direction, wheel-floor contact and payload can change the final distribution. One route and one average cannot represent retry and failure behaviour across the real task envelope.

What is the key difference between FAT and site SAT?

FAT can quantify docking, registration and end-task layers under controlled conditions. SAT must add the real floor, fixture, lighting, traffic, payload, operating routine and recovery path.

Sources

These primary sources support the material facts and engineering boundaries discussed above.

Evaluating robot control, bimanual manipulation or a mobile platform?

Talk to Matrix Dimension →