Robot-to-instrument alignment: building the positioning error budget

When we review a mobile manipulator, we start at the instrument handoff point. The held plate or rack has to arrive within the allowed position and angle, using the actual gripper, load and station fixture. That is the requirement we work back from when checking the base, arm and mounting geometry.

Define the allowed error at the handoff point

For laboratory automation, the useful question is: What is the maximum allowable position and orientation error at the task interface? That interface may be the center of an SBS plate nest, a tube-rack guide, a drawer opening, a capper fixture, a reader loading position or another constrained handoff point.

A mobile manipulator combines at least two motion systems: a mobile base and a manipulator. The final task pose also depends on the mechanical relationship between them, the tool center point (TCP), the gripper, the object being handled and the instrument-side fixture. Each element can be individually repeatable while the complete system still misses the handoff.

This is why component datasheets are necessary but insufficient. ISO 18646-2:2024 describes navigation-performance test methods for mobile service robots, including pose accuracy and repeatability, while ISO 9283 addresses performance criteria and test methods for manipulating industrial robots. These are useful engineering references for component-level performance, but their applicability depends on how the final platform is classified and they do not replace an end-to-end task-level error budget. ISO 18646-2 also explicitly does not verify or validate safety requirements.

1. Start with a task coordinate frame

Before discussing tolerances, define the frame in which success is measured. For an instrument handoff, a practical frame is usually attached to a physical datum on the instrument or its loading fixture. The required pose is then expressed as X/Y/Z position, roll/pitch/yaw orientation and the permitted approach direction.

A single “± mm” value is often misleading because a laboratory handoff can have very different tolerance in different directions. A plate may tolerate more vertical error before final seating but very little lateral error at guide features. A tube rack may have adequate X/Y clearance but be sensitive to yaw. A drawer may require both a position window and a constrained approach vector.

2. Separate navigation, docking and manipulation errors

3. AMR repeatability is not instrument alignment

An AMR can repeatably navigate to a map pose and still be insufficient for direct manipulation. The navigation system estimates the base pose in a map; the instrument task is referenced to a physical opening or fixture. Any mismatch between the map frame and the physical station becomes part of the error chain.

For this reason, a mobile manipulator often benefits from a second, local alignment layer near the instrument. Depending on the application, that may be a mechanical docking datum, a fiducial, a camera-based correction, a laser feature, a station marker or another local reference. The purpose is to reduce the uncertainty between the mobile base frame and the instrument frame immediately before manipulation.

Docking should therefore be tested independently from free navigation: approach the same station repeatedly from different starting positions and headings, then measure the resulting base pose relative to the station datum.

4. The arm datasheet still does not give task accuracy

Manipulator repeatability describes how consistently the robot can return to a commanded pose under defined conditions. It is not the same as absolute accuracy, and neither value automatically includes the mobile base, mounting interface, tool calibration or instrument geometry. Workspace location also matters: the arm may behave differently near singular configurations, at large extension, or with different payload and center of gravity.

For an OEM platform, the relevant question is not the best specification in the brochure but whether the chosen arm can meet the task window in the actual installed geometry.

5. TCP calibration is a real part of the budget

The Tool Center Point is the point on the end effector that the robot controller uses for positioning. If the configured TCP does not match the real gripper geometry, the robot can repeat its programmed motion very well and still put the physical tool in the wrong place.

TCP error can come from machining tolerance, assembly, tool replacement, finger replacement, collision, service work or an incorrect calibration procedure. A small angular error can become a meaningful lateral error at the tip of a long gripper. For small angles, a useful approximation is: lateral error ≈ tool length × angular error (radians).

6. Gripper repeatability and labware seating are often underestimated

The robot can place its TCP repeatably while the object inside the gripper moves differently from cycle to cycle. Two-finger grippers are sensitive to finger geometry, pad compliance, part surface, closing sequence and whether the object is fully seated against a locating feature.

• What surfaces define the object position inside the gripper?
• Can the object rotate or slide after pickup?
• Does grip force deform the object?
• Do different labware suppliers or lots change the seating position?
• Can the gripper confirm that the object is present and correctly seated?

A good task-level test therefore measures the position of the handled object, not only the empty robot flange or empty gripper.

7. Instrument tolerance belongs in the same calculation

The receiving station is not a perfect coordinate system. Its nest, tray, drawer, guide rails and mounting structure all have tolerances. The instrument may also move slightly during service or relocation. Where practical, the station should provide clear datums or self-locating geometry. Chamfers, lead-ins, tapered guides, compliant insertion features and kinematic locating elements can make the task tolerant of small residual errors.

The best engineering solution is often not to demand a much more accurate robot, but to design the robot and instrument interface so the permitted error window is larger.

8. How should the errors be combined?

Do not add raw supplier numbers directly. Each contributor first has to be expressed in, or propagated into, the same task coordinate frame. Translational and angular errors should be treated separately, and angular uncertainty can create translational error through lever arms such as the tool length, gripper offset or distance from the arm base to the instrument.

• Worst-case stack: propagate bounded contributors conservatively when the requirement must be guaranteed and several contributors could align in the same direction.
• Statistical / RSS estimate: use root-sum-square only for contributors that can reasonably be treated as independent, approximately zero-mean random variables after systematic offsets have been removed.
• Non-linear or strongly coupled chains: use a sensitivity/Jacobian model, simulation or Monte Carlo propagation when frame transforms, orientation errors, compliance or correlations make simple scalar stacking misleading.

Systematic offsets such as a bad TCP calibration, an incorrect base-to-arm transform or a shifted station datum should not be hidden inside a random RSS calculation. They should be calibrated out or explicitly bounded.

A practical worksheet should classify contributors as systematic, repeatable, random, correlated, configuration-dependent or environment-dependent. That classification tells the team whether the right response is calibration, mechanical redesign, a local sensor, tighter manufacturing, software compensation, re-teaching or a larger handoff tolerance.

9. Validate the complete chain at the task interface

1. Start the mobile platform from several positions and headings.
2. Navigate to the station and perform the normal docking or local alignment sequence.
3. Move the arm using the production TCP, payload and gripper configuration.
4. Pick or place representative labware using the intended approach path.
5. Measure success at the instrument datum, not only in robot coordinates.
6. Repeat enough cycles to characterize the distribution and identify outliers.
7. Repeat after relevant disturbances such as restart, charging, tool change or service adjustment if they are part of the use case.

The acceptance criterion should be defined before testing. “The robot reached the station” is not a measurable pass condition. Examples include maximum X/Y/Z error at a datum, angular error, successful insertion rate, no-contact clearance, or successful placement into a defined fixture window.

10. Where should ownership sit?

We take responsibility for the agreed hardware configuration, base-to-arm relationship, gripper integration, calibration and hardware acceptance checks. Any local alignment function is defined in that scope. Your scheduling software decides when the robot moves to a station and what the laboratory workflow does with the reported result.

The instrument OEM or system integrator may own the physical station datum and access geometry. These boundaries should be written into the project interface document because an unexplained alignment failure can otherwise become a dispute between the AMR supplier, arm supplier, gripper supplier, instrument supplier and scheduler team.

What should be agreed before the build?

• The instrument datum and required handoff pose.
• Allowed translational and angular error by axis.
• Representative labware and payload range.
• Required approach direction and collision-clearance envelope.
• How the platform establishes local alignment at the station.
• TCP and base-to-arm calibration method.
• Which error contributors are measured during FAT.
• The end-to-end pass/fail criterion and test cycle.

Technical references

• ISO 18646-2:2024 — Robotics — Performance criteria and related test methods for service robots — Part 2: Navigation
• ISO 9283 — Manipulating industrial robots — Performance criteria and related test methods
• Universal Robots — TCP Configuration