6-Axis Force Sensor Axis Sign Convention
A 6-axis force sensor axis sign convention defines positive force and moment directions in a stated frame. Compare a known load using the drawing axes and reporting setup.
By HexaForce · Published · Evidence reviewed · Illustrative static screen; no device connection.
1. Enter a known load and six readings
Use the drawing axes. Keep motion disabled and stay within the device’s rated combined-load limits.
2. Illustrative example result
These are prefilled example readings, not measurements from your sensor. Replace all six readings to run your check.
Full coordinate handedness: NOT VERIFIED. A single step cannot establish all axes, calibration accuracy, overload margin or control stability. Active channels use an illustrative ±10% magnitude screen; idle force ±max(2% applied load, 0.01 N), idle torque ±max(2% F×arm, 0.001 N·m). These are not supplier acceptance limits. See method and limitations.
| Channel | Expected | Measured | Screen | Interpretation |
|---|---|---|---|---|
| Fx | 0 N | 0 N | NEGLIGIBLE | Idle screening window ±0.98 N; not a crosstalk specification. |
| Fy | 0 N | 0 N | NEGLIGIBLE | Idle screening window ±0.98 N; not a crosstalk specification. |
| Fz | -49 N | -49 N | PASS | Matches this load step within the illustrative ±10% magnitude window. |
| Mx | 0 N·m | 0 N·m | NEGLIGIBLE | Idle screening window ±0.098 N·m; not a crosstalk specification. |
| My | 4.9 N·m | 4.9 N·m | PASS | Matches this load step within the illustrative ±10% magnitude window. |
| Mz | 0 N·m | 0 N·m | NEGLIGIBLE | Idle screening window ±0.098 N·m; not a crosstalk specification. |
What to decide before changing a sign
Define the loaded body first
The checker defaults to force applied to the tool side, with +Z outward. Equal/opposite body reaction reverses all six wrench components at the same origin and basis.
Checker assumption; static action–reaction balance.Calculate moments at a stated origin
With r = (0.1, 0, 0) m and F = (0, 0, −49) N, My = +4.9 N·m. This is a derived example, not a sensor test result.
Cross-product expansion in section 2.Verify units before interpreting signs
Force and torque can use separate scales. Decode values into named axes before entering them here; packet order is not inferred by this checker.
ATI manual §9.3; ROS REP-103.One step cannot validate the frame
Independent force and moment directions, mechanical drawing checks and device-specific acceptance limits are still required. This tool does not validate control stability.
Observability limit: unloaded axes remain untested.Evidence: ATI Net F/T manual, 9620-05-NET FT-23, §§4.6 and 9.3; ROS REP-103 — Units and coordinate conventions (official source). The numerical examples below are derived from the stated geometry.
1. Choose the physical basis and reporting convention
A label such as “compression positive” is incomplete without the loaded body, basis and moment definition. The options below are explicit checker assumptions, not universal manufacturer defaults.
| Checker option | Centered compression | −Z at +X | Boundary |
|---|---|---|---|
| Applied wrench; +Z outward | Fz = −F | Fz = −F, My = +FL | Default physical right-hand basis; tool-side applied load. |
| Fz reporting polarity inverted only | Reported Fz = +F | Reported Fz = +F, My = +FL | Only Fz output is negated; decode it before spatial cross products. This is not a different Cartesian basis. |
| Opposite-body reaction | Fz = +F | Fz = +F, My = −FL | All six components negated; same origin and same basis. |
| A rotated or reflected installation | Not inferred | Not inferred | Use the drawing and a complete spatial mapping. Mounting selection here records context only. |
Scope check: ISO 9787:2013 — Robot coordinate systems and motion nomenclatures defines robot coordinate systems; the sensor drawing and active reporting configuration determine the actual axes and output signs.
2. Derive the expected force and moment signs
Derived values below use the default applied-wrench convention, F > 0 and perpendicular arm L > 0. All unlisted components are zero only for the idealized fixture stated.
| Physical test | Force vector | Moment vector | Fixture condition |
|---|---|---|---|
| Centered compression | (0, 0, −F) | (0, 0, 0) | Line of force passes through sensor origin. |
| Centered tension | (0, 0, +F) | (0, 0, 0) | Line of force passes through sensor origin. |
| Centered +X / +Y shear | (+F, 0, 0) / (0, +F, 0) | (0, 0, 0) | Force line passes through origin; an offset adds a moment. |
| −Z force at r = (L, 0, 0) | (0, 0, −F) | (0, +FL, 0) | My = rzFx − rxFz. |
| −Z force at r = (0, L, 0) | (0, 0, −F) | (−FL, 0, 0) | Mx = ryFz − rzFy. |
| Pure Z couple | (0, 0, 0) | (0, 0, ±FL) | Two equal/opposite forces separated by L; CCW is positive viewed FROM +Z toward origin. A single tangential push also produces force. |
Mx = ryFz − rzFy; My = rzFx − rxFz; Mz = rxFy − ryFxForce acts on the tool side. An opposite-body reaction reverses both force and moment at the same origin. For an arbitrarily rotated installation, first express all vectors in one physical basis.
3. Check units, payload order and reference frame
Use device-specific protocol evidence before interpreting a sign. ATI §9.3 illustrates separate force/torque scaling; ROS REP-103 specifies SI conventions. No generic register map, rate or byte order is assumed here.
| Check | Known rule or calculation | Evidence to obtain | If unknown |
|---|---|---|---|
| Force units | 1 kN = 1000 N; 1 lbf ≈ 4.4482216 N; 1 kgf = 9.80665 N | Active calibration unit and counts-per-force factor | Retain raw data; do not treat counts as N. |
| Torque units | 1 N·mm = 0.001 N·m; 1 in·lbf ≈ 0.112984829 N·m; 1 kgf·cm = 0.0980665 N·m | Active torque unit and counts-per-torque factor | Check conversion direction before changing scale. |
| Axis names and order | Mx/My/Mz may be labelled Tx/Ty/Tz | Firmware-matched payload definition and serial-linked calibration row order | Enter no inferred pin or register assignments. |
| ROS 2 wrench | force and torque are separate vectors; stamped data adds frame/time context | Publisher configuration, actual frame_id and verified sensor transform | A changed frame_id alone does not rotate values. |
| Raw analog bridges | Coupled bridge signals need the correct calibration mapping | Serial-linked matrix, input order, excitation and scaling | This checker accepts calibrated named channels only. |
Primary references: ATI Net F/T manual, 9620-05-NET FT-23, §§4.6 and 9.3, ROS REP-103 — Units and coordinate conventions (official source), ROS 2 Jazzy geometry_msgs — WrenchStamped definition. Protocol-specific offsets, packet timing and available HXF drivers are unknown here and require configuration evidence.
4. Understand the screen and its limits
These transparent heuristic windows support a sign sanity check. They are editorial screening settings, not calibration accuracy, crosstalk ratings or supplier acceptance limits. No uncertainty budget or rated-load validation is performed.
| State | Checker rule | Meaning | Next action |
|---|---|---|---|
| Loaded channel match | Correct sign and measured magnitude within ±10% of expected | Only this component in this load step matches | Repeat independent directions; compare against the device’s own limits. |
| Weak/zero loaded response | Magnitude below 5% of expected | No reliable polarity evidence, even if the sign matches | Check load path, tare, scaling and decoding. |
| Idle force / idle torque | ±max(2% applied force, 0.01 N) / ±max(2% F×L, 0.001 N·m) | Unexpected response screen, not a measured crosstalk specification | Measure baseline noise; use configuration-specific uncertainty for acceptance. |
| Opposite sign / unexpected axis | Loaded response opposite to expectation / shear response on the other axis | Candidate convention, mapping or orientation discrepancy | Inspect all rows and run an independent test before changing anything. |
| Approximately 1000× scale | Measured/expected near 1000 or 0.001, within ±10% | Possible unit mismatch, not a confirmed cause | Compare source and destination unit labels. |
| Invalid or incomplete input | Blank, nonnumeric, nonfinite, nonpositive force or zero moment-test arm | No result generated | Correct fields or load a complete example. Numeric bounds protect the software, not the device. |
Small signals near noise, arbitrary multi-axis loads, different origins and pose-dependent gravity fall outside the simple fixture assumptions. Use a measured uncertainty budget and device limits for acceptance. ISO 21612:2021 — Road vehicles: crosstalk determination for multi-axis load cell concerns vehicle-test crosstalk; it is not the source of these screening windows.
5. Repeatable static installation verification protocol
This editorial procedure checks a known static load path. It is not a standardized calibration certificate or robot safety acceptance procedure.
- Document the sensor drawing, origin, axis directions, loaded body, serial number, calibration revision and active software transforms. With motion disabled, secure a fixture and confirm allowable combined loads.
- Record the baseline with the fixture and cable routing unchanged. Apply a centered compression and tension load; subtract the baseline and record both Fz signs. Do not tare away the test load.
- Apply separate centered +X and +Y forces using the drawing directions. Record all channels; response on the other shear axis may indicate mapping or a rotated frame.
- Apply −Z at a measured +X offset, then at a +Y offset. Under the default convention expect +My and −Mx respectively. Subtract lever self-weight baseline and use perpendicular distances.
- Apply a controlled pure Z couple in both directions. View FROM +Z toward the origin: CCW gives +Mz. Do not substitute a single tangential force without accounting for its net force.
- Repeat independent force/moment directions as needed, save the settings and compare to device-specific uncertainty and limits. Verify the full rotation and origin mapping before any controller commissioning.
6. Keep sensor origin and TCP moment separate
Let R rotate sensor components into TCP axes. Let r be the vector from the sensor origin to the TCP, expressed in sensor axes. For the same applied wrench:
F_tcp = R F_sM_tcp = R (M_s − r × F_s)The subtraction follows from M_s = M_tcp + r×F_s for parallel axes. If using a plus-sign formula, define p = −Rr, the vector from TCP to sensor expressed in TCP axes: M_tcp = R M_s + p×F_tcp. Leaving the vector direction unstated changes the answer.
The checker compares values at the sensor origin and does not perform this transform. See ATI Net F/T manual, 9620-05-NET FT-23, §§4.6 and 9.3 §4.6 for a device-specific example of factory origins and configured transforms; its settings must not be copied to another device without checking the manual.
7. Resolve risk without guessing configuration values
Device drawings and calibration evidence control the hardware limits. Generic brand comparisons cannot determine a particular robot’s flange clocking or sensor polarity.
| Risk / cost | Why it matters | Mitigation | Alternative |
|---|---|---|---|
| Misuse: sign change in a live control loop | A wrong feedback mapping can produce unintended motion | Disable motion, verify static mapping, then follow the controller’s commissioning procedure | Ask the integration owner to review the complete frame and feedback convention. |
| Cost: adapter rework or repeated calibration | Unspecified seating, tightening or cable loads can alter the baseline | Obtain approved fasteners, torque, surface and combined-load specifications; record the baseline | Request configuration review before machining or remounting. |
| Scenario mismatch: gravity on a rotated wrist | “Down” in the world is not necessarily −Z in the sensor | Resolve the known load into sensor axes; subtract the recorded no-test-load baseline | Use a controlled fixture aligned to the sensor drawing. |
| Scenario mismatch: a single tangential push | It creates net force as well as torque; a pure-couple prediction will disagree | Use an equal/opposite force pair with measured separation | Calculate the complete wrench for the real load point outside this checker. |
| Unverified product support | No model-specific tolerance, adapter alignment, driver or stability claim is established here | Request serial-specific drawing, calibration revision and tested interface evidence | Treat all HXF configuration capability as pending written confirmation. |
8. Four worked scenarios with explicit assumptions
These are calculated teaching scenarios, not measured customer deployments, verified compatibility claims or product performance results.
A. Default cantilever calculation
Premise: Illustrative: +Z up, applied load on tool side, r = (0.1, 0, 0) m; load = (0, 0, −49) N.
Calculation / check: My = rzFx − rxFz = 0 − 0.1×(−49).
Result / next step: Fz = −49 N and My = +4.9 N·m. Subtract the unchanged fixture baseline before entering measured readings.
B. Same wrench in different units
Premise: Illustrative: same load and geometry; force readout in kN and torque in N·mm.
Calculation / check: 49 N = 0.049 kN; 4.9 N·m = 4900 N·mm.
Result / next step: Enter applied force 0.049 kN, measured Fz −0.049 kN and My 4900 N·mm. The step should match; a correctly declared unit change is not a scale fault.
C. Unexpected shear direction
Premise: Illustrative: centered +20 N along sensor +X; displayed Fx = 0 and Fy = +20 N.
Calculation / check: Compare the physical load direction, packet fields and any enabled spatial rotation; repeat a +Y load.
Result / next step: A swap and a rotated frame remain competing hypotheses. This one observation cannot authorize changing raw wires or driver signs.
D. Move the reference origin
Premise: Illustrative: F_s = (0, 0, −49) N and M_s = (0, 4.9, 0) N·m; TCP is 0.1 m along sensor +X.
Calculation / check: For parallel axes, r points sensor → TCP. M_tcp = M_s − r×F_s.
Result / next step: M_tcp = 0 at the force application point. The sensor still experiences 4.9 N·m at its own origin; TCP values cannot be used alone to assess sensor overload.
9. Frequently asked engineering questions
Frame and sign
Is compression always negative on a 6-axis force sensor?
No. It is negative in this checker’s default tool-side applied-wrench example with +Z outward. Confirm the physical axes, loaded body, output polarity and transforms from the specific drawing and manual.
Does ISO 9787 define every sensor’s force output polarity?
Its public scope defines robot coordinate systems and motion nomenclature. It is not evidence for a universal sensor output polarity. The checker’s applied, reaction and Fz-reporting options are explicitly stated assumptions.
Why does −Z at a positive X offset give positive My?
Expand M = r×F. With rx = L > 0 and Fz = −F, My = rzFx − rxFz = +LF. This assumes the stated right-hand basis and moment measured at the sensor origin.
From which side is positive Z torque counterclockwise?
View FROM the positive Z end toward the origin. Counterclockwise is +Mz. Looking along +Z from the opposite end reverses the apparent rotation direction.
Can a single successful test prove a right-handed frame?
No. A single wrench leaves other independent directions unobserved. Check the drawing and complete mapping, including independent force and moment directions. This checker always reports full handedness as not verified.
What does the Fz-inverted reporting option mean?
Only the reported Fz sign is reversed relative to the physical default. The moment basis remains unchanged. Undo that reporting polarity before using the force vector in a physical cross product; it is not a new spatial coordinate system.
Data and units
What changes when I select kN or N·mm?
Applied force and force readings use the selected force unit; moment readings use the torque unit. Calculations compare values in SI internally. Changing a selector reinterprets existing numbers, so edit the numbers if you intend to preserve the physical load.
Will selecting a payload order rearrange my readings?
No. Enter readings by the named Fx/Fy/Fz/Mx/My/Mz axes after documented decoding. Payload order is recorded for review only; the checker does not decode packets, registers or raw bridges.
Can zero or blank readings pass a loaded channel?
A blank is invalid. An actual zero on an expected loaded channel is a missing response and cannot pass. An unloaded channel may be negligible within the explicitly displayed illustrative idle window.
Does a 1000× discrepancy prove a torque unit error?
No. A near-1000 or near-0.001 ratio is only a diagnostic clue. Check units, scale factors, actual load and calibration together. Convert N·mm to N·m with 0.001; the reverse direction uses 1000.
Does ROS WrenchStamped automatically transform a wrench?
No. The Jazzy message definition supplies a header and wrench, not a transformation algorithm. Verify the publisher’s frame_id, timestamp, units and actual transform. Renaming a frame does not rotate or shift the values.
Can I enter raw strain-gauge voltages?
No. Enter calibrated, named engineering-unit channels. Raw bridges require the serial-linked matrix, correct input ordering and scaling first. Do not swap raw bridge leads to imitate an axis remap.
Commissioning limits
Are the ±10% and idle windows acceptance specifications?
No. They are transparent heuristic screens for this tool. They do not replace calibration uncertainty, rated ranges, noise measurements or configuration-specific acceptance criteria, and they may be unsuitable for very small signals.
What changes when I choose a different mounting configuration?
Only the context saved in the record changes. The checker does not infer spatial rotation or gravity from a mounting label. Resolve your load into the sensor drawing axes before entering it.
How should I remove gravity and fixture preload?
Record the no-test-load baseline at the same pose with fixture, lever and cable routing unchanged, then use the difference under the known test load. A baseline from one pose does not compensate changing gravity at other poses.
What bolt torque, flatness and controller offset should I use?
Those are configuration-specific unknowns here. Obtain the approved sensor and robot drawings, fastener instructions and adapter transform. No generic brand label or screw size determines those values.
Does a step match mean I can enable robot force control?
No. It establishes only the displayed screen for one submitted load. Complete frame mapping, rated-load checks and controller-specific commissioning remain necessary. No stability margin, response time or device compatibility is established here.
What should I send for an engineering review?
Send the selectable record, drawing axes and origin, loaded body, sensor serial/calibration revision, units, decoded payload map, baseline and actual load geometry. Inquiry and brief links carry the submitted readings for review; capabilities remain subject to confirmation.
10. Sources, evidence scope and next steps
Reviewed 2026-11-02. ISO citations use public catalogue scopes only; no paid standard clauses or universal manufacturer settings are asserted. ROS examples refer to Jazzy message definitions; no HXF driver release has been tested for this guide.
- ISO 9787:2013 — Robot coordinate systems and motion nomenclatures
Public scope: robot coordinate systems and motion notation. It does not establish a universal sensor tension/compression output polarity.
- ATI Net F/T manual, 9620-05-NET FT-23, §§4.6 and 9.3
Device-specific factory origins, configurable tool transforms and separate force/torque count scaling. These details do not establish HexaForce product compatibility.
- ROS REP-103 — Units and coordinate conventions (official source)
SI units and right-handed frames. Body axes are x forward, y left, z up; a sensor frame still needs its documented transform.
- ROS 2 Jazzy geometry_msgs — WrenchStamped definition
A header (frame and timestamp) plus a Wrench containing force and torque vectors; no proof that a publisher has transformed the values.
- ISO 21612:2021 — Road vehicles: crosstalk determination for multi-axis load cell
Public abstract concerns crosstalk comparisons for vehicle crash testing. It is not used here as a generic calibration or robot acceptance specification.
For an installation review, include the sensor drawing and origin, serial-linked calibration, decoded units/order, baseline and actual test geometry. Use the result’s inquiry link to carry your submitted readings. Model ratings, CAD availability and software support require written confirmation.