Best Six-Axis Force Torque Sensors for Robot Integration

For a supported cobot, shortlist Robotiq FT 300-S, OnRobot HEX and the relevant Bota integration kit before committing to a custom driver. For a machine you control end to end, start with ATI Mini45 or ME-Systeme K6D and choose the measurement range, interface and calibration together. An easy-to-install sensor and a fast-enough measurement chain solve different problems.

This guide uses product documentation. It does not report independently measured noise, accuracy or controller latency. A six-axis sensor reports forces and moments around its own coordinate system. A robot that already estimates contact through its joints may still benefit from direct wrist measurement, but the new sensor adds stack height, cable routing and its own calibration work.

Compare the integration paths

Candidate	Where it belongs on the shortlist	What to verify before ordering
ATI Mini45	Custom wrist where the calibration and electronics must fit the application	Selected calibration, combined load limits, interface and tool-flange adapter
Robotiq FT 300-S	Supported cobot where integration time is a constraint	Robot generation, driver, supported force operations and data access
OnRobot HEX	Cobot end-of-arm package for force-sensitive tasks	Exact HEX variant, robot support and the available data path
Bota Systems	Compact instrumentation or a supported robot kit	Range, connector, data protocol and software generation
ME-Systeme K6D	Machine or test fixture with specific mechanical and range requirements	Flange, electronics and calibration for the selected configuration

ATI Mini45 for custom calibration and integration

ATI’s Mini45 specification lists calibrations with different force and moment ranges. Quote the calibration code, electronics and cable together. One headline force range cannot tell you whether the sensor can tolerate the wrench generated by an offset gripper. Read the load limits and check the expected combination across axes with the vendor. Use the published drawing to confirm the alignment features and mounting hardware fit the actual wrist and tool plate.

Robotiq and OnRobot for supported cobots

Robotiq’s FT 300-S documentation lists supported robot families. Check the robot generation, controller software and available integration components. A robot-program instruction to zero the sensor can be useful for a particular operation. It does not remove gravity and inertia for an arbitrary tool pose. Robotiq’s calibration notes distinguish factory calibration and a temporary offset.

OnRobot HEX is another route when you want vendor-supported end-of-arm integration. Ask for the data path available on your controller for the task you actually plan to run. A polished setup screen, raw samples for analysis and a force loop that meets your timing requirement are three separate deliverables. Test those in the intended controller environment, with the tooling connected.

Bota and ME-Systeme for different packaging constraints

Bota’s MiniONE is a compact six-axis option when distal mass or mounting space is tight. The Bota shop also lists robot-specific kits using different transducers. Choose the sensor for the load and measurement job first, then check which integration hardware is available. Small sensors can be valuable for fingertips and light tooling, but their size says nothing about whether they can survive a wrist collision.

ME-Systeme’s K6D range lets you compare distinct mechanical and range configurations. Start with a drawing of the sensor and mating parts. Check whether the ordered device comes with the required amplifier or digital interface, calibration files and a way to reproduce the same measurement after servicing the cell.

Work backward from the tool tip

A 2 kg tool produces roughly 20 N of force under gravity. With the center of mass 0.15 m from the measurement origin and gravity perpendicular to that offset, that alone produces about 3 N·m of moment. A lateral process force, robot acceleration or a collision can add more. Those example numbers are for sizing intuition. Estimate all relevant poses and contact directions for your actual tool. Check the allowable combined loads as well as the continuous and overload limits for each individual axis.

Then define the smallest force change that would affect the task. Ask for data over the expected temperature range and a clear definition of noise and usable bandwidth. Log timestamped samples with the intended cable routing and motion nearby. Look at age of data and jitter at the controller. A fast transducer feeding a slow or irregular fieldbus can still make a poor feedback signal.

Qualify the assembled cell

1. Confirm axis direction using known loads applied independently along the tool axes. Test an offset load whose moment you can calculate.
2. Move through working poses with no contact and check the residual after tool-weight compensation.
3. Record the contact transient on a representative part. Compare what reaches the robot program with what the acquisition tool shows.
4. Check behavior on stale data, disconnection and saturation. Use the cell’s engineered safety functions independently.
5. Save the serial number, calibration, fixture and controller configuration so a replacement is reproducible.

See surface-finishing programming tools for path generation, robot tool changers for repeatable stack changes, adaptive grippers for grasp selection, the robot API directory for integration research and robot runtime safety for the safety boundary.

FAQ

Can I use a wrist sensor as a safety-rated device?

Use a sensor for safety only where its certification and the full safety architecture support the intended function. A normal process-force alarm is not an independent protective function.

Why does the same force produce a different torque after changing tools?

Moving the point of application relative to the sensor changes the moment arm. Update the tool transform and verify it with an offset load after installing the new tool.

What is the difference between zeroing and payload compensation?

Zeroing offsets the current reading. Payload compensation accounts for changes in how the tool’s weight projects onto the sensor as the robot moves.

Does a larger range make a better sensor?

It gives more headroom when the range and load limits fit. Compare that headroom with the resolution, noise and drift your task can tolerate.