humanoidsdata.com

Search

Search datasets, articles, and glossary terms for humanoids and embodied AI.

← All glossary terms

Hardware & control

Force–torque sensing

Force–torque sensing is the measurement of forces and rotational moments transmitted through a robot interface, commonly with a six-axis sensor reporting three force components and three torque components. The combined quantity is a wrench expressed at a specified origin and in a specified coordinate frame. It measures the resultant load at the sensor, not a spatial pressure map across every contact.

Also known as: force-torque sensing, force/torque sensing, F/T sensing, force-torque sensor, force/torque sensor, F/T sensor, wrench sensing

Updated

A six-axis sensor reports one resultant wrench

Modern Robotics packages three-dimensional force and moment into a six-component wrench. Commercial six-axis sensors such as those documented by ATI report forces along x, y and z and torques about those axes.

The reading depends on the sensor origin and coordinate frame. Moving the reference point changes the reported torque because the same force has a different moment arm. Gravity from a tool or payload can also appear in the measurement unless it is compensated.

Force–torque and tactile sensing are complementary

A wrist sensor can measure the net load transmitted through an arm but cannot normally show which fingertip or surface region produced it. A tactile array can preserve local pressure or shear while missing loads outside its instrumented area. Joint-torque estimates infer interaction through the mechanism and are not identical to a calibrated wrench measurement.

Controllers often use force–torque feedback for insertion, polishing, compliant motion, collision detection or payload estimation.

Wrench data needs more than six numbers

The record should include force and torque units, axis order, coordinate frame, reference point, sampling rate, calibration matrix, zeroing procedure, filtering, range and saturation. ROS 2's Wrench message separates force and torque; a stamped form is needed when timestamp and frame identity matter.

Dataset users should know whether gravity and tool weight were removed, whether values are raw or compensated, and where the sensor is mounted. Without that context, equal six-vectors can describe different physical loads.

Sources