A force sensor is a transducer that converts a mechanical load into an electrical signal proportional to the force, and in many designs the torque, acting on it. Among robotics sensors, it gives a robot a sense of how hard it is pushing or being pushed, which position-based sensing alone cannot provide.
In robot arms, the most common form is the six-axis force/torque sensor mounted between the wrist flange and the tool. It reports three force components and three torque components, so the controller knows both the magnitude and the direction of every contact. Single-axis load cells, joint torque sensors and distributed tactile arrays complete the family.
How a force sensor measures load
Nearly every force sensor measures force indirectly through deformation. A carefully shaped elastic body, often aluminum, steel or titanium, deflects by micrometers under load, and a sensing element converts that tiny strain or displacement into a signal.
The classic element is the foil strain gauge in a Wheatstone bridge, whose resistance changes with strain and produces a millivolt output that is amplified and digitized. Six-axis sensors place several bridges on a cross-shaped or spoke-shaped body and use a calibration matrix to separate the six load components from the raw bridge signals.
The table below compares the main sensing principles found in robotic force sensors.
| Principle | How it works | Strengths | Weaknesses |
|---|---|---|---|
| Strain gauge | Resistance changes with surface strain | Accurate, mature, static loads | Temperature drift, fragile wiring |
| Piezoelectric | Crystal produces charge under load | Very stiff, high bandwidth | Cannot hold a static reading |
| Capacitive | Gap between plates changes with load | Low power, compact | Sensitive to humidity and noise |
| Optical | Light intensity or wavelength shifts with deflection | Immune to electrical noise | Higher cost, complex readout |
| Motor current estimation | Force inferred from joint currents and a model | No added hardware | Low accuracy due to friction |
Key specifications of a force sensor
A force sensor datasheet is read in terms of range, resolution, accuracy, bandwidth and overload capacity. These parameters interact: a sensor with a large range usually has coarser resolution, and a stiffer body generally means less signal per newton.
| Specification | Typical range | What it determines |
|---|---|---|
| Force range | About 10 N to several kN | Largest load measured without saturation |
| Torque range | About 0.5 to several hundred N·m | Largest moment from offset contact |
| Resolution | Fractions of a newton or finer | Smallest detectable change |
| Sampling rate | About 100 Hz to 7 kHz | How fast the control loop can react |
| Overload capacity | Often 5 to 20 times rated range | Survival after a collision |
| Crosstalk | A few percent before calibration | Error leaking between axes |
Overload capacity is often the deciding factor in robotics, because collisions and dropped tools create brief peaks far above working loads. Choosing a sensor sized only for the nominal task force is a common reason for damaged sensors.
Force sensor compared with related robot sensors
A force sensor answers a different question than the other sensors on a robot. The comparison shows what each one measures and when it is the right choice.
| Sensor | Measures | Typical role |
|---|---|---|
| Force sensor | Force and torque at one interface | Contact control, weighing, collision detection |
| Tactile sensor | Pressure distribution over a surface | Slip detection and object shape in a grasp |
| Encoder | Joint or motor position | Position and velocity control |
| IMU | Acceleration and angular rate | Orientation and motion estimation |
| Depth camera | Distance to surfaces in a scene | Locating objects before contact |
Force and tactile sensing are complementary. A wrist force sensor reports the total load on the tool, while a tactile array reports where on a finger the contact sits. Vision and depth sensing guide the robot until contact, and force sensing takes over at the moment of contact.
How robots use force sensing
Force sensing enables force control, in which the robot regulates contact force instead of only position. The two main schemes are impedance control, which makes the robot behave like a programmable spring and damper, and hybrid force/position control, which controls force along some axes and position along others.
- Assembly and insertion: peg-in-hole and connector mating with clearances below 0.05 mm.
- Surface finishing: grinding, polishing and sanding at a constant contact force.
- Hand guiding: the operator pushes the tool and the robot follows.
- Payload identification: estimating mass and center of gravity of a gripped part.
- Collision detection: stopping when unexpected force exceeds a threshold.
Placement matters for these tasks. A wrist sensor sees only loads beyond the wrist, so it cannot detect a collision on the upper arm, while joint torque sensors cover the whole arm but with lower precision at the tool.
Errors and pitfalls in force measurement
Raw force readings contain effects that are not contact forces. The most important is the weight of the tool itself, whose apparent direction changes as the wrist rotates, so readings must be gravity-compensated using the tool mass and center of gravity.
| Problem | Cause | Remedy |
|---|---|---|
| Slow zero drift | Temperature change in gauges and electronics | Re-bias before each contact task |
| Reading changes with orientation | Uncompensated tool weight | Identify payload and subtract gravity |
| Oscillation on contact | Control gains too high for stiff surfaces | Lower gains or add damping |
| Noisy signal during motion | Inertial forces and vibration | Filter and compensate acceleration |
A frequent mistake is re-biasing the sensor while it is already in contact, which hides a real load and causes the robot to push harder than intended once the contact ends.
The short version
A force sensor lets a robot feel how hard it is pressing. It sits between the robot arm and its tool, and when something pushes on the tool, a metal part inside bends very slightly. Tiny electronic elements notice that bending and turn it into a number.
With that number, a robot can push gently instead of blindly. It can slide a part into a tight hole, polish a surface with steady pressure, or stop when it bumps into something unexpected, much as a person adjusts grip and pressure without looking.
Where it makes a difference
An electronics manufacturer automates the insertion of a stiff connector into a circuit board. Position control alone fails, because the connector and board tolerances add up to more than the 0.1 mm clearance available, and the robot either jams or bends pins.
The integrator mounts a six-axis force/torque sensor at the wrist and programs a search strategy: the robot presses down with 10 N, moves in a small spiral until the force drops as the connector aligns, then pushes in while limiting force to 40 N. If the force exceeds the limit, the robot backs out and retries. The jam rate falls sharply and bent pins become rare.