Robot Force Control: From Contact Detection to Stable Process

Position control works well when the workpiece, fixture, tool, and robot remain where the program expects them to be. Grinding, polishing, deburring, insertion, surface following, and product testing violate that assumption: contact force changes with part tolerance, fixture error, abrasive wear, thermal growth, and local surface geometry.

Force control adds feedback from the interaction itself. The controller can detect contact, permit compliant motion, regulate pressure, or adapt speed when force departs from a target. That makes a process tolerant of limited variation, but it does not eliminate the need for sound mechanics, calibrated payload data, a stable tool, or a safety-rated risk reduction system.

A wrist sensor measures process contact close to the tool, while also seeing payload, acceleration, cable, and inertial effects.

Choose the feedback source around the process

Some robots estimate external torque from joint-motor current and a dynamic model. This can support broad collision detection and compliance without adding hardware at the wrist. Accuracy depends on drivetrain friction, gravity compensation, payload mass, center of gravity, robot posture, temperature, and acceleration. Small contact forces can be difficult to separate from the robot’s own dynamics.

An external six-axis force-torque sensor measures forces and moments nearer the process. It can provide better sensitivity and direct information about tool contact, but it adds mass and stack height, changes payload and inertia, requires wiring and signal integration, and has finite overload capacity. The sensor must be sized for accidental loads as well as the normal process range; a device selected only for fine resolution may be damaged during a crash or tool change.

A third option is mechanical compliance, such as a floating tool, constant-force actuator, or compliant wrist. Passive compliance can be robust and fast, while active force feedback provides programmability and diagnostic data. Many successful cells combine them: mechanical compliance absorbs high-frequency variation, and the robot loop corrects slower changes.

Match the control mode to the task

Contact detection watches for a force or torque threshold, then changes state or stops an approach. It is useful for finding a surface or confirming seating, but the threshold must account for noise, tool weight, approach speed, and inertial load. A threshold that works during a slow test may trigger too late during production-speed contact.

Compliance allows motion under load, often along selected axes while constraining the others. Constant-force control regulates a target normal force as the robot follows a surface. Search routines combine limited force with spiral, raster, or directional motion to locate an insertion feature. Speed adaptation can slow a machining path when the measured load rises, reducing overload while maintaining progress.

Each mode needs boundaries: maximum displacement, maximum force and torque, permitted axes, velocity limits, timeout, escape direction, and behavior after sensor loss. Without those limits, a force loop may continue pushing because the commanded value was never reached.

Stable contact depends on robot dynamics, tool stiffness, process speed, surface variation, filtering, and loop tuning.

Control-loop bandwidth meets mechanical compliance

Contact changes the dynamics of the system. A stiff tool against a stiff part can create a high natural frequency and amplify small delays or measurement noise. Excessive gain may cause chatter; heavy filtering reduces noise but adds phase delay and can destabilize or slow the response. A soft interface is easier to control but may reduce path accuracy or process consistency.

Commissioning should begin at low speed and conservative force, with verified tool mass and center of gravity. Bias the sensor in the correct pose and temperature condition, then evaluate multiple robot postures because gravity compensation and structural stiffness change across the workspace. Trend actual force, force target, position correction, speed, and saturation flags. A smooth average can conceal damaging peaks, so capture data fast enough for the contact event.

Separate process regulation from machine safety

A force threshold used to control quality is not automatically a safety function. Personnel hazards can remain from sharp tools, hot parts, stored pneumatic energy, crushing points, and robot or payload momentum. Collaborative operation likewise depends on an application-specific risk assessment, validated limits, and safety-rated functions—not merely the presence of torque sensing.

The machine design should define what happens during emergency stop, protective stop, sensor failure, network interruption, tool breakage, and loss of the workpiece. If stored force can release motion after power removal, the recovery procedure must address it. Restart should require the cell to establish a known tool, payload, sensor status, and contact-free state before force control resumes.

Give the PLC a clear interface

The high-rate force loop belongs in the robot or dedicated motion controller. The PLC should exchange machine-level states such as ready, approach permitted, contact found, process active, process complete, quality result, and fault. It should also receive diagnostic values needed for production records and maintenance, without attempting to close the force loop through a slower industrial network.

Define ownership of recipes and limits. If the PLC selects a part recipe, the robot should confirm the loaded force, speed, tool, and search parameters before motion. Changes to thresholds and filters need revision control because they affect both quality and equipment loading.

Use force data as process evidence

Force traces can reveal more than pass or fail. Increasing insertion peak may indicate misalignment, burrs, lubrication loss, or fixture drift. A rising average polishing force at the same command can indicate tool wear or compliance change. A sudden shift in a six-axis signature can identify a loose tool or incorrect part orientation.

Statistical limits should come from known-good process data and engineering tolerances, not arbitrary percentages. Retain the recipe, tool identity, robot position, and relevant force features with the result. That context helps distinguish product variation from sensor bias or maintenance work.

Motion-system components can be reviewed under Drives & Motion Control, with robot-specific hardware in the ABB Robotics collection.

Engineering perspective

Force control is most valuable when the fixture and process are already repeatable enough to define a stable contact task, yet still contain variation that pure position control cannot absorb economically. It should not become a software patch for loose fixturing or uncontrolled tooling. A strong implementation combines appropriate sensing, bounded control modes, mechanical compliance, safety separation, diagnostic trends, and tests of credible failure states.