Power and force limiting in collaborative robot arms

By Tolerance Labs engineering team · Updated 18 October 2026

A collaborative arm that may touch a person must keep the force and pressure of that contact low enough to avoid injury. Power and force limiting is how most collaborative arms do it, and the arm’s mass, speed and joint sensing decide how well it works.

What power and force limiting is

ISO 10218-1 covers the safety of the robot itself, and ISO 10218-2 covers robot systems and their integration. Together they set the safety requirements for industrial robots and their applications, including applications where people and robots share a space. The 2025 editions of ISO 10218-1 and ISO 10218-2 take in most of what ISO/TS 15066 gave as separate guidance for collaborative applications, including the limits for contact with the human body. These documents have been revised over time, so work from the current edition of each.

The documents describe several ways a robot and a person can share a space, commonly grouped as a safety-rated monitored stop, hand guiding, speed and separation monitoring, and power and force limiting (PFL). In PFL, contact between the robot and a person can happen, by design or by accident, and the robot keeps the forces of that contact within limits. The other methods avoid contact or stop before it.

Transient and quasi-static contact

A transient contact is a brief impact in free space: the arm hits a hand or an arm that can move away. The body part recoils, and the contact lasts a fraction of a second. Force depends mainly on the robot’s speed and effective mass at the moment of impact.

A quasi-static contact is a clamping or crushing case: the body part is trapped between the robot and a fixed object, such as a table, a fixture or another link. It cannot move away, so force keeps rising while the robot keeps pushing. Force then depends on how quickly the robot detects the contact and stops or backs off, and on how much force the joints can make.

Quasi-static contact is usually the harder case. Look for pinch points between links, between the tool and a fixture, and near the base, and remove them by design where you can. Allowed values differ between transient and quasi-static contact and between body regions. Take them from the current documents, not from a datasheet or from this guide.

Effective mass

The mass that matters in an impact is the effective mass: the mass the arm presents at the contact point in the direction of motion. It depends on pose and direction. It includes the moving links, the tool and the payload, and the rotor inertia of each joint multiplied by the square of its gear ratio.

Rotor inertia often dominates in a high-ratio joint. Example: a rotor of 1 × 10⁻⁴ kg·m² behind a 100:1 reducer appears at the output as 1 × 10⁻⁴ × 100² = 1 kg·m². At a contact point 0.5 m from the joint axis, that inertia acts like a mass of J ÷ r² = 1 ÷ 0.5² = 4 kg, more than many cobot forearms. Low-ratio joints reflect far less; see backdrivability in robot joints.

The arm’s mass matrix at each pose gives the effective mass in each direction. Tool and payload mass add to it almost directly near the flange, so a heavier gripper can mean a lower allowed speed.

Worked example: speed and contact force

A simple model treats the body region as a spring of stiffness k, hit by the effective mass m at speed v. All the kinetic energy goes into the spring, so ½mv² = ½kx², and the peak force F = kx becomes F = v × √(m × k).

The values below are assumptions chosen to show the arithmetic. They are not limits from any standard. Take real body-region values from the current documents and from your risk assessment.

• Assumed: effective mass m = 10 kg, contact stiffness k = 20,000 N/m, speed v = 0.5 m/s.
• √(m × k) = √(10 × 20,000) = √200,000 ≈ 447 kg/s.
• Peak force: 0.5 × 447 ≈ 224 N. Deflection: 224 ÷ 20,000 ≈ 0.011 m, about 11 mm.
• At half the speed, 0.25 m/s: 0.25 × 447 ≈ 112 N. Force halves when speed halves.
• At 0.5 m/s with a quarter of the mass, 2.5 kg: √(2.5 × 20,000) ≈ 224 kg/s, and 0.5 × 224 = 112 N. To halve force through mass alone, you must cut the mass to a quarter.

Speed is the strongest lever: force scales with v, and energy with v². Soft padding lowers k, which lowers force with √k but adds deflection. The model ignores the time the robot needs to detect contact and stop. During that time the motors keep pushing, so real peak forces can exceed the estimate, and in a clamping case they keep rising. Validate on the real application with a measuring device made for collaborative robot force and pressure tests.

Detecting contact in the joints

• Joint torque sensors. A sensor at each joint output measures the torque the link carries. The controller compares it with the torque its model predicts for the motion and flags the difference. Sensitivity is high because reducer friction sits outside the measurement; see joint torque sensors.
• Motor current. The controller estimates external torque as the torque from motor current minus the modeled gravity, inertia and friction torques. No extra sensors are needed, but friction in high-ratio reducers changes with temperature, speed and load, so the threshold must sit above that uncertainty. Low-ratio, backdrivable joints make current-based detection much more sensitive.

Thresholds trade sensitivity against false stops. Fast moves and heavy payloads make model errors larger, so many controllers raise thresholds during fast motion or rely on a payload setting that matches the real tool. A wrong payload setting is a common cause of both false stops and missed contacts. Collision detection in robot joints covers detection methods and tuning in more detail.

After detection, the arm can stop, stop and back off a short distance, or switch to a compliant mode so a trapped person can push it away. Releasing matters most in clamping cases. A robot that stops while still pressing keeps the force on.

Reducing contact force by design

• Reduce speed in zones where contact is possible, and enforce that limit with safety-rated monitoring, not only the motion program.
• Reduce mass at the tool: light grippers, short adapters, and payloads held close to the flange.
• Round and pad edges and corners, so force spreads over a larger area and pressure falls.
• Remove pinch points between links, the tool, fixtures and walls, or keep the gaps wide enough for a hand.
• Limit joint torque where the task allows, so a clamping force cannot build as high.
• Guard the tool when it is sharp, hot or heavy. PFL limits the robot’s forces; it does nothing about a blade in the gripper.

The application is assessed, not the arm

A cobot arm can offer PFL functions, but no arm is safe or collaborative by itself. The integrator assesses the whole application: the robot, its tool and payload, the fixtures, the task, and the people nearby and how they work. A risk assessment identifies each possible contact and its body region, and validation measures the real forces and pressures. Change the tool, payload, speed or layout, and the assessment must be reviewed.

When you buy joints or an arm for a collaborative build, ask which safety functions the controller provides, how they are rated, and what documentation supports them. A torque sensor does not make a joint module or a part collaborative. Detection, the rating of the stop and the rest of the application decide the outcome. Safe torque off for robot joints covers how a joint stops safely.

Checklist

• Work from the current editions of ISO 10218-1 and ISO 10218-2 (the 2025 editions take in most of ISO/TS 15066), and a risk assessment of the whole application.
• List each possible contact, its body region, and whether it is transient or quasi-static.
• Remove pinch points and clamping cases by design first.
• Estimate effective mass at the tool, including reflected rotor inertia, and keep tool mass low.
• Set speeds from force estimates, then validate with measurements on the real application.
• Confirm how the arm detects contact, how its thresholds depend on the payload setting, and how it releases a clamp.
• Repeat the assessment when the tool, payload, speed or layout changes.

Browse robot arms and robot arm and cobot joints, and compare torque sensing and payload in Robot Parts.