Compliance and Safe Physical Interaction
A traditional industrial robot arm is a powerful, precise, and completely indiscriminate machine. Put your hand in the wrong place and it will break your bones without hesitation or awareness. That is fine when the robot operates inside a safety cage inaccessible to humans, but it is incompatible with a world where robots work side-by-side with people. The shift from isolated industrial robots to collaborative robots (cobots) that share space and tasks with humans is one of the most consequential transitions in modern robotics. Making it safe requires rethinking control from the ground up.
Why Stiff Position Control Is Dangerous Near People
A position controller's goal is to reach and maintain the commanded position regardless of external forces. This rigidity is a feature in precision manufacturing: the robot holds position against workpiece resistance, vibration, and load changes with unwavering accuracy. But rigidity becomes lethality in human environments. If a stiff position-controlled robot arm collides with a person, the controller's response is to push harder — to overcome the 'disturbance' (the human) and reach the commanded position. The force applied can be enormous: industrial robot motors routinely produce peak torques of hundreds of Newton-meters. The human loses this contest. Three physical quantities determine the injury risk in a robot-human collision: the contact force, the contact duration, and the contact area. High force over a small area over a short duration can cause injury even from a robot that quickly backs off. Safety standards like ISO/TS 15066 specify maximum permissible contact forces and pressures for different body parts — for example, the maximum allowable clamping force on a hand is around 140 N, and even lower (65 N) for the head and neck. For a robot to operate safely near people, it must do one or more of the following: move slowly enough that kinetic energy at impact is low; detect contact before force builds; yield to contact forces rather than overpower them; or be designed with inherently compliant (soft) mechanisms that limit peak forces physically.
The Universal Robots UR10 and KUKA LBR iiwa are canonical collaborative robots. Both feature torque sensors in every joint, rounded external surfaces without pinch points, joint torque limits enforced in software and hardware, and power-and-force limiting (PFL) control modes that continuously monitor contact forces and halt motion when limits are exceeded. Safety is a system property — it requires the controller, the hardware, the sensors, and the software architecture working together.
Impedance control, introduced in Lesson 3 in the context of dynamics, is the primary control framework for safe physical interaction. Instead of commanding a fixed position, the controller shapes the robot's mechanical impedance — how it responds to external forces. The impedance relation is: F_contact = Md * (x_ddot - x_ddot_d) + Bd * (x_dot - x_dot_d) + Kd * (x - x_d) where: F_contact is the external contact force x is the current end-effector position, x_d is the desired position Md is the desired inertia (virtual mass) Bd is the desired damping coefficient Kd is the desired stiffness By choosing low Kd (soft spring), the robot yields easily when contacted — contact forces stay low even with significant position deviation. By choosing high Bd (strong damping), oscillations after contact are suppressed. The robot behaves like a physical spring-mass-damper with programmable parameters. The power of impedance control is that the impedance parameters can be task-dependent and even direction-dependent. A robot polishing a surface might be stiff perpendicular to the surface (to maintain contact) and compliant along the surface (to follow its contour). An assembly robot inserting a peg might be stiff axially (to push the peg in) and compliant laterally (to accommodate small alignment errors without jamming — this is called remote center compliance).
Admittance control is the dual of impedance control, and the distinction matters for implementation. Impedance control is implemented in torque-controlled joints: the robot measures contact force and outputs torque. Admittance control is implemented in position-controlled joints: the robot measures contact force and outputs a modified position setpoint. Most commercial cobots (which have position-controlled joint drives as their low-level interface) use admittance control. The admittance law is: M * x_ddot_cmd = F_measured - B * x_dot - K * (x - x_ref) The measured external force drives a virtual second-order system, and the resulting position is sent as the reference to the underlying position controller. The human pushes the robot; the admittance controller computes how much the robot's reference position should shift in response, and the position controller tracks that shifted reference. From the outside, the robot feels soft and yielding. Collision detection is the final piece. Even a compliant controller needs to distinguish between intentional contact (a human guiding the arm) and dangerous unintended collision. Collision detection algorithms compare measured joint torques to torques predicted by the dynamics model. If the difference (the external torque residual) exceeds a threshold, a collision is detected and the robot stops or retracts. Reaction time matters enormously: ISO/TS 15066 requires reaction times below 10 ms for the highest safety categories.
Match each safety and compliance concept to its defining description.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
A cobot is in admittance control mode. A worker accidentally bumps the arm sideways. The robot's torque sensors read a 25 N lateral force. In admittance control, what does the controller do with this measurement?
An engineer is programming a cobot to insert cylindrical pins into holes. The pin-hole clearance is 0.1 mm, meaning precise lateral alignment is critical. The engineer wants the robot to push the pin in firmly but automatically accommodate small alignment errors without jamming. Which impedance configuration achieves this?
Design a Safe Human-Robot Interaction Scenario
- You are a safety engineer designing a collaborative robot workcell for a new application. Choose one of the following:
- Option A: A cobot that assists a surgeon by holding a camera steady during minimally invasive surgery, with the surgeon occasionally repositioning it by hand.
- Option B: A cobot that works alongside a factory worker assembling electronics, handing the worker parts and tools, with the worker's hand sometimes in the robot's workspace.
- Option C: A cobot in a physical therapy clinic that supports a patient's arm during rehabilitation exercises, responding to the patient's voluntary movements while maintaining a prescribed motion range.
- For your chosen scenario, answer the following:
- 1. Sensing: What sensors does the robot need to monitor forces, detect unexpected contact, and localize the human? List at least three.
- 2. Control mode: Should the robot use impedance or admittance control in this application? Justify your choice based on the joint drive hardware and the nature of human contact expected.
- 3. Impedance parameters: Should the desired stiffness Kd be high or low? Should damping Bd be high or low? Reason from the task physics.
- 4. Failure mode analysis: Describe two specific scenarios where even a well-designed compliant controller could fail to prevent injury. How would you engineer safeguards against each?
- 5. Regulatory: ISO/TS 15066 limits clamping forces. How would you verify in testing that your system meets these limits before deploying it near real humans?