Admittance vs Impedance Control: How to Choose for a Robot

Quick answer

This page is a system-level decision guide: compare command interfaces, sensing, and validation evidence. For a shorter first-pass tool, use the interactive admittance-versus-impedance selector; treat its output as a prompt for testing, not design approval. For broader foundations, see our guides to impedance control in robotics and force control in robotics.

The distinction is about the dynamic relationship a controller creates between motion and force, not a guarantee that one architecture is safer, faster, or better for a particular gearbox. Hogan’s foundational treatment emphasizes controlling dynamic interaction rather than position or force in isolation. Keemink and colleagues analyze how filtering, internal flexibility, motion-loop bandwidth, and virtual damping affect admittance performance and stability.¹²

1. What each controller commands

For a one-dimensional illustration, an impedance target can be written as F = M_d · (ẍ_d − ẍ) + B_d · (ẋ_d − ẋ) + K_d · (x_d − x). The controller uses motion error to calculate a desired force or torque. A physical robot still needs an actuator and inner loop capable of producing that command with sufficient accuracy.

An admittance target can be written as M_a · ẍ_r + B_a · ẋ_r + K_a · (x_r − x_0) = F_ext. The measured interaction force F_ext generates a motion reference x_r; an inner position or velocity controller tracks it. These simplified equations show the intended mapping, not complete robot dynamics or a stability proof. Sign conventions and coordinate frames depend on the implementation.

The diagram describes common controller implementations, not a requirement that impedance use no force sensing or that admittance use a particular sensor. Both can include feedback, observers, cascaded loops, and additional safety layers.

2. Compare the implementation, not just the names

These are screening questions rather than categorical properties. For example, admittance is commonly used with stiff, high-friction devices, while impedance implementations can also use force feedback. Keemink et al. describe trade-offs involving sensor location, drive-train flexibility, filtering, and inner-loop bandwidth; those factors are more informative than a single gear-ratio cutoff.² External force sensors are not mandatory in every design: Yao et al. report a sensorless force-estimation and admittance implementation for an industrial robot, though that result does not validate a different robot or task.³

3. Use a four-check selection process

Do not select a controller from nominal gear ratio alone. A high-ratio transmission can make transparent torque control harder, but the result depends on actual friction, compliance, sensing, compensation, and actuator control. Likewise, using an external force sensor does not by itself establish safe or stable admittance behavior.

If the decision is moving toward procurement, translate these requirements into the humanoid joint RFQ checklist for OEM buyers so suppliers can respond to the same interfaces and acceptance evidence.

4. Two practical screening examples

Backdrivable joint for leg or arm interaction

For a low-friction joint with a useful torque interface, impedance is a reasonable candidate when the task needs a shaped torque response to motion error. Before committing, measure torque tracking at low speed, reversal, expected load, and operating temperature. Include impact/contact traces for a leg; a favorable free-space test alone does not establish contact performance. A quasi-direct-drive or series-elastic design may help, but neither label substitutes for measurements. For a hardware example, see this series-elastic actuator design screening guide.

Stiff position-controlled joint for insertion or guided motion

For a stiff actuator with a position servo, admittance is a reasonable candidate if contact force can be measured or estimated reliably. Check sensor location and tool/payload compensation, then test filtering and the complete force-to-motion delay against representative contact stiffness. If hard-contact oscillation appears, tune and retest within the system’s validated stability envelope; simply increasing loop rate or reducing virtual mass is not a general remedy.²

These examples identify candidates for evaluation, not a universal mapping of robot limb to controller. A hybrid architecture can assign different controllers by joint or task, but mode changes need bumpless references, bounded commands, and contact-transition tests.

5. Validation plan before design freeze

Use the same payload, environment, approach speed, supply voltage, and thermal condition when comparing candidate controllers. Record raw signals with synchronized timestamps so that delay and saturation can be distinguished from controller tuning effects.

Set numerical acceptance limits from the application’s risk analysis and system requirements. The literature supports checking coupled stability, filtering, bandwidth, and mechanical flexibility; it does not establish a universal 1 kHz loop-rate threshold for every robot or task.²

Once these signals and limits are defined, request an engineering review by email with the joint interface and test conditions.

6. Risks and limits to keep visible

Neither approach removes the need for independent safety limits. A force-control law is not a safety-rated protective function unless the complete system has been designed and validated for that role.

7. Method, evidence, and limitations

This comparison uses the classical dynamic-interaction framing from Hogan (1985) and the admittance design and stability analysis from Keemink et al. (2018). The recommendations are engineering screening guidance, not a performance guarantee for a particular motor, gearbox, sensor, robot, or application. No universal gear-ratio threshold, force-sensor requirement, or controller bandwidth can be inferred from these sources alone.

The source literature is stable foundational work; the page’s review date is 2026-10. Product interfaces and current actuator specifications vary by supplier and must be checked against the exact hardware revision and test conditions.

8. Frequently asked questions

What is the main difference between impedance and admittance control?

In a common implementation, impedance control maps motion error to a force or torque command. Admittance control maps measured interaction force to a motion reference that an inner position or velocity loop tracks. Both shape dynamic behavior during contact.

Does admittance control always need an external force-torque sensor?

Not always. A wrist or joint sensor is common, but torque sensing or a model-based observer may provide the force estimate in some designs. Evaluate accuracy at the contact point, payload compensation, friction sensitivity, noise, drift, and delay. The sensor type should follow the task and validation results.

Does a high gear ratio mean I must use admittance control?

Potentially. A high ratio may increase friction and make motor-current torque estimates less representative of joint torque. A joint torque sensor, series elasticity, compensation, or a suitable force-control loop may change the result. Require measured torque tracking and contact stability evidence instead of accepting a gear-ratio cutoff.

Is admittance control inherently slower or unstable on hard contact?

No universal ranking applies. Sensor filtering, computation, communication, actuator response, contact stiffness, and controller tuning affect phase and stability. Measure end-to-end delay and test the intended contact conditions on the actual system.

Can a robot use impedance and admittance control together?

Yes. A robot can use different interaction controllers across joints or tasks, or switch modes. The transition logic, force limits, state continuity, and coupled stability need their own validation, including contact-transition tests.

What should a supplier provide before I choose a control architecture?

Request the command and feedback interfaces, torque/current calibration, backdrive or friction data across load and temperature, sensor timing, inner-loop response, saturation limits, and contact-test traces under conditions close to your application.

Next steps

• Controls teams: document the requested interaction behavior, sensed signals, complete loop delay, force/torque limits, and acceptance criteria.
• Mechanical and actuator teams: provide measured friction/backdrive behavior, compliance, torque calibration, and performance across load and temperature.
• OEM procurement teams: put the command and feedback interfaces, raw validation traces, fault behavior, and test conditions into the RFQ; do not specify “impedance-ready” or “admittance-ready” without defining evidence.

For an architecture discussion, email [email protected], use WhatsApp, or share your requirements through the contact page.

References

Footnotes

1. Neville Hogan, “Impedance Control: An Approach to Manipulation: Part I—Theory”, Journal of Dynamic Systems, Measurement, and Control, 1985. Foundational treatment of robot-environment dynamic interaction. ↩

2. Arvid Q. L. Keemink, Herman van der Kooij, and Arno H. A. Stienen, “Admittance control for physical human–robot interaction”, The International Journal of Robotics Research, 2018. Reviews admittance architecture and analyzes filtering, flexibility, motion-loop bandwidth, damping, and stability. ↩ ↩² ↩³ ↩⁴

3. Yao Bitao et al., “Sensorless and adaptive admittance control of industrial robot in physical human–robot interaction”, Robotics and Computer-Integrated Manufacturing, 2018. Experimental research on model-based external-force estimation and admittance; application-specific evidence. ↩