Skip to main content

Series Elastic Actuators — Compliance by Design

Duration: 45 min · Level: Intermediate · Module: 2. Actuator Architecture · Focus: SEA, compliance, force-control, safety

In the last lesson you chose how much gearing to use. This one adds a deliberate spring to the drivetrain — and that single mechanical decision is what lets a robot control force as precisely as it controls position. If your robot will ever press, hold, or touch a person, you need to understand the Series Elastic Actuator (SEA), because it is the dominant choice for exactly those joints.

The idea: measure a spring, know the force​

A SEA, introduced by Gill Pratt and Matt Williamson at MIT in 1995, places a physical spring in series between the gearbox output and the joint. That sounds like it would make the joint sloppy. In fact it gives you a precise force sensor for free, because of one equation you should commit to memory:

force = spring displacement × spring stiffness

Put an encoder on each side of the spring, measure how much it deflects, multiply by the known stiffness, and you know the output force directly — no expensive load cell required. You then close a control loop on that force. The spring also does mechanical good on its own: it absorbs impact energy and makes every contact gentler, because the robot physically gives before the controller even reacts.

The trade-off you are buying: bandwidth for safety​

Compliance is not free, and the cost is force-control bandwidth. A softer spring yields more compliance and safer contact but responds to force commands more slowly; a typical SEA tops out around 50 Hz of force-control bandwidth. A stiffer spring is faster but less forgiving. So the spring stiffness is the design parameter — it sets where you sit on the safety-versus-speed curve.

This is why you don't put SEAs everywhere. The practical mapping by joint:

  • Ankles and knees → QDD. Locomotion needs stiffness and fast response; a soft spring there hurts you.
  • Wrists and shoulders → SEA. Manipulation near people needs compliance and clean force control; the spring is exactly what you want.

That single sentence — stiff legs, compliant arms — will guide most of your actuator layout.

Who builds them this way, and why​

The pattern is not theoretical. NASA Valkyrie, MIT HERMES, and Apptronik Apollo all use SEAs, chosen specifically for human-safe operation. Apollo (2023) is the cleanest reference: all 28 joints use in-house SEAs rated at 300 Nm peak, designed from the start for ISO/TS 15066 collaborative operation near humans (the safety standard you'll study in Module 8). The lesson to take from the Apollo actuator work (Apptronik, ICRA 2023) and Pratt and Williamson's original paper is that force-controllable contact is an actuator-level decision — you cannot fully recover it in software if the hardware is rigid.

The frontier option: variable stiffness​

If you want both worlds, the emerging answer is the variable stiffness actuator (VSA), which can shift between a compliant SEA-like mode and a near-rigid mode on demand — soft while handing a patient a cup, stiff while carrying a load. VSAs are mechanically complex and heavier, so treat them as an advanced option for a specific high-value joint, not a default. For most builds in 2026, a well-chosen fixed spring plus good force control gets you most of the benefit.

Putting it into practice​

Design and reason about one SEA joint.

  1. Pick a shoulder joint that must support a 5 kg arm raise (~80 Nm). Choose a spring stiffness and compute the spring deflection at that torque from force = displacement × stiffness. Is the deflection large enough to measure cleanly with your encoders?
  2. Estimate the force-control bandwidth your stiffness gives you (softer → lower). Decide whether ~50 Hz is fast enough for the manipulation tasks you care about.
  3. Now justify the architecture: write one sentence each for why this shoulder is an SEA and why the knee on the same robot is a QDD.
  4. Optional stretch: sketch where, if anywhere, a VSA would earn its added mass on your robot — and where it would not.

Key takeaways​

  • A SEA puts a spring between gearbox and joint; measuring its deflection gives precise force from force = displacement × stiffness, plus mechanical impact absorption.
  • The cost is bandwidth: softer springs are safer but slower (~50 Hz typical), so spring stiffness is the key safety-vs-speed design knob.
  • Map by joint: QDD for stiff, fast legs (ankles, knees); SEA for compliant arms near people (wrists, shoulders).
  • SEAs are the human-safe standard — NASA Valkyrie, MIT HERMES, and Apptronik Apollo (28 in-house SEAs, 300 Nm, ISO/TS 15066-ready) all rely on them.
  • Variable stiffness actuators offer both modes at the cost of complexity and mass — an advanced, per-joint option rather than a default.

References​

  • Series Elastic Actuators — Pratt and Williamson (1995). IROS 1995
  • Actuator Design for the Apollo Humanoid Robot — Apptronik Team (2023). ICRA 2023

← Previous: 2.1 The Actuator Revolution: From Hydraulics to QDD Electric · Next: 2.3 Joint-Level Architecture for a Full Humanoid →

Part of Module 2: Actuator Architecture.