Series Elastic Actuator: Design & Control

• Series Elastic Actuators are systems that integrate a purposely compliant spring between the motor and load to enable precise force sensing and controlled interaction.
• Optimized spring geometry and material selection, validated with finite element analysis, provide modular stiffness and efficient energy transfer in various applications.
• Advanced control strategies, including multi-loop PID and neural network controllers, enhance SEA performance by balancing compliance with high bandwidth in dynamic environments.

A Series Elastic Actuator (SEA) is an actuation architecture characterized by the intentional placement of a compliant (elastic) element—typically a spring—in series between the power source (such as an electric motor) and the load. This configuration fundamentally modifies the actuator’s mechanical impedance, enables high-fidelity force sensing, and allows control strategies that improve safety, robustness, and performance in physical human–robot interaction, prosthetics, legged locomotion, and other environments requiring compliant interaction.

1. Mechanical Architecture and Spring Design

The SEA’s defining mechanical feature is a spring element through which all load force is transmitted. The geometry, material, and placement of this spring are critical. In high-performance SEAs, such as those designed for powered ankle prostheses, the elastic element may take the form of a planar torsional spring constructed from maraging steel and optimized for high yield strength and endurance. Relevant geometric parameters include outer radius ($R_\mathrm{out}$), inner thickness ($T_\mathrm{in}$), branch length ($L$), and branch width ($d$), with total actuator stiffness increased by stacking multiple such springs:

$k_\mathrm{total} = n \cdot k_\mathrm{single}$

where $n$ is the number of stacked spring elements, each characterized and optimized via finite element analysis and experimental validation. Key spring metrics from advanced designs include measured total linear stiffness up to $$7.5\,\mathrm{Nm/^\circ} \approx 430\,\mathrm{Nm/rad}$$, and endurance demonstrated by maximum von Mises stress below 50% of yield under nominal load cycles (Li et al., 2023).

The force or torque delivered by the SEA is computed directly from measured spring deflection, exploiting Hooke’s law:

$$\tau = k_\mathrm{total} \cdot (\theta_\mathrm{motor} - \theta_\mathrm{load})$$

This approach ensures high-resolution, low-noise force estimation, essential for accurate force/impedance control (0912.3956).

2. Dynamic Modeling and Compliance Characteristics

SEA-equipped systems are modeled as cascaded inertial bodies (motor and load) coupled via the series spring (and optionally, parallel damping):

$$\begin{align*} J_m\,\ddot\theta_m + b_m\,\dot\theta_m + k_\mathrm{total}\,(\theta_m - \theta_l) &= \tau_m \ J_l\,\ddot\theta_l + b_l\,\dot\theta_l - k_\mathrm{total}\,(\theta_m - \theta_l) &= \tau_\mathrm{ext} \end{align*}$$

where $J_m, J_l$ and $b_m, b_l$ denote motor/load inertias and viscous damping, respectively; $\tau_m$ is the actuator torque, and $\tau_\mathrm{ext}$ covers external/environmental torque. The series compliance $1/k_\mathrm{total}$ achieves several functional goals:

• Inertia Decoupling: The compliance reduces reflected inertia at the interface, increasing backdrivability.
• Energy Storage: Elastic deformation enables storage and subsequent release of mechanical work, particularly crucial in applications such as prosthetic gait during stance and push-off.
• Signal Filtering: The spring attenuates high-frequency disturbances and decouples motor transients from the human–machine interface.

Experimental and simulation studies confirm SEA implementations delivering continuous torque above 80 Nm, peak torques above 200 Nm, closed-loop control bandwidths exceeding 10 Hz, and deflection/rotation ranges supporting biological joint motion (Li et al., 2023).

3. Control Architectures and Integration

SEAs enable advanced force and impedance control, leveraging direct force measurement via spring deflection. Architectures typically employ multi-layered control loops:

• High-level reference generator: In prosthetic and gait applications, a neural network or MLP (Multi-Layer Perceptron) may predict desired joint state trajectories (angle, velocity, phase, speed) from inertial data (Li et al., 2023).
• Intermediate impedance/force loops: Reference torques or positions are generated to match the virtual impedance model or prescribed trajectory.
• Low-level motor control: A PID (or cascaded PID) controller regulates actuator position/torque, commanding the motor to track the reference while implicitly or explicitly compensating for anticipated spring deflection:

$$\theta_\mathrm{motor,\,ref} = a_\mathrm{ref}(t) + \Delta_\textrm{offset}$$

• Feedback and feedforward tuning: Measured motor position and spring deflection close the feedback loop, directly implementing torque–angle laws.

In prosthetic SEAs, the control update rate is matched to sensor acquisition (e.g., 100 Hz), with built-in compliance ensuring phase-stable assistance and robustness to unmodeled shocks (Li et al., 2023).

4. Performance Metrics, Bandwidth, and Tunability

Performance of SEAs is tightly coupled to both hardware design and control policy. Notable metrics include:

Characteristic	Value / Description
Peak stiffness	$$k_{\mathrm{total}} = 7.5\,\mathrm{Nm/^\circ} \approx 430\,\mathrm{Nm/rad}$$
Spring stack limit	Up to $20\,\mathrm{Nm}$ per stack, $\sim 10^\circ$ deflection
Torque rating	Continuous $>80\,\mathrm{Nm}$, peak $>200\,\mathrm{Nm}$
Control bandwidth	$>$10 Hz closed-loop (100 Hz IMU/MLP sampling)
Mobility design	$-50^\circ$ (plantarflexion) to $+20^\circ$ (dorsiflexion)
Endurance	$\sigma_\mathrm{max}=972\,\mathrm{MPa}$ under $20\,\mathrm{Nm}$ load ($<50\%$ yield)

Adjustability is achieved through:

• Spring geometry/material optimization: $R_\mathrm{out}$, $T_\mathrm{in}$, $L$, and $d$ control $k_\mathrm{total}$, guided by systematic correlation studies.
• Stack count modification: Number of planar springs ($n$) sets actuator stiffness in modular form.
• Cartridge swapping: Modular cartridges allow rapid mechanical adaptation for subject-specific needs (Li et al., 2023).

5. Advanced Variations and Adaptation

SEAs are not limited to single-joint or prosthetic applications. Implementations span:

• Cable-driven SEAs: Flexible cable transmissions with series compliance enable remote actuation and high compliance (Zou et al., 2016, Zou et al., 2017).
• Fluid-based SEAs: Soft robotic grippers integrate rolling-diaphragm actuators and compliant hoses as the elastic element, enabling both high Z-width impedance range and intrinsic force estimation from internal pressures (Wang et al., 2020).
• Continuous-compliance SEAs: Modern research explores compliant elements distributed across structural components, such as curved steel plate shanks in bipedal robots, morphing the “spring” into an infinite-dimensional elastic continuum (Bendfeld et al., 2024).
• Application-specific tuning: SEAs are central to devices that implement biologically-inspired compliance in exoskeletons, legged robots, and haptic devices, leveraging their shock tolerance, transparency, and force-fidelity for a range of dynamic and unstructured tasks (0912.3956, Koda et al., 2024).

6. Key Implementation Trade-Offs and Guidelines

Engineering an effective SEA requires balancing several trade-offs:

• Compliance vs. Bandwidth: Softer springs provide greater shock absorption and safety but reduce achievable force/torque bandwidth. Stiffer springs raise bandwidth at the cost of compliance, requiring careful selection according to application task envelopes.
• Motor-sizing and transmission: Motor and gear selection must account for both static torque requirements and dynamic, compliance-filtered loads to avoid under- or over-sizing, with guidelines often leveraging “maximum torque transmissibility” analyses (Lee et al., 2019).
• Controller structure: Multi-loop control (velocity, torque, impedance) enables independent tuning of tracking and robustness. Use of disturbance observers, model-based feedforward, and structured $H_\infty$ synthesis enhances performance and passivity guarantees (Rampeltshammer et al., 2022, Yu et al., 2019).
• Safety and endurance: Fatigue performance, modularity for rapid reconfiguration, and sensor selection (e.g., high-resolution encoders) are critical to ensure long-term safe operation and precise force feedback.

In state-of-the-art designs, modular architectures, fully characterized by explicit geometric and material parameters, enable a high degree of adaptation, analytic modeling, and robust control for human-centric, high-performance robotics (Li et al., 2023).

7. Application Case Study: Powered Ankle Prosthesis with Planar Torsional SEA

A representative cutting-edge system employs a planar, stackable torsional spring formed from maraging steel. The actuator exhibits:

• Linear torque–angle relation within the working range,
• High torque density and endurance ($\sigma_\text{max} <50\%$ of yield under max load),
• Stackable/modular stiffness (user adjustable),
• Integration with an MLP-based gait recognizer for real-time, phase-synchronous reference generation,
• Biologically realistic torque/power output and energy return enabling locomotion up to $4\,\mathrm{m/s}$,
• Built-in compliance that passively stabilizes the foot–ground interface and mitigates load transients,

demonstrating that advanced SEA design and control can produce actuators functionally and dynamically comparable to or exceeding biological limbs (Li et al., 2023).