Active 2-DoF Ankle Control

Updated 19 July 2025

Active 2-DoF ankle control is a system that independently manages plantarflexion/dorsiflexion and inversion/eversion to emulate natural ankle biomechanics.
It employs real-time sensor feedback and adaptive algorithms to minimize gait tracking errors and improve user stability.
This approach enhances mobility by reducing muscle effort and optimizing assistive device performance in exoskeletons, orthoses, and humanoid robots.

Active 2-DoF ankle control refers to the intentional, real-time manipulation of two independent anatomical axes of the ankle joint—typically plantarflexion/dorsiflexion (sagittal plane) and inversion/eversion (frontal plane)—via actuators, sensors, and control algorithms in wearable robots (such as exoskeletons, orthoses, and prostheses) or humanoid robots. This approach aims to replicate or augment the complex biomechanical roles of the natural ankle, providing enhanced stability, adaptation to terrain, energy efficiency, and user-specific gait modulation, especially for paraplegic, amputee, or neurologically impaired populations.

1. Underlying Principles and Rationale

Active 2-DoF ankle control is motivated by the biomechanical and functional significance of the ankle joint in natural locomotion. The human ankle possesses up to three degrees of rotational freedom, with two principal axes of movement:

Sagittal (plantarflexion/dorsiflexion; PF/DF)
Frontal (inversion/eversion; INEV)

Most traditional wearable robots or assistive devices employ either purely passive mechanisms or limit active control to a single DoF (usually PF/DF). However, limiting the controlled DoFs constrains the user's natural motion and impairs postural stability, particularly during dynamic or perturbed walking.

Recent findings show that the inclusion of a second actively controlled DoF— permitting both PF/DF and INEV—markedly improves kinematic compatibility, reduces misalignment-induced errors (quantified, for example, by reduced cuff rotation), and brings user stability and trajectory tracking significantly closer to unassisted gait. The transition from 1 DoF to 2 DoF produces a larger improvement in stability and tracking error than the subsequent transition from 2 DoF to 3 DoF, suggesting that a two-DoF design captures most of the essential functional features required for effective gait assistance (Dezman et al., 10 Jun 2024).

2. Control Strategies and Real-Time Algorithms

Control of 2-DoF active ankles spans a range of architectures and methodologies, all requiring real-time estimation of user state, terrain, and interaction forces:

Direct Kinematic and Force-Based Control

Implementation often splits control into swing-phase and stance-phase directives. During swing, inverse kinematics ensures that the foot is parallel to the ground (flat-foot landing), while during stance, constraints on the center of pressure (COP) are enforced to keep the foot rigidly in contact with the ground. For example, the mapping from ankle joint angles to 3D foot orientation is computed by forward kinematics, and joint targets are solved numerically (e.g., Newton–Raphson algorithm) (Gurriet et al., 2019).
Rigid contact constraints are imposed by saturating joint commands according to ground reaction force measurements, ensuring COP remains within safe limits.

Closed-Loop Stabilization and IMU Feedback

Feedback from inertial measurement units (IMUs), specifically pelvis or foot orientation, is used to adjust ankle commands and maintain desired pelvis trajectory—even compensating for small disturbances or terrain irregularities (Gurriet et al., 2019).

Bioinspired and EMG-Driven Controllers

Control strategies may emulate neuromechanical or musculoskeletal models. This includes controllers using winding filament muscle models, which infer human muscle forces from joint kinematics and muscle lengths, or Hill-type models driven by real-time EMG signals from antagonistic muscle pairs (e.g., tibialis anterior and gastrocnemius), with activation dynamics formalized via difference equations (Bishe et al., 2020, Shah et al., 2022).
In patient populations lacking reliable EMG, bioinspired controllers using generalized activation curves and muscle length as inputs allow universal applicability without invasive or unreliable sensing (Bishe et al., 2020).

Learning-Based and Adaptive Control

Model-based adaptive controllers (such as disturbance observers or extended Kalman filter-based estimators) track gait phase, stride length, and ground incline and adapt assistive torques biomimetically in real time to match biological needs (Medrano et al., 2022).
Neural network or Koopman-operator-based predictive schemes are trained to map proximal joint kinematics or sensor data to desired ankle angles and moments, supporting context-aware control and automatic adaptation across multiple activities (e.g., stairs, slopes) without explicit mode classification (Dey et al., 2021, Singh et al., 10 Jan 2025).

Human-Robot Interaction (HRI) Feedback

Real-time measurement of human-robot interaction torques is used as a feedback channel to detect mismatches, synchronize device and user trajectories, and minimize user effort—especially important in load-carrying tasks or when the device’s trajectory lags user intent (Almeida et al., 14 Apr 2025).

3. Mechanical Implementations and Actuation Schemes

The mechanical design of 2-DoF active ankles must ensure precise, low-latency actuation along at least two anatomical axes, while minimizing distal mass and misalignment:

Series Elastic Actuators (SEA): Enable compliant torque control and can independently modulate PF/DF and INEV (Lora-Millan et al., 2023).
Cable-Driven and Pneumatic Designs: Reduce distal mass by remote actuation.
Parallel Mechanisms: Systems such as the 3-RRR spherical parallel mechanism provide inherent multi-axis motion naturally aligned with human biomechanics (Zhang et al., 19 May 2025).
Electro-Hydraulic Actuation: Integrates compact, energy-regenerative circuits to match human joint power in a lightweight form—requiring dynamic models that accommodate rapid cyclic pressurization, energy storage, and release synchronized with gait (Wei et al., 2023).

Innovative mechanical integration includes topology optimization and additive manufacturing of fluid channels or skeletal elements to balance structural strength against weight constraints, with some designs achieving up to 55% mass reduction while maintaining output and kinematic accuracy (Wei et al., 2023).

4. Experimental Validation and Performance Metrics

Empirical evaluations of active 2-DoF ankles focus on their impact on:

Gait Kinematics: Range of motion (RoM), trajectory tracking error (frequently quantified as RMSE against natural walking), and trunk acceleration-based stability measures. As shown in (Dezman et al., 10 Jun 2024), increasing controlled DoFs from 1 to 2 provides a substantial reduction in RMSE and improved stability, with diminishing returns beyond 2 DoF.
Muscle Activity: EMG measurements (e.g., gastrocnemius medialis, tibialis anterior) before and after device use show that active assistance significantly reduces muscle activations during relevant gait phases (up to 40% reduction in GM for terminal stance with exoskeleton assistance) (Almeida et al., 14 Apr 2025).
Biological Torque and Power Reduction: Estimation of joint torque reductions under active assistance, with neuromechanical and disturbance observer-based approaches yielding mean reductions of 20–24% relative to non-assisted conditions, and up to 40% in nonstandard tasks (e.g., moonwalking) (Durandau et al., 2021).
User-Device Synchronization: HRI torque profiles and lag/synchronization measures indicate improved comfort and naturalness with active, adaptive controllers; excessive unidirectional or large-magnitude HRI torques flag the need for further closed-loop adaptation (Almeida et al., 14 Apr 2025).

Robustness to “unseen” conditions—such as changing speed, incline, or load—is a key performance consideration, with model- and data-driven controllers showing strong potential for generalization beyond training scenarios (Medrano et al., 2022, Durandau et al., 2021).

5. Limitations and Emerging Challenges

Some persistent limitations and challenges in active 2-DoF ankle control include:

Soft-Tissue Compliance and Kinematic Misalignment: Even with added DoFs, exoskeletons may not fully eliminate kinematic mismatch due to the compliance and deformation of human tissue, as indicated by residual cuff rotation and measured angle discrepancies (Dezman et al., 10 Jun 2024). Sensor fusion and adaptive compensation algorithms are needed.
Complexity and Energy Consumption: Multi-DoF actuation increases device mass, control and sensing complexity, and energy demands. Lightweight structural integration and energy recovery/optimization are essential for long-duration use (Wei et al., 2023).
Control Personalization and Adaptation: Significant inter-user differences in kinematics and muscle activation (especially in neurologically impaired populations) require real-time model adaptation, possibly via machine learning, EMG-based personalization, or real-time impedance tuning (Durandau et al., 2021, Singh et al., 10 Jan 2025).
Incomplete Validation in Clinical Populations: Many systems are validated primarily on healthy individuals, with limited evidence from patient or out-of-lab studies. Broader clinical and ecological validation is needed (Lora-Millan et al., 2023).

6. Representative Applications and Future Directions

Active 2-DoF ankle control has enabled several practical applications:

Fully Actuated Exoskeletons and Crutch-Less Gait: As demonstrated in the ATALANTE exoskeleton, robust sagittal plane stabilization and improved pelvis tracking were achieved through independent control of PF/DF and INEV, with significant extension of step number and static balancing capability (Gurriet et al., 2019).
Bioinspired Rehabilitation and Neuroprosthetics: Functional electrical stimulation (FES) controllers based on Koopman operator theory effectively coordinate activation of plantar- and dorsiflexors in real time, maintaining ankle motion within a natural range and tracking targets within an RMSE of ~1.625° under optimal conditions (Singh et al., 10 Jan 2025).
Humanoid Robot Locomotion: Robot designs leveraging nonlinear transmission kinematics (e.g., parallel mechanisms) can analytically account for variable reduction ratios, supporting more dynamic, energy-efficient motions that faithfully map hardware limitations into whole-leg trajectory optimization (Lutz et al., 28 Mar 2025).
Clinical and Industrial AAFOs: Assistive ankle-foot orthoses with 2-DoF control, integrated sensor feedback, and adaptive algorithms show promise for general mobility aid, rehabilitation, and load-carrying assistance—lowering metabolic cost and increasing gait naturalness (Lora-Millan et al., 2023, Almeida et al., 14 Apr 2025).

Anticipated future avenues include:

Multimodal and adaptive feedback integration (EMG, IMU, force/torque, plantar pressure)
Real-time model personalization and online learning
Optimized lightweight mechanical designs for wearable, versatile assistance in ecological (out-of-lab) contexts
Expanded validation on patient populations and during complex, non-cyclic tasks.

Active 2-DoF ankle control thus constitutes a technological and methodological foundation for next-generation gait-assistive devices, aiming for robust, adaptive, and personalized locomotion support across a diverse range of users and environments.