An approach for combining transparency and motion assistance of a lower body exoskeleton

AI summary

Overview

Research area: Robotics — wearable robotics, specifically lower-body gait exoskeleton control.

Technical level: Intermediate. The paper uses nonlinear dynamical systems (adaptive oscillators) and torque control laws, but presents the underlying ideas at an accessible level.

Scope: The paper describes a control approach that combines a "transparent" mode (the device does not resist the wearer) with a superimposed gait-assistance mode on a prototype lower-body exoskeleton.

What This Paper Is About

Wearable exoskeletons must both stay out of the way when the user moves freely and actively help when assistance is wanted. This paper proposes a way to do both on the same device without adding force or torque sensors, by exploiting the mechanical backlash inside the gearboxes and by learning the wearer's periodic hip motion with adaptive oscillators. The goal is to produce a smooth assistive torque that pushes the legs toward their estimated future position while the device stays otherwise transparent to the user.

Key Contributions

1. A sensor-free transparency mode. The gear backlash of the actuation units (an angular range of about ±0.5°) is used as the source of a control signal: the difference between the motor position and the gear-output position drives a virtual-stiffness torque that compensates inertia and friction, so the device does not significantly hinder free motion.
2. A gait-assistance mode built on adaptive oscillators. A coupled set of adaptive oscillators performs a dynamic Fourier-series-like decomposition of the measured hip angle, learning its frequency, phase and amplitude components so an assistive torque can be computed from the predicted future position.
3. A weighting scheme that gates assistance on estimation quality. A tanh-based weight w_{A,j} switches assistance off when the discrepancy p between the measured and estimated signal exceeds a tunable maximum p_max, so help is only applied when the learned signal actually resembles the measured motion.
4. Combination of both torques and an exoskeleton prototype implementation. The final actuator torque is the sum of the transparency torque and the weighted assistive torque, tested on a modular four-actuation-unit prototype at the Institute of Robotics, Johannes Kepler University Linz.

Main Findings

• Transparency works as a pendulum test. A hip joint initially deflected by a certain angle and then controlled with the transparency torque converged quickly to its equilibrium; because of an eccentrical center of mass, that equilibrium is not the vertical. The strong damping of the oscillation comes from the inertia and friction of the planetary gear.
• The oscillator learns the gait signal within roughly one cycle for the left leg. In the reported measurement, the adaptive oscillator system was re-activated and reset to defined initial conditions at about 0.5 s. The learned signal for the left leg was close enough to the measured joint angle after one gait cycle, and the assistive torque was smoothly switched on.
• The right leg required more cycles. The learned signal of the right leg was not accurate enough during the first gait cycle, so the assistive torque was switched off again, and motion assistance started with the third gait cycle.
• Safety torque limit applied. The total desired torque was limited to ±0.275 Nm during the experiments.
• Assessment was subjective, not benchmarked. The paper states that, due to a lack of usable standard tests, the effects of transparency and assistance were assessed just by subjective evaluation, supported by measurements of actuator positions and torques. No quantitative performance metrics or standard benchmark scores are reported.
• Learning gains were tuned on synthetic gait. Artificially generated hip angle trajectories, produced with an approach from a previous publication to simulate a person walking at variable gait speed, were used to tune the learning gains and to bound the system parameters.

Methodology in Plain English

The team started from a hardware constraint: their exoskeleton has no force or torque sensors. Every gearbox has a small amount of play (backlash). Within that play, the output side can rotate a little without the motor turning. The system measures both the motor position and the position on the output side; their difference tells the controller how far the joint has drifted into the backlash. Multiplying that difference by a tunable virtual stiffness produces a torque that damps unwanted inertial effects and makes the joint feel light to the wearer.

For assistance, the researchers use adaptive oscillators — a set of coupled nonlinear dynamical equations whose intrinsic frequencies are integer multiples of a fundamental frequency. Fed with the measured hip angle, the oscillators pull apart the signal into amplitude and phase components and track the fundamental frequency on their own, which lets the system reconstruct the current joint angle and predict its value a fixed time horizon ahead. The assistive torque is proportional to the difference between the predicted future angle and the current estimate, so the leg is guided toward where it is expected to go next.

Because a wrong prediction would mean a wrong push, a weighting term based on a hyperbolic tangent compares the actual signal against the estimate. If the mismatch is small, the weight is high and assistance is applied; if the mismatch grows past an adjustable threshold, assistance fades out. The final command sent to each actuator is simply the transparency torque plus this weighted assistive torque. The whole scheme was implemented on the Linz prototype, using the two-hip-joint configuration with the knee actuators detached, and tried by people walking on level ground, including the transition from standing to walking.

Why This Matters

Research impact: the paper offers a cheap route to transparency. Instead of adding force/torque sensors, it reuses the backlash that gearboxes already have, which lowers cost and complexity for exoskeleton control research. It also demonstrates combining a passive-feeling mode with an assistive mode rather than treating them as separate devices.

Real-world applications:

• Robotic gait rehabilitation for patients recovering walking ability, where the device must not fight the patient between therapy movements.
• Mobility assistance for older adults affected by demographic change and declining mobility.
• Everyday walking support and sit-to-stand or stance-to-walk transitions for people with lower-limb weakness.
• Repurposing as a research or prosthetics/orthotics testbed where new periodic-motion control strategies can be swapped in.

Industry relevance: the approach targets a common industry trade-off — commercial exoskeletons are typically closed platforms that are hard to modify for research, and adding sensors raises cost and weight. A sensor-free transparency strategy that fades assistance in and out based on signal confidence is directly relevant to makers of rehabilitation and assistive exoskeletons. The work was supported by the "LCM - K2 Center for Symbiotic Mechatronics" within the Austrian COMET-K2 program.

Future Directions

• Extensive experiments beyond level-ground walking, specifically stair ascent and descent and other cyclic movement tasks.
• Designing objective performance measures, since the authors note that no usable standard tests exist and current assessment was subjective.
• Testing the four-degree-of-freedom configuration with the knee actuators reattached, since the reported experiments used only the hip joints.
• Investigating how robustly the adaptive oscillator gains and the tanh gating threshold generalize across users, gait speeds and non-periodic movement, given that tuning relied on artificially generated gait patterns.

Target Audience

Researchers and engineers working on wearable robotics, exoskeleton control and robotic gait rehabilitation will benefit most, particularly those interested in sensor-free transparency or in oscillator-based assistance torque generation. The paper is also useful for graduate students entering the field because it lays out the control laws, the gating logic and the practical hardware constraints in a compact form. Clinicians and device developers evaluating low-cost transparency strategies will find the conceptual argument relevant, though they should note the reported results are initial and subjectively evaluated rather than benchmarked.

Authors’ abstract

In this paper, an approach for gait assistance with a lower body exoskeleton is described. Two concepts, transparency and motion assistance, are combined. The transparent mode, where the system is following the user's free motion with a minimum of perceived interaction forces, is realized by exploiting the gear backlash of the actuation units. During walking a superimposed assistance mode applies an additional torque guiding the legs to their estimated future position. The concept of adaptive oscillators is utilized to learn the quasi-periodic signals typical for locomotion. First experiments showed promising results.