TL;DR
This paper introduces a novel framework for variable assistance in lower body exoskeletons, enabling users to actively participate in walking while ensuring safety. By leveraging control barrier functions, the system provides just enough assistance to maintain a stable gait, allowing users to recover muscle function and learn to walk more naturally.
What Problem Does This Paper Address?
Traditional lower body exoskeletons often provide full assistance, rigidly following pre-programmed trajectories. While effective for restoring basic mobility, this approach can hinder rehabilitation for individuals aiming to regain muscle function. Studies suggest that allowing users some control and variability in their movements can accelerate the learning process and prevent learned helplessness. However, providing this freedom while guaranteeing the user's safety and maintaining a functional walking pattern is a significant challenge. The paper addresses this by developing a control strategy that dynamically adjusts assistance levels, allowing for user participation within safe boundaries.
Methods
The core of the paper's contribution lies in its variable assistance framework, built upon the Atalante exoskeleton. This framework integrates several key components:
- Patient-Exoskeleton Model Generation: A personalized dynamical model is created by fusing human anthropometric data with the exoskeleton's physical properties.
- Gait Library Generation: Dynamically stable walking gaits are generated using the Partial Hybrid Zero Dynamics (PHZD) method across various patient and gait parameters. This library is then fitted using a neural network for real-time trajectory generation.
- Baseline Walking Approach: This involves trajectory tracking using PID controllers with a deadbeat mechanism to handle early impacts. It also includes flat-foot ankle control for stable ground contact and minor adjustments for hardware flexibilities.
- Variable Assistance Architecture: This is the novel contribution. It uses control barrier functions (CBFs) to define "virtual guides" – safe trajectory tubes.
- Joint Idealization: Torques are computed to make assisted joints feel transparent, compensating for gravity and friction.
- Feedforward Assistance: Torques are provided to compensate for the exoskeleton's inertia, making it easier for the user to move the legs.
- Virtual Guide Filter: This is the key safety mechanism. It computes torques to limit the discrepancy between the actual and desired joint trajectories, ensuring they stay within the defined virtual guide (safe tube). The "assistance factor" (Ξ) controls the width of this virtual guide, with Ξ=1 being full assistance (no user input allowed) and Ξ=0 being minimal assistance.
- Haptic Feedback: Vibration motors provide real-time feedback to the user about their joint positions relative to the virtual guide boundaries, aiding them in staying within the safe zone.
The CBF approach is crucial here as it provides formal guarantees of safety, ensuring that the system remains within the desired safe set (the virtual guide) despite user perturbations.
Datasets
Not discussed in the paper. The experiments were conducted with able-bodied human subjects and the Atalante exoskeleton.
Key Insights and Ideas
- Safety-Critical Control for Human-Exoskeleton Interaction: The paper successfully applies control barrier functions, a technique from safety-critical control, to ensure user safety during dynamic walking with an exoskeleton. This is a significant step towards more robust and reliable human-robot interaction.
- Variable Assistance as a Rehabilitation Tool: The framework's ability to dynamically adjust assistance levels is highlighted as a key enabler for rehabilitation. By allowing users to actively participate, it promotes muscle strengthening and natural gait learning, avoiding the pitfalls of learned helplessness associated with purely passive systems.
- The "Virtual Guide" as an Intuitive Safety Constraint: The concept of a virtual guide provides a clear and intuitive way to define safe operating regions for the exoskeleton joints. The assistance factor directly correlates with the guide's width, offering a tunable parameter for controlling user effort and autonomy.
- Metabolic Cost as a Measure of Effort: The use of metabolic rate measurements provides a quantitative and physiologically relevant way to assess the user's effort and the effectiveness of different assistance levels.
Key Results
- Experimental Validation: The framework was demonstrated experimentally with eight able-bodied subjects.
- Trajectory Adherence: In all cases, actual joint trajectories remained within the virtual guide bounds, demonstrating the effectiveness of the virtual guide filter in maintaining safety.
- Metabolic Rate and Assistance:
- When subjects were passive, metabolic rate was similar between full and partial assistance.
- When subjects actively tried to track the nominal gait, partial assistance (Ξ=0.5) led to a higher metabolic rate compared to full assistance, indicating increased user contribution.
- A clear trend was observed: as the assistance factor decreased, both normalized metabolic rate and tracking error increased, suggesting that lower assistance incentivizes user effort.
- Tracking Accuracy: Subjects were generally able to improve tracking accuracy when actively trying to follow the gait in partial assistance.
Limitations and Future Work
- Gait Anthropomorphism: The generated gaits were noted as not being very anthropomorphic, which made it difficult for subjects to track them.
- Haptic Feedback Interpretation: The proposed haptic feedback was sometimes hard for users to interpret and act upon.
- Assisted Joints Selection: The current framework assists only specific joints (sagittal hip and knee of the swing leg).
- External Stabilization: As assistance factor decreased, the need for an operator to help stabilize the exoskeleton increased.
Future work suggested includes:
- Developing more anthropomorphic walking gaits.
- Investigating better haptic feedback strategies.
- Adding an active stabilization layer to the framework to address stability issues at lower assistance levels.
- Adapting walking gaits to match the user's natural gait.
- Investigating the use of variable assistance in true clinical rehabilitation settings.
Summary updated