Publication
Augmentative lower-limb exoskeletons and exosuits (collectively: exos) hold transformative potential for individuals across the mobility spectrum, from people affected by impairments to able-bodied persons aiming to enhance their physical capabilities. However, the widespread use of exos for daily ambulation is an aspiration yet to be fulfilled, with various challenges remaining to be addressed. A primary difficulty arises from the dynamic nature of human locomotion, which is constantly adapting to factors such as the environment and situational demands. For exos to keep pace with these continuous adjustments, a flexible control system that can seamlessly adapt to various locomotion tasks is essential. Therefore, the aim of this thesis is to contribute toward realizing the vision of widespread exo adoption by addressing the challenge of task-adaptive control, focusing specifically on hip exoskeletons.
Theoretical frameworks for adaptive exo control have advanced considerably in recent years, and promising results have been achieved in research. However, many of these solutions are either too complex and/or face practical constraints for real-world application, hindering their translation from the lab into everyday use. Recognizing the interplay between theoretical advancements and practical constraints, we explore a range of control strategies, with principles of practicality and parsimony guiding our decisions about algorithms and sensors.
We first focused on the timing of assistance, a critical factor in ensuring that the exoskeleton's actions are synchronized with the user's intentions. Targeting the basic problem of real-time gait phase estimation, we investigated the potential of phase variables for robust and computationally efficient timing adaptation. Following an in-depth performance evaluation of the existing methods, we proposed an improved approach using geometric transformations and online optimization. This approach achieved estimation accuracy levels comparable to state-of-the-art methods, with relatively lower computational and sensing requirements.
Next, we integrated gait phase estimation into control and contrasted this "explicit synchronization" approach with a method based on "implicit synchronization", a paradigm of directly mapping sensory inputs to assistive torques. The implicitly synchronized approach led to higher metabolic benefits and showed better robustness, highlighting its potential for real-world applications. Building upon this method, we complemented its adaptivity by incorporating heart rate feedback for amplitude modulation. Testing in a challenging real-world scenario demonstrated the effectiveness of this approach, and more broadly, the potential of heart rate feedback for assistance adaptation.
Finally, using the same idea of leveraging the information embedded in the user's state for adaptation, we turned to exploring strategies based on bio-inspired neuromuscular models. We refined the commonly used