Model-based optimisation for the personalisation of robot-assisted gait training

TL;DR

This work addresses personalised robot-assisted gait rehabilitation by introducing an offline, model-based optimisation framework that tunes a path-control impedance controller through musculoskeletal modelling and forward-dynamics simulation. Human torques are estimated from motion data in a feedforward OpenSim model, and a multi-objective objective combines tracking error and exoskeleton effort to yield subject-specific stiffness parameters via a gradient-free optimiser. Simulations predict substantial improvements with personalised parameters, but experimental results in 18 healthy participants show high inter- and intra-personal variability, with only some individuals benefiting and several performing worse, underscoring the need for more accurate human-behaviour estimation and robust interaction modelling. The study demonstrates feasibility and highlights key challenges for translating offline personalisation to reliable clinical gains in robot-assisted gait training.

Abstract

Personalised rehabilitation can be key to promoting gait independence and quality of life. Robots can enhance therapy by systematically delivering support in gait training, but often use one-size-fits-all control methods, which can be suboptimal. Here, we describe a model-based optimisation method for designing and fine-tuning personalised robotic controllers. As a case study, we formulate the objective of providing assistance as needed as an optimisation problem, and we demonstrate how musculoskeletal modelling can be used to develop personalised interventions. Eighteen healthy participants (age = 26 +/- 4) were recruited and the personalised control parameters for each were obtained to provide assistance as needed during a unilateral tracking task. A comparison was carried out between the personalised controller and the non-personalised controller. In simulation, a significant improvement was predicted when the personalised parameters were used. Experimentally, responses varied: six subjects showed significant improvements with the personalised parameters, eight subjects showed no obvious change, while four subjects performed worse. High interpersonal and intra-personal variability was observed with both controllers. This study highlights the importance of personalised control in robot-assisted gait training, and the need for a better estimation of human-robot interaction and human behaviour to realise the benefits of model-based optimisation.

Figures (10)

• Figure 1: The offline model-based pipeline for control personalisation: (A) The physical properties and the motion of both the user and the robot are measured using (B) motion capture technologies to (C) create a personalised human-robot model and (D) obtain an estimate of the controls of the human for the completion of a task. (E) Together with the controller constraints and (F) the objectives of the collaborative task, (G) the optimal controller structure and/or controller parameters are obtained offline. (H) The outputs obtained from the offline optimisation are then used to design and/or fine tune the real-life robot controller to provide personalised assistance.
• Figure 2: Placement of reflective markers for model scaling. Tracking markers were placed on the right (1) and the left (2) shoulder and an anatomical marker was placed on the sternum (3). Three anatomical markers were used to define the pelvis with one marker on the right anterior superior iliac spine (ASIS) (4), one on the left ASIS (5) and one on the sacrum (6). A cluster of three tracking markers was used on the thighs (7a-7c) and anatomical markers were placed on the lateral (8) and medial (9) femoral epicondyles. A cluster of three tracking markers was also placed on the shanks (10a-10c). The feet were defined with five anatomical markers placed on the lateral (11) and medial (12) malleoli, the fifth (13) and the first (14) metatarsal heads and the heel (15).
• Figure 3: Illustration of the dual-phase reference kinematic path, ${\bf Q}_{ref}$, defined in joint space, and an instance of the dynamic allocation of the reference point, ${\bf q}_{ref}$, based on the pose of the exoskeleton, ${\bf q}_{act}$.
• Figure 4: (a) Data collection setup. Participant wearing an exoskeleton and performing a unilateral tracking task with the help of visual feedback. (b) Experimental protocol including four phases of pretest training with visual feedback and the experimental validation where the baseline controller and the optimised controller were tested in a randomised order.
• Figure 5: Personalised stiffness obtained for each participant for the two phases of the cycle for both the hip joint and the knee joint of the exoskeleton.