2024 Theses Doctoral

Robot-Assisted Posture Training Using Boundary-Based Assist-as-Needed Force Fields

Ai, Xupeng

Dynamic postural control requires regulating body alignment to achieve postural stability and orientation during functional movements. This ability may be impaired in people with neuromotor disorders, challenging them in performing daily activities. Conventional training strategies, such as muscle strengthening, joint locking, and proprioceptive training, are known to improve posture control. However, providing sufficiently rich intervention and maintaining high training intensity can be labor-intensive and expensive. Therefore, novel technologies are being explored to overcome the challenges.

Robot-assisted training is an emerging technology in posture rehabilitation. To maximize motor improvement, the assist-as-needed strategy is widely used in robotic platforms to provide adaptive assistance based on patients' functional ability. A prevailing paradigm employing the assist-as-needed strategy is the boundary-based assist-as-needed (BAAN) controller, which provides assistive forces when the center of mass moves beyond the stability boundary. This dissertation investigates the mechanisms underlying the efficacy of BAAN force fields and explores novel approaches to enhance the therapeutic effectiveness of BAAN robotic posture training protocols.

In Chapter 1, we outline the research background and introduce the main content of the following chapters in this dissertation. We also describe two cable-driven robotic platforms with BAAN controllers: the Robotic Upright Stand Trainer (RobUST) for standing posture training and the Trunk Support Trainer (TruST) for sitting posture training. In Chapter 2, we present a study using the RobUST platform to investigate how the BAAN force field impacts muscle synergy in the lower limbs during standing posture training. This pilot study provides insights into understanding the neuromuscular basis of the BAAN robotic rehabilitation strategy and helps explain its effectiveness. In Chapter 3, we present a deep learning-based dynamic boundary design for the BAAN controller. We conducted a controlled experiment with 20 healthy subjects using the TruST platform to test the dynamic boundary's effectiveness. This study highlights the clinical potential of the dynamic boundary design in BAAN robotic training.

Extended reality (XR) technology, including Virtual reality (VR) and augmented reality (AR), is gaining popularity in posture rehabilitation. XR has the potential to be combined with BAAN robotic training protocols to maximize postural control improvement. In Chapter 4, we conducted a randomized control experiment with sixty-three healthy subjects to compare the effectiveness of TruST intervention combined with VR or AR against TruST training alone. This study provides novel insights into the added value of XR to BAAN robot-assisted training and the differences between AR and VR when integrated into robotic training protocols.

Motor skills acquired through BAAN robot-assisted training necessitate consistent follow-up practice for long-term maintenance. However, due to portability limitations, BAAN robot-assisted training faces challenges in providing follow-up training after high-intensity in-lab robotic interventions. In Chapter 5, we present a remote XR rehabilitation system with markerless motion tracking for sitting posture training. This remote XR framework holds promise as an adjunctive training approach to complement existing BAAN robot-assisted training methods, maximizing motor improvements.