Comparison between Self-Tuned and Benchmark Assistance in Healthy Subjects Using an Active Pelvis Orthosis

Lower-limb exoskeletons are emerging as cutting-edge technology to enhance human mobility. Despite recent advances, yet major challenges in realizing effective control system to have synergistic cooperation between the user and the exoskeletons. It can be achieved by adapting the control parameters to the user. Human-In-The-Loop Optimization methods propose to automatically customize and optimize exoskeleton control parameters closing the loop on human performance, such as metabolic cost or muscle activity. However, optimization methods do not consider the user’s preference, that may capture a broader spectrum of factors associated to functional assistance. Considering user preference into exoskeleton control systems could increase functional benefits from exoskeletons use, user satisfaction, adoption, and acceptability of the technology. In this framework, the aim of this thesis was to investigate the users’ preferences in the tuning of the robotic assistance of a bilateral hip exoskeleton, an Active Pelvis Orthosis (APO), which provides hip flexion-extension assistance while walking. Specifically, the present thesis compares self-tuned assistance based on the user’s preference with a benchmark assistance tuned by an expert. Both types of assistance were modeled as phase-locked torque profiles shaped as sum of two Gaussian functions, one positive for hip flexion and one negative for hip extension. The benchmark assistance estimates the hip torque profile by amplifying the difference between the hip angular profile measured by hip encoders and the profile predicted by the adaptive oscillators. While the self-tuned assistance was blindly tuned by the participants, through a graphical user interface in which six knobs were associated with six control parameters. i.e., the amplitudes, the phase of the peaks, and the duration of the two Gaussian functions. Experiments were carried with a pool of subjects naïve to the device and the control system. Each subject self-tuned the assistance searching to maximize comfort, stability, and minimize perceived exertion while they are walking with the APO in assistive mode. To explore the biomechanical effects of the two types of assistance, electromyography, and physiological cost index (PCI) were measured while participants walked on a treadmill with 6% of slope for 20 minutes at 1.25 m/s of speed with the APO with both the assistances and without the APO; while at the end of each walking task, participants were asked to attribute a value in the Borg scale. Furthermore, to assess the effect of the assistance in terms of functional outcome, the self-selected speed preferred by subjects was measured. On average, the self-tuned assistances resulted different from the benchmark ones. Across participants average variations of PCI, self-selected speed and Borg scale score obtained with self-tuned assistance were respectively 11% higher, 8% smaller and 3% higher than those obtained with expert-tuned assistance. Experimental tests have shown that each subject prefers different robotic assistance, which deviates from a kinematically advantageous one, confirming the need for custom-made control. In this study, the benchmark assistance demonstrated to be more efficient than the self-tuned one, which was considered by every subject more comfortable and stable at the beginning, but less effective in the long term. This study can therefore be regarded as a pilot work, which can open the way for further approaches of preference-based control strategies.