Force control and EMG-based adaptive assistance for enhanced human-exoskeleton interaction

The increasing diffusion of exoskeletons in industrial, domestic, and medical environments highlights their potential to reduce work-related musculoskeletal disorders, support rehabilitation, and improve the mobility of people with disabilities. Unfortunately, control technologies capable of ensuring a reliable physical human-exoskeleton interaction and providing optimal assistance based on the load lifted by the user are still missing. This thesis investigates advanced control strategies that address key limitations of existing approaches by acting at both the low-level and high-level layers of exoskeleton control architectures. The first part of the thesis focuses on low-level force control. First, a force control benchmarking framework is developed to highlight well-known critical challenges in force control and to evaluate force control algorithms by taking into account diverse interacting environments. Next, a high-performance friction compensation solution for benchmarking applications is proposed to guarantee non-biased benchmarking results. Since the latter solution cannot be exploited outside the benchmarking scenario and given the importance of accurate friction compensation for reliable human-exoskeleton interaction, this thesis proposes a novel force control architecture that includes a model-reference friction observer. Since passivity is a fundamental requirement for interaction control, the conditions under which the proposed observer preserves passivity at the environment, control, and friction ports were investigated. This implies that the proposed architecture can perform passive friction compensation for any friction dynamics and can be combined with any passive force controller while guaranteeing a stable interaction with any passive environment. Experimental validation shows that the proposed approach outperforms existing friction compensation solutions in force control applications. The second part of the thesis focuses on high-level Electromyography (EMG)-based control strategies for Adaptive Gravity Compensation (AGC). To introduce the reader, a brief overview of Myoelectric Control Strategies (MCSs) is presented, showing that MCSs found in the literature are generally composed of three distinct functional modules: a decoder to extract the movement intention from EMG signals, a controller to accomplish the desired motion through a command given to the actuators, and a shaper to connect them. Subsequently, two EMG-based AGC approaches are proposed. Differently from existing AGC solutions, the proposed approaches (1) do not require knowledge of the anthropomorphic properties of the human arm and (2) recognize the importance of accounting for the human-exoskeleton dynamics within the adaptation law to improve payload estimation performance in dynamic conditions. The feasibility of the proposed approaches is experimentally validated on a 1-degree-of-freedom upper-limb exoskeleton.