Innovative control solutions for wearable robotics using machine learning

The enormous advances in robotics performed over the last few years are enabling wearable robots to increasingly move out of the laboratories toward real-world applications, proving to be an invaluable resource for the evolution of a society that moves at the speed of its technological advances. Wearable robots add quality to our time and will enable our society to be more active, productive, creative, and independent throughout all stages of life. However, the symbiotic cooperation between the robot and the human body, essential for a safe and effective user experience, can only be obtained through an efficient and synergic interaction. As a consequence, any potential benefit of this interaction highly depends on the efficacy of the ergonomics and the transmission of power, as well as a stable and intuitive flow of information. This thesis presents the design and development of innovative control methods for the detection and recognition of the motor intentions of the user in lower-limb wearable robots. In particular, the presented work addressed this topic with a particular focus on the use of potent and versatile machine learning algorithms to minimize the number and complexity of the sensory system and to achieve robustness to intra- and inter-subject variabilities. The thesis also presents a novel concept of a low-power ankle-foot prosthetic module, conceived to enhance the energy storage and command the energy return of a standard passive foot. The results of the experimental activities advance the current state of the art and investigate new technological solutions that can serve and improve the quality of life of impaired individuals.