Abstract
In this dissertation, a control relation for the hip, knee and ankle joint torques for both legs during normal human walking is characterized with respect to the joint angles and velocities of the right and left hip, knee and ankle joints. The walking data of seven healthy subjects are used to quantify this controller. A 20% sparse phase invariant multivariable linear controller is shown to be one of the simplest controller that is able to explain the walking data of individual subjects, because this controller does not require multiple phases unlike current state-of-the-art multiphase controllers used to explain human walking data. Due to the phase invariant structure of the obtained controller, one does not require a finite state machine and phase detection that are commonly used in the control of human augmentation and assistive devices. This property of a phase invariant controller could bypass potential errors during phase detection that originate from environmental uncertainties. Following this, a four phase multivariable linear controller that switches according to the events of toe-off and heel-strike is shown to generalize to the walking data of seven subjects. Due to the variance that is present in the walking data of multiple subjects, a multiphase and multivariable linear controller is proposed and an algorithm which could estimate this controller is presented. For the walking data of these seven subjects, it is shown that both a multiphase and multivariable controller is necessary to generalize to the walking data of multiple subjects. Such a control approach extends the current state-of-the-art multiphase but single joint controllers to a multiphase and multivariable controller. Because the obtained controller is able to better explain the walking data of multiple subjects, this controller could improve the control of next generation assistive devices used to augment and assist humans. Finally, a phase invariant multivariable linear controller is shown to generate the walking motion of a moderately complex seven-link biped in simulation. Such a control structure is atypical due to the hybrid dynamics of biped robots that contain multiple phases within a full step and current state-of-the-art controllers that adopt a phase dependent approach to generate robotic locomotion. Our proposed phase invariant controller could simplify the implementation of current state-of-the-art multiphase controllers on a biped robot, because this controller does not require a finite state machine, phase detection and retuning of the control parameters for each motion phase.