The use of exoskeletons for clinical lower-limb stroke rehabilitation offers the potential of improved and customized rehabilitation that reduces the requirements and demands placed on multiple staff members. Initial research with lower-limb exoskeletons show potential to alleviate this problem. Conventional assistance-based exoskeleton devices simply enforce the desired gait trajectory for the patient in order to ensure safety and stability. Unfortunately, if the end-user does not have to work to contribute to successful motion, rehabilitation often does not occur. Recent evidence has suggested that assistance-asneeded control prevents users from slacking, facilitating functional motor recovery. Assistance-as-needed control turns off assistive torques during periods when the patient is able to execute a desired gait pattern but if the patients gait deviates sufficiently from a desired trajectory then assistive torques are generated to compensate for the patients loss of strength. This strategy encourages the patient to contribute effort while still enabling the exoskeleton to guide movements. Assistance-as-needed control inherently leads to aperiodic gait patterns and has accordingly been difficult to employ in lower-limb exoskeletons due to the need to ensure stability. This work demonstrates how virtual constraint control-a method with robust stability properties used in prostheses and assistive exoskeletons control-can be combined with a velocity-modulated deadzone to ensure stability. Simulations suggest that the method can accommodate a large deadzone while remaining stable across a range of unanticipated gait pathologies, as demonstrated using Lorenz mappings that can accommodate the aperiodic nature of the resulting gait.