Stroke is a leading cause of disability, and post-stroke individuals often experience hemiparesis, which can result in compensatory movements that affect their overall gait pattern. Lower-body exoskeletons allow for repetitive, intense, and task-based practice of movement patterns that can promote gait recovery. Validating exoskeleton gait patterns in simulation is vital for ensuring patient safety and comfort;​ however, current simulation tools possess challenges in controlling exoskeleton hardware. Additionally, present model-based adaptive controllers rely on complex dynamics models for the user and exoskeleton that are difficult to accurately define. In this thesis, a centralized exoskeleton simulation and control framework is presented, where desired gait patterns can be validated and customized on-the-fly in physiotherapy sessions. A model-free, end-to-end deep reinforcement learning controller is also presented, to allow for exoskeleton control that is robust to desired gait patterns and a user's interaction with the device. Several experiments were conducted to validate these approaches.