Lower-limb exoskeletons have emerged as a promising technology to assist individuals with spinal cord injury in regaining walking abilities. Patients can be provided with assistance in various daily life activities: sitting, standing, walking in level-ground, ramps, and stairs. However, achieving intuitive control of these exoskeletons remains a significant challenge. In the context of rehabilitation, providing assistance-as-needed can be accomplished through the personalized control of the exoskeleton. However, this process often requires time-consuming calibration for each user and activity as well as recalibration throughout the progression of patient recovery. The goal of this research is to create an intuitive, machine learning-based controller for a lower-limb exoskeleton that generalizes across new users and novel activities without calibration.