Robotic assistive devices, such as exoskeletons and prostheses, are poised to fundamentally improve mobility and quality of life by augmenting and extending human capability for both able-bodied people and for people affected by stroke, polio, multiple sclerosis, spinal cord injury, and cerebral palsy. However, current approaches to control of powered exoskeletons and prostheses are rudimentary with numerous tunable parameters that are specific to each user, and offer no formal guarantees of either stability or safety. This is in stark contrast to the surge in control technology for highly dynamic bipedal locomotion. Our research generalizes formal control methodologies from bipedal robots to prostheses and exoskeletons, and explicitly accounts for the partial human model information through decentralized control and sensed external forces.