Stair Ascending Stair Descending Standing Walking Intention Detection Kinematics Human + Exoskeleton AOs Kinematics Control Mode Frequency Phase (a) Torques Weights Primitives Hip Joint Torques Knee Ankle % Gait Cycle Phase Frequency Stimulations (b) Weights HFL Musculoskeletal Model GLU Primitives HAM Joint Torques VAS GAS TA % Gait Cycle SOL Figure 1. A bioinspired controller: (a) a controller based on dynamic primitives (DLMPs) and (b) a controller based on neural primitives (NLMPs). The components of the developed controllers (AOs and primitive-based controllers) are shown in red. The blue box represents the algorithm detecting the motor intention of the user. The green box represents the human-orthosis coupled system. HFL: hip flexor; GLU: gluteus; HAM: hamstring; VAS: vastus; GAS: gastrocnemius; TA: tibialis anterior; SOL: soleus. These advantages make the physiological concept of motor primitives an inspiring foundation for the control of artificial devices such as orthoses, prostheses, and even autonomous robots. These devices may benefit from a primitivebased control in complex environments where different locomotion modes are requested [21]. Furthermore, the use of the same reduced set of primitive signals for different locomotion tasks presents the desirable advantage of generating smoother transitions between them. In this article, the generation of artificial primitives is studied at two different levels, the dynamic and neural levels. The controller based on dynamic-level primitives combines a set of march 2016 * IEEE ROBOTICS & AUTOMATION MAGAZINE * 85