Model predictive impedance control : a model for joint movement control
A human is capable of making various movements under different environmental conditions including highly accurate movements, low accurate movements, ballistic movements, or learning new skills. This flexibility results from highly integrated neural centers in the brain and the spinal cord acting as higher and lower level controllers and information processors, a large number of muscles and bones constructing the actuators required to generate movements, and numerous sensory receptors informing the neural centers about the results of movement and the environment. For a comprehensive understanding of the motor control system, it is necessary to have a global model. This model can also be used to test the function of motor components under natural or pathological conditions. This thesis is an effort toward developing such a global model based on the most currently accepted theories and hypotheses in biological motor control and control engineering. The proposed model, called model predictive impedance control,specifically combines the equilibrium point hypothesis (α and $\lambda$ models), the impedance control strategy (including stiffness and viscosity), servomechanism control theory, and the optimization-based model predictive control algorithm as a unified model applicable in the study of different types of movements. The model is adaptive with learning ability and operates in open-loop or closed-loop manners. The focus is on the overall function of motor centers instead of individual realization structurally or functionally. Acting as a supervisory and higher level controller, the model predictive controller presents a new approach to determine the joint impedance and the equilibrium point based on use of a priori knowledge of the neuromusculoskeletal system and environment. To evaluate the performance of the model, it was applied to three different types of joint movements: a tracking movement with an unpredicted disturbance, a rhythmic movement, and an unstable biped model of human walking. Computer simulation results showed excellent performance of the model in all three cases for optimal values of active joint impedances and a perfect match between the musculoskeletal system and the model internal to the model predictive controller. The controller was also able to maintain acceptable performance in the presence of a 25% mismatch between the musculoskeletal system and its internal model.