Feedforward and deterministic fuzzy control of balance and posture during human gait

The human body routinely performs tasks that require multi-segment limb control and inter-limb coordination which presents a substantial challenge to researchers. In particular, locomotor movements require balance and posture of the body which is especially complex because of the unstable nature of a bipedal system. However, locomotor activity is essential for the prosperity and well-being of virtually all people. Thus, modelling these functions of the central nervous system (CNS) could lead to profound benefits. The objective of this research endeavour is to develop a specialized control algorithm that models the posture and balance mechanisms employed by the CNS to regulate the upper body or HAT (head-arms-trunk) during gait.

Pertinent physiological variables are measured experimentally during slow, normal and fast walking speeds to establish bounds on key parameters related to posture and balance. In particular, it has been found that HAT oscillations are less than $\pm2\sp\circ$ during all speeds studied, agreeing with the results of other researchers. A model of the HAT is developed which incorporates characteristics of the hip musculature, pure time delays in the neural pathways and inertial properties of the HAT segment in the sagittal plane. An initial investigation shows that postural adjustments can be achieved with a linear quadratic regulator (LQR) although poor responses were observed when the model of the HAT was subjected to measured disturbances that occur at the base of the HAT. It is concluded that any purely reactive control strategy would not be able to perform adequately and that the CNS must use predictive or feedforward capabilities to overcome slow muscle responses and pure time delays. An investigation is conducted to find an appropriate feedforward control signal. Results of this study suggest that vestibular system would be an inappropriate feedforward balance indicator because the accelerations measured by the vestibular system lag the desired control signal. It is shown analytically that the disturbances that affect the HAT are related to the third derivative or jerk of the ankle position. It is speculated that this signal could be used as the root signal for controlling the entire lower limb during gait. Simulations show that utilizing this approach in conjunction with the LQR represents a more effective balancing mechanism than the LQR alone. Fuzzy control is investigated for further improvements in performance. This choice of control is supported with biological evidence which suggests that certain kinesthetic receptors function in a fuzzy-like manner. A deterministic design method is developed that allows a fuzzy controller to be generated directly from a linear controller. The mapping is exact and this allows the fuzzy controller to retain local stability characteristics of the linear controller while simultaneously incorporating fuzzy heuristics. Application of this deterministic approach leads to HAT responses that are similar to those found experimentally where the CNS has control, and thus satisfies the main objective of this work. Conclusions are made and an outline is given for future research.