Humanoid postural control based on two biologically-inspired approaches

This paper presents two approaches for controlling a humanoid upright posture. The first approach directly mimics the human postural control as understood and identified by biologist and neurologists while the second approach constitutes standard control engineering techniques arranged in a control architecture that is inspired by relevant biological findings. The former architecture is composed of an ankle angle feedback loop for body-foot control. Simple sensor fusions are used to yield estimates of external disturbances such as body support motion or contact forces having impact on the body. They are fed into the loop for disturbance compensation and upgrade it with the help of body-space signals from the equilibrium (vestibular) organs into a body-space control. A body-space command signal is used as set point signal for desired body position or motion and the error signal is fed into a linear PID controller. The latter architecture is composed of an extended observer that can estimate unknown external stimuli based on a dynamic model, a compensation mechanism, and a linear PID controller with the possibility of introducing PD feed-forward terms. Experiments on a special-purpose humanoid robot show the applicability of the two presented approaches in terms of stabilizing the system and tracking a desired voluntary motion in the presence of superimposed external stimuli. Although theoretically a large set of individual sensor combinations can be employed for estimation, using the vestibular sensors demonstrates a particularly robust estimation.