Model-free human-like humanoid push recovery

In this paper we present a model-free system for humanoid push recovery. The concept is inspired by human proprioceptory sensory input used for detecting balance disturbances. In our implementation, the foot pressure sensors found on a humanoid are used to detect the degree of sway in any direction. This difference is fed to a PID controller which sets the proper ankle torque to be applied in order to keep the humanoid standing in the upright position. This approach imitates the proprioceptive sensor information used by humans which dominates posture assessment. The system is implemented on the NAO H25 V3.3 humanoid in Webots against push disturbances in various directions. The system is also tested on the actual Nao humanoid.