Less computationally intensive fuzzy logic (type-1)-based controller for humanoid push recovery

Abstract

In this study, we present a new type-1 fuzzy logic-based controller for push recovery by humanoids. The objective of this study is to develop an intelligent controller and to implement biologically inspired push recovery for humanoid robots. The fuzzy inference system takes two crisp values as inputs, which are fuzzified, before a number of rules are applied, and finally the output is defuzzified to convert it into a crisp value. We apply fuzzy rules to our model, which we simulate in an unstructured environment. The objective is to reduce the fuzzy rules and make the fuzzy inference set less computationally intensive and fast, as well as exploiting the advantages of easy trainability and high generalizability. The fuzzy logic-based controller can predict the requisite push recovery strategy and whether the robot will be able to recover or fall. The architecture has a hierarchical design. The first fuzzy inference system (FIS1) is based on two input variables: the force and the direction of motion, where the result depends on the magnitude of the force applied to the body and direction in which the body moves after being pushed. FIS1 can determine small, medium, and large forces in term of roll and pitch effects on the body. These outputs are the input variables employed by FIS2 to predict the push recovery strategy that will be applied, and eventually the robot will be able to recover from a push or fall. The term “auto-leaning” from human autonomy is introduced in the context of push recovery. We extend the Gordon model for balancing humanoids using fuzzy logic by considering the effects of roll, pitch, and yaw. This extends earlier research in the field of humanoid push recovery, where the force was only applied in one direction, to studies of pushing in different directions. In a cluttered environment, pushing is a very common experience, from which humans can readily recover whereas humanoid robots cannot. Humanoid robots have more degrees of freedom, thus finding solutions using alternative methods is not a simple task. The novel feature of this study is the introduction of an intuitive fuzzy logic-based learning approach and we show that it is fast and effective. This exhaustive fuzzy logic-based design is the main focus of this study.