Accuracy Preserving Interpretability with Hybrid Hierarchical Genetic Fuzzy Modeling: Case of Motion Planning Robot Controller

Abstract

Design of robot controller for motion planning, using fuzzy logic control, requires formulation of rules that are collectively responsible for necessary levels of intelligent behaviors. To ensure the model interpretability, this collection of rules can be naturally decomposed and efficiently implemented as a hierarchical fuzzy model. This paper describes how this can be done using hybrid hierarchical genetic fuzzy modeling. The idea is to combine, in a hierarchical design, "mapping" for sub-goal behavior (SGB), and "reactivity" for local avoiding obstacles behavior (LAOB), to have at the same time, an interpretable and precise communicating system for robot motion planning controller. The design of each fuzzy unit of the hierarchical model is automatically ensured by MAGAD-BFS method (multi-agent genetic algorithm for the design of beta fuzzy systems), promoting itself as an interpretability-accuracy trade-off. A proposed reduced version of generalized local Voronoi diagram (RGLVD) comes to guarantee a high degree of precision for robot motion to attempt destinations (sub-goals). Compared to the navigation using only fuzzy rules controller, the hybrid hierarchical model is more efficient in terms of saving time and optimizing path