Bio-inspired optimization of kinematic models for multi-legged walking robots

摘要

Designing a walking robot for a specified task is a complex problem. It requires numerous calculations and evaluations to find the desired shape and topology to perform correctly. The abstraction of such a task is a multi-dimensional, multi-goal optimization problem. This paper proposes a bio-inspired solution to optimize such a kinematic model: A genetic algorithm aims to free the designer from the cumbersome procedure of calculating forces, evaluating models and selecting certain hardware components. Through a set of predefined preferences, it can be tasked to develop a light, stable or fast robot. It uses the classic evolutionary mechanisms of selection, recombination and mutation and adapts at runtime. The algorithm also allows the definition of a morphological type, a blueprint for walking robots, derived from common robot classes and calculates their dynamic model with a physics engine.