Kinematic arrangement optimization of a quadruped robot with genetic algorithms

Abstract

Background: As research on quadruped robots grows, so does the variety of designs available. These designs are often inspired by nature and finalized around various technical, instrumentation-based constraints. However, no systematic methodology of kinematic parameter selection to reach performance specifications is reported so far. Kinematic design optimization with objective functions derived from performance metrics in dynamic tasks is an underexplored, yet promising area. Methods: This article proposes to use genetic algorithms to handle the designing process. Given the dynamic tasks of jumping and trotting, body and leg link dimensions are optimized. The performance of a design in genetic algorithm search iterations is evaluated via full-dynamics simulations of the task. Results: The article presents comparisons of design results optimized for jumping and trotting separately. Significant dimensional dissimilarities and associated performance differences are observed in this comparison. A combined performance measure for jumping and trotting tasks is studied too. It is discussed how significantly various structural lengths affect dynamic performances in these tasks. Results are compared to a relatively more conventional quadruped design too. Conclusions: The task-specific nature of this optimization process improves the performances dramatically. This is a significant advantage of the systematic kinematic parameter optimization over straight mimicking of nature in quadruped designs. The performance improvements obtained by the genetic algorithm optimization with dynamic performance indices indicate that the proposed approach can find application area in the design process of a variety of robots with dynamic tasks.