Toward Faster Parameter-Tuning of Sampling-Based Motion Planners

Abstract

This paper provides an alternative parametertuning method for sampling-based motion planners, while handling case studies of global path planning for multi–Degrees Of-Freedom DOF manipulators. Observations of the behavior of average-planning-time as function to parameters’ values are presented for the Open Motion Planning Library OMPL bidirectional planners. Based on the presented observations, recommendations are made of which parameters are worth to be tuned. It is also discussed how time-efficient is to move the parameter tuning problem from searching for multiple parameters together, to be searching for a single parameter at a time. Moreover, an alternative single-parameter-tuning method is proposed depending on polynomial interpolation of a pre-chosen performance-estimation function. The proposed method has been formulated using average-planning-time as a performance-estimation function, and has been tested in tuning bidirectional planners’ parameters in three different scenarios equipped with three different manipulators (LBR IIWA 14 R820, KUKA KR3 R540, and Kinova JACO). The proposed method was compared and outperformed tuning results of the Sequential Model-based Algorithm Configuration SMAC where it is used as a multiple-parameter- tuning method.