Glen Berseth

SteerFit uses the SteerSuite framework and additional plugins to optimize parameter settings for dynamic navigation algorithms.

Abstract

We propose a statistical framework and a methodology for automatically characterizing the influence that a steering algorithm's parameters have on its performance. Our approach uses three performance criteria: the success rate of an algorithm in solving representative scenarios, the quality of the simulations solution, and the algorithm's computational efficiency. Given an objective defined as a weighted combination of the performance criteria, we formulate an optimization problem in the space of the algorithm's parameters that we solve with an evolutionary-based approach. Although our framework can analyze an algorithm from many perspectives, we present two demonstrative studies: a univariate analysis that studies the effects of each of the algorithm's parameters in isolation, and a multivariate analysis that studies the combined effect of the algorithm's parameters on the objectives. % We apply our methodology to two established steering approaches, and show that fitting optimal values to the parameters can significantly improve the performance of both algorithms over the commonly used default values.

You can find the poster describing the project here

This video demonstrates some of the example results of parameter optimization.