Performance Models in Robotics With a Use Case on SLAM

Abstract

The performance of a software component implementing a robotic functionality depends on many factors ranging from the system configuration (e.g., available sensors and robot kinematics) to the operating environment, passing by the component configuration parameters. A naive approach to model the performance of such a software component is to measure its performance on every possible combination of such variables. However, this is not possible as the number of combinations would not be tractable, considering also that multiple measurements should be performed for each of them. To make the problem tractable, we propose to sample a relatively small number of combinations, conduct experiments for each of them, and from these results estimate a statistical model of the software component performance, which we call <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">component performance model</i> . A performance model allows the comparison of different components implementing the same functionality to determine the best one to be used in a given setting and its optimal configuration. Moreover, performance models of multiple functionalities may be composed to predict the performance of an entire system at design-time. Besides the general framework to extract performance models, here we present an operational use case in Simultaneous Localization and Mapping (SLAM).