System interdependence analysis for autonomous robots

Abstract

With the increasing complexity of robotic systems, system robustness and efficiency are harder to achieve, since they are determined by the interplay of all of a system’s components. In order to improve the robustness of such systems, it is essential to identify the system components that are crucial for each task and the extent to which they are affected by other components and the environment. Such knowledge will help developers to improve their systems, and can also be directly utilized by the systems themselves, for example, to detect failures and thereby correctly adjust the system’s behavior. In this article a method of system interdependence analysis is presented. The basic idea is to learn and quantitatively evaluate the coherence between performance indicators of different system components, as well as the influence of environmental parameters on the system. To validate the proposed approach, system interdependence analysis is applied to the navigation system of an autonomous mobile robot. Its navigational methods are presented and suitable indicators are derived. The results of using the method, based on experimental data from an extended field experiment, are given.

A Framework for Predicting Future System Performance in Autonomous Unmanned Ground Vehicles

“…Other efforts [20], [21] have investigated the interdependencies between the principal components of autonomous robots: 1) perception; 2) planning; and 3) task execution. However, it is still difficult to assess the effect of environmental variations on system performance given the scarcity of performance indicators.…”

“…However, it is still difficult to assess the effect of environmental variations on system performance given the scarcity of performance indicators. One effort [20] developed a probabilistic model that learns the interdependencies between the three above components based on learning a Bayesian network that identifies the coherence between performance indicators and system outputs. However, the approach requires subsystem performance indicators that were not available for our work, but are likely applicable to our proposed future work.…”