Friday, September 30, 2005

Paper: XOMO: Understanding Development Options for Autonomy

(With Julian Richardson)

Autonomy is a key capability for future space exploration, but uncertain risks in the development and deployment of autonomy systems create a barrier to its use.

Our broad goal is to remove this barrier by developing a model which can quantify autonomy risk, and developing tools and techniques which analyze this and other models, e.g. COCOMO-II, to determine actions which can be taken during project development to improve project outcome, for example reducing cost and/or risk.

In this paper, we focus on the second of these --- the analysis of models to determine beneficial project actions. We demonstrate that the trade space of project development options can be analyzed by combining XOMO --- a general framework we have developed for Monte Carlo simulation of COCOMO-like models --- with data mining (in particular, treatment learning), to determine those actions which most improve project outcome. In this paper, we use XOMO to simulate development options and measure their effects according to three models: the COCOMO effort estimation model, the COQUALMO defect model, and Ray Madachy's schedule risk model.

In a sample case study, the combination of XOMO and treatment learning finds process options which halve the mean development effort, halve the mean risk of schedule over run, reduce the mean defect density by 85%, and significantly reduce the variance on the above measures.


To appear, 20th International Forum on COCOMO and Software Cost Modeling, October 25-28, 2005, Los Angeles, California.

Available on-line at http://timmenzies.net/pdf/05xomo.pdf.

No comments: