The Challenges of Sensing and Repairing Software Defects in Autonomous Systems
Abstract
A small study investigated the potential benefits and research challenges related to the software components of future space vehicle designs. The study identified two potential research thrusts aimed at improving the resilience and reliability of software deployed on space vehicles: (1) improving software resiliency through proactive diversity and (2) reducing costs and schedule overruns through automated software repair. Both thrusts rely on recently developed technology known as GenProg. GenProg uses genetic programming (GP), an iterated stochastic search technique, to search for program repairs. The search space of possible repairs is infinitely large, and GenProg employs five strategies to render the search tractable: (1) coarse-grained, statement-level patches to reduce search space size; (2) fault localization to focus edit locations; (3) existing code to provide the seed of new repairs; (4) fitness approximation to reduce required test suite evaluations; and (5) parallelism to obtain results faster. The study focused on automated software transformations for repair and resiliency, because there is extensive prior work on the related topics of anomaly detection, intrusion detection and fault isolation, which could also be adapted to software in the space vehicles domain.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 09, 2014
- Accession Number
- ADA602024
Entities
People
- Stephanie Forrest
- Weimer
- Westley
Organizations
- University of New Mexico