Validation of an Autonomous Adaptive Safety-Critical System: Lessons Learned
The functionality of adaptive systems evolves over time, as they improve their performance by online learning. Through judicious learning, a deployed system may be able to react to situations that were never identified and analyzed by the designer. Online adaptive systems are attracting increasing attention in application domains where autonomy is an important requirement. Long term space missions, where communication delays to ground stations are prohibitively long, and flight control systems, which deal with a wide range of environmental factors, are among the typical application domains.
Traditional software validation techniques cannot guarantee safe behavior of online adaptive systems. We will discuss challenges that this type of systems present for software verification and validation experts. Furthermore, we will present a validation methodology developed in the context of NASA Intelligent Flight Control Systems program. This methodology includes a flexible failure detection scheme and stability analysis of a learning algorithm based on Lyapunov theory. Even though our case study is very specific, the theoretical foundation of the presented validation methodology makes it generally applicable to a wide range of online adaptive systems with embedded soft-computing components.
Bojan Cukic is an Associate Professor at the Lane Department of
Computer Science and Electrical Engineering,
Dr. Cukic served as the Program Committee co-chair for the 14th IEEE International Symposium
on Software Reliability Engineering (ISSRE
2003) and 8th IEEE International
Symposium on High Assurance Systems Engineering (HASE 2004). Up until
recently, he served as WVU research lead at the NASA IV&V facility in