Autonomy Software: V& V Challenges and Characteristics

Abstract

Abstract-The successful operation of unmanned air vehicles requires software with a high degree of autonomy. Only if high level functions can be carried out without human control and intervention, complex missions in a changing and potentially unknown environment can be carried out successfully.Autonomy software is highly mission and safety critical: failures, caused by flaws in the software cannot only jeopardize the mission, but could also endanger human life (e.g., a crash of an UAV in a densely populated area). Due to its large size, high complexity, and use of specialized algorithms (planner, constraint-solver, etc.)g autonomy software poses specific challenges for its verification, validation, and certification.---we have carried out a survey among researchers aid scientists at NASA to study these issues. In this paper, we will present major results of this study, discussing the broad spectrum.of notions and characteristics of autonomy software and its challenges for design and development. A main focus of this survey was to evaluate verification and validation (V&V) issues and challenges, compared to the development of "traditional" safety-critical software. We will discuss important issues in V&V of autonomous software and advanced V&V tools which can help to mitigate software risks. Results of this survey will help to identify and understand safety concerns in autonomy software and will lead to improved strategies for mitigation of these risks.

Certification and Software Verification Considerations for Autonomous Unmanned Aircraft

“…It is acknowledged that such software is complex and especially adaptive, and it is an extreme case for conventional V&Vapproaches; but with formal techniques (run-time monitoring, static analysis, model checking, and theorem proving), there exist sufficient new methodologies for the V&Vof autonomous software, although the required level of expertise to master these techniques increases significantly. In a NASA case study, the specific characteristics of autonomy are described and current V&V techniques are analyzed [21]. The results showed that current techniques can be used to find errors in such software, but conventional V&V techniques focus mainly on finding implementation and integration errors.…”

“…Conventional testing methods can and should be used, but in practice for autonomy functions, they are reaching their limits. Tests should not only be performed on a functional level but also on a behavior level [21]. New, more formal, V&V methodologies [20] promise to cope with the growing software complexity, but they are difficult to handle, and although there have been some studies, the industry has still to gather more experience with these techniques.…”

Model-Based Verification and Validation of Safety-Critical Embedded Real-Time Systems: Formation and Tools

“…Autonomous, intelligent and adaptive ERT control algorithms VV&T is a major challenge for uninhabited aerial vehicles (UAVs), automobiles, aircrafts, satellite vehicles, and space robots certification [23][24][25][26][27].…”

Model-Based Product-Oriented Certification

“…As stated in the Columbia Accident Investigation Board Report [3], the inability to thoroughly apply the required certification protocols had been determined to be a contributing factor to the loss of STS-107, Space Shuttle Columbia. Schumann and Visser's discussion in [25] suggests that the current process-oriented certification methodologies are prohibitively expensive for systems of such complexity. A detailed analysis by Lowry [18] indicates that at the present moment the certification cost of mission-critical space software exceeds its development cost.…”