Run-Time Assurance for Learning-Enabled Systems

Cofer, Darren; Amundson, Isaac; Sattigeri, Ramachandra; Passi, Arjun; Boggs, Christopher; Smith, Eric; Gilham, Limei; Byun, Taejoon; Rayadurgam, Sanjai

doi:10.1007/978-3-030-55754-6_21

Cited by 18 publications

(6 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There has been considerable previous work done in the area of formal methods for assurance [9], [10], [11]. In [9] the research discusses a method to perform run-time assurance for learning systems with an assurance architecture designed in Architecture Analysis and Design Language (AADL) and formal contracts for each of the components modeled and verified in Assume Guarantee (AGREE) annex. In [12] Davis discusses an approach to use architectural analysis to prove that the protocol designed for multi-agents satisfies the specified properties.…”

Section: A Formal Methods or Assurance Methodsmentioning

confidence: 99%

Can Model Checking Assure, Distributed Autonomous Systems Agree? An Urban Air Mobility Case Study

Gupta¹,

Bhattacharyya²,

Vadivel³

2020

IJACSA

View full text Add to dashboard Cite

Section: A Formal Methods or Assurance Methodsmentioning

confidence: 99%

Can Model Checking Assure, Distributed Autonomous Systems Agree? An Urban Air Mobility Case Study

Gupta¹,

Bhattacharyya²,

Vadivel³

2020

IJACSA

View full text Add to dashboard Cite

“…Within the domains of cyberphysical systems and aerospace, in particular, there is a significant number of problems where ML can be applied to applications possessing actionable specifications. 15 One instance is a battery management system. We know how to construct battery management systems via conventional means and hence can construct actionable specifications that can be used to construct test oracles and properties to check at runtime.…”

Section: A Path Forwardmentioning

confidence: 99%

Assuring Safety-Critical Machine Learning-Enabled Systems: Challenges and Promise

Goodloe

2023

Computer

View full text Add to dashboard Cite

afety-critical systems, such as automobiles and medical devices, are systems whose failure could result in loss of life, significant property damage, and damage to the environment. The grave consequences of failure have compelled industry and regulatory authorities to adopt conservative design approaches and exhaustive verification and validation (V&V) procedures to prevent mishaps. In addition, strict licensing requirements are often placed on human operators of many safety-critical systems. Ultracritical systems, such as civil transport aircraft and nuclear power plants, are safety-critical systems that society deems should never suffer a catastrophic failure during their operating lifetime and whose development is subject to particularly strict regulatory constraints and rigorous operator training requirements. In practice, the V&V of avionics and other ultracritical software systems

show abstract

“…Thus, the generalization and non-deterministic behavior problems could be tackled directly. In this line, [29] presents a RTA architecture based on a NN aircraft taxiing application showing how RTA could be used to ensure safe operation.…”

Section: Run-time Assurance (Rta)mentioning

confidence: 99%

Machine Learning Based Visual Navigation System Architecture for Aam Operations with A Discussion on its Certifiability

Escudero

Costas

Hardt

et al. 2022

2022 Integrated Communication, Navigation and Surveillance Conference (ICNS)

View full text Add to dashboard Cite

Advanced Air Mobility (AAM) is expected to revolutionize the future of general transportation expanding the conventional notion of air traffic to include several services carried out by autonomous aerial platforms. However, the significant challenges associated with such complex scenarios require the introduction of sophisticated technologies able to deliver the resilience, robustness, and accuracy needed to achieve safe, autonomous operations [39]. In this context, solutions based on Artificial Intelligence (AI), able to overcome some limitations found in traditional approaches, are becoming a major opportunity for the aviation industry, but, at the same time, a significant challenge with respect to the certification standards.With the focal point on further proposing a certifiable architecture for AI-enhanced vision navigation in AAM operations, this paper first, summarizes the current technologies and fusion methods applied to date to navigation purposes, to later address the certification problem. Regarding certification, it explores three specific points: 1) traditional certification procedures; 2) current status of AI homologation recommendations; and 3) other certification factors to be considered for future discussion.

show abstract

Run-Time Assurance for Learning-Enabled Systems

Cited by 18 publications

References 4 publications

Can Model Checking Assure, Distributed Autonomous Systems Agree? An Urban Air Mobility Case Study

Can Model Checking Assure, Distributed Autonomous Systems Agree? An Urban Air Mobility Case Study

Assuring Safety-Critical Machine Learning-Enabled Systems: Challenges and Promise

Machine Learning Based Visual Navigation System Architecture for Aam Operations with A Discussion on its Certifiability

Contact Info

Product

Resources

About