Test and Evaluation of Reinforcement Learning via Robustness Testing and Explainable AI for High-Speed Aerospace Vehicles

Raz, Ali K.; Nolan, Sean M.; Levin, Winston C.; Mall, Kshitij; Mia, Ahmad; Mockus, Linas; Ezra, Kris; Williams, Kyle A.

doi:10.1109/aero53065.2022.9843563

Cited by 6 publications

(2 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This section presents a conceptual example for the T&E of an aircraft flight control system using classical control system methods and contrast it with a ML model, in particular Reinforcement Learning (RL), which has been shown to accomplish a similar capability in a variety of control related problems (Raz et al 2022). Figure 4 contrasts the two implementations by extending the control of an aircraft to compensate for random disturbances, such as wind gusts.…”

Section: Conceptual Example Of Test and Evaluation With And Without D...mentioning

confidence: 99%

Explainable AI and Counterfactuals for Test and Evaluation of Intelligent Engineered Systems

Raz,

Miller,

Chang

et al. 2023

INCOSE International Symp

Self Cite

View full text Add to dashboard Cite

Systems Engineering (SE) based test and evaluation (T&E) approaches have proven crucial for successful realization of most modern‐day complex systems and system‐of‐systems solutions. The design, test, and evaluation of engineered systems with many technologically advanced and complex components hinges on well‐structured integration and life cycle evolution process models, clearly defined requirements, along with controllability, observability, and stability (COS) of components that transcend to the system‐level. However, integration of machine learning (ML), deep learning (DL) and artificially intelligent (AI) components invalidates some SE foundations as the DL algorithms are primarily data‐driven with opaque decision‐making constructs. As a result, the current SE T&E approaches, although necessary, have become insufficient to evaluate the adoption of ML/DL/AI methods in engineered systems. This paper proposes two new approaches—explainable AI (xAI) and counterfactual testing and evaluation (cT&E)—as addition to the SE tool set for T&E of intelligent engineered systems. An example of SE T&E considerations is provided based on a conceptual aircraft control system implementation comparing as a classical controller and a DL‐based reinforcement learning controller.

show abstract

Section: Conceptual Example Of Test and Evaluation With And Without D...mentioning

confidence: 99%

Explainable AI and Counterfactuals for Test and Evaluation of Intelligent Engineered Systems

Raz,

Miller,

Chang

et al. 2023

INCOSE International Symp

Self Cite

View full text Add to dashboard Cite

show abstract

“…The authors note that this paper is a significantly revised version of a conference paper that was published as part of the 2022 IEEE Aerospace conference proceedings [16]. In particular, the changes introduced include: 1) a re-organized T&E framework (Section III) to establish logical connections between the three steps, highlight their synergy, and explain how each one collectively aids the T&E of RL; 2) an expanded literature review of SHAP with an RL example; 3) an added discussion of the benefits of RL to high-speed aerospace system problems; 4) the addition of new results for XAI with SHAP along with new dependency plots and related discussion, 5) a new test case considering RL response to unmodeled environmental disturbances; and 6) added discussion of the new results to highlight the connections between the steps of the T&E framework and underscore how they collectively help to explain our RL problem exemplar.…”

Section: Introductionmentioning

confidence: 99%

Explainable AI and Robustness-Based Test and Evaluation of Reinforcement Learning

Raz,

Mall,

Nolan

et al. 2024

IEEE Trans. Aerosp. Electron. Syst.

View full text Add to dashboard Cite

Reinforcement Learning is a powerful and proven approach to generating near-optimal decision policies across domains, though characterizing performance boundaries, explaining decisions, and quantifying output uncertainties are major barriers to widespread adoption of Reinforcement Learning for real-time use. This is particularly true for high-risk and safety-critical aerospace systems where the cost of failure is high and performance envelopes for systems of interest may be small. To address these issues, this paper presents a three-part Test and Evaluation framework for Reinforcement Learning which is purpose-built from a systems engineering perspective on artificial intelligence. This framework employs Explainable AI techniques-namely Shapley Additive Explanations-to examine opaque decision-making, introduces robustness testing to characterize performance bounds and sensitivities, and incorporates output validation against accepted solutions. In this paper, we consider an example problem of a high-speed aerospace vehicle emergency descent problem where a Reinforcement Learning agent is trained to control vehicle angle of attack (AoA). Shapley Additive Explanations expose the most significant features that impact the selection of AoA command while robustness testing characterizes the acceptable range of disturbances to in flight parameters the trained vehicle can accommodate.

show abstract

Sensor Tasking for Space Situation Awareness: Combining Reinforcement Learning and Causality

Fan

Chang

Raz

et al. 2023

2023 IEEE Aerospace Conference

View full text Add to dashboard Cite

Test and Evaluation of Reinforcement Learning via Robustness Testing and Explainable AI for High-Speed Aerospace Vehicles

Cited by 6 publications

References 23 publications

Explainable AI and Counterfactuals for Test and Evaluation of Intelligent Engineered Systems

Explainable AI and Counterfactuals for Test and Evaluation of Intelligent Engineered Systems

Explainable AI and Robustness-Based Test and Evaluation of Reinforcement Learning

Sensor Tasking for Space Situation Awareness: Combining Reinforcement Learning and Causality

Contact Info

Product

Resources

About