Towards Assurance Cases for Resilient Control Systems

James Weimer, Oleg Sokolsky, Nicola Bezzo, Insup Lee

TL;DR

This paper tackles the challenge of building convincing arguments (assurance cases) that complex, resilient control systems for autonomous vehicles are safe and reliable. They propose a model-manipulation strategy that links evidence from different levels of abstraction, from first-principles physics to software implementation, using a case study of a resilient speed estimator. This approach aims to provide regulators and stakeholders with a clear, evidence-backed justification for system guarantees.

What Problem Does This Paper Address?

As cyber-physical systems (CPS) like autonomous vehicles become more complex, ensuring their resilience to unpredictable conditions (faults, attacks, environmental changes) is paramount. This requires rigorous design and verification, generating a wealth of evidence (requirements, test results, design reviews). However, organizing and presenting this evidence to demonstrate that the system actually meets its safety and resilience guarantees is a significant challenge. Assurance cases are a proposed solution, but there's a lack of clear methods for building them, especially for resilience properties that span multiple levels of system design and reasoning. This paper addresses this gap by demonstrating how to construct an assurance case for a specific component – a resilient speed estimator – by systematically linking evidence across different modeling and implementation stages.

Methods

The core of the paper's contribution is a model-manipulation strategy for constructing assurance cases. This strategy is visualized in Figure 2 and works by making a claim about applying an analysis algorithm or transformation to a given model. This claim is then broken down into two sub-claims:

  1. Technique sub-claim: This argues that the technique itself is sound (e.g., the algorithm is proven correct). Evidence here comes from proofs or tool qualifications.
  2. Model sub-claim: This argues that the model being manipulated is suitable and accurate. This sub-claim can be recursively argued using the same model-manipulation strategy.

This iterative approach allows for a chain of reasoning from high-level, first-principles models down to the specific software implementation. The paper also leverages existing notations like Goal-Structuring Notation (GSN) for representing arguments graphically.

Datasets

Not discussed in the paper. The paper focuses on a case study using a specific robotic platform (LandShark) and its associated sensor data and models.

Key Insights and Ideas

  • Bridging the Abstraction Gap: The most significant insight is the systematic way the paper connects evidence from different levels of abstraction. They show how to argue that a discrete-time model accurately represents a continuous-time model, and how that, in turn, is derived from first-principles physics, ultimately leading to the software implementation.
  • Evidence for Resilience: The paper demonstrates how to build an assurance case for a resilience property (bounded speed estimation under attack), not just basic functionality. This involves explicitly modeling potential attacks and arguing that the system remains robust.
  • Model Suitability is Key: The authors emphasize that the assurance case isn't just about proving algorithms are correct, but also about proving that the models used in the design and analysis are accurate and suitable representations of reality.
  • Iterative Argumentation: The recursive nature of the model-manipulation strategy is crucial. It allows for breaking down complex claims into manageable, evidence-backed sub-claims, building confidence step-by-step.

Key Results

The paper presents a detailed assurance case for the LandShark resilient speed estimator. While quantitative results are not the primary focus, the paper demonstrates:

  • A structured argument that the resilient speed estimator (RSE) algorithm computes bounded speed estimates, even under sensor attacks.
  • A chain of reasoning that links the RSE algorithm's mathematical model to a reduced-order continuous-time model, then to a full-order continuous-time model derived from first-principles physics of the LandShark vehicle.
  • An argument for the correct implementation of the RSE algorithm on the LandShark platform, including its deployment within the ROS middleware.
  • The identification and justification of various assumptions, including attack models, environmental assumptions, and platform assumptions.

Limitations and Future Work

  • Case Study Specificity: While the authors believe the modeling levels and assumptions are typical, the case study is specific to a speed estimator on a ground vehicle. Generalizing this approach to other CPS components or system types might require further investigation.
  • Bisimulation Analysis: The paper mentions that approximate bisimulation analysis was not performed for the reduced-order model in their case study, but rather dominant eigenvalues were identified. This suggests that a more rigorous bisimulation analysis could strengthen the argument.
  • Integration into Larger Cases: The assurance case presented is for a single module. The paper acknowledges that this will be part of a larger assurance case for the whole vehicle, and the integration of such modular cases is a future direction.
  • Automation: While tools like CVXGEN are used, the overall assurance case construction is still largely manual. Future work could explore more automated methods for generating and verifying assurance cases.

Summary updated

Abstract

The paper studies the problem of constructing assurance cases for embedded control systems developed using a model-based approach. Assurance cases aim to provide a convincing argument that the system delivers certain guarantees, based on the evidence obtained during the design and evaluation of the system. We suggest an argument strategy centered around properties of models used in the development and properties of tools that manipulate these models. The paper presents the case study of a resilient speed estimator for an autonomous ground vehicle and takes the reader through a detailed assurance case arguing that the estimator computes speed estimates with bounded error.
BibTeX
@inproceedings{weimer2014towards,
  author = {James Weimer and Oleg Sokolsky and Nicola Bezzo and Insup Lee},
  title = {{Towards Assurance Cases for Resilient Control Systems}},
  booktitle = {CPSNA},
  year = {2014},
  doi = {10.1109/cpsna.2014.19},
}

Loading related content...

Figure