François Terrier

Fabio Arnez, Huascar Espinoza, Ansgar Radermacher et al.

As autonomous systems increasingly rely on Deep Neural Networks (DNN) to implement the navigation pipeline functions, uncertainty estimation methods have become paramount for estimating confidence in DNN predictions. Bayesian Deep Learning (BDL) offers a principled approach to model uncertainties in DNNs. However, in DNN-based systems, not all the components use uncertainty estimation methods and typically ignore the uncertainty propagation between them. This paper provides a method that considers the uncertainty and the interaction between BDL components to capture the overall system uncertainty. We study the effect of uncertainty propagation in a BDL-based system for autonomous aerial navigation. Experiments show that our approach allows us to capture useful uncertainty estimates while slightly improving the system's performance in its final task. In addition, we discuss the benefits, challenges, and implications of adopting BDL to build dependable autonomous systems.

François Terrier

The explosion in the performance of Machine Learning (ML) and the potential of its applications are strongly encouraging us to consider its use in industrial systems, including for critical functions such as decision-making in autonomous systems. While the AI community is well aware of the need to ensure the trustworthiness of AI-based applications, it is still leaving too much to one side the issue of safety and its corollary, regulation and standards, without which it is not possible to certify any level of safety, whether the systems are slightly or very critical.The process of developing and qualifying safety-critical software and systems in regulated industries such as aerospace, nuclear power stations, railways or automotive industry has long been well rationalized and mastered. They use well-defined standards, regulatory frameworks and processes, as well as formal techniques to assess and demonstrate the quality and safety of the systems and software they develop. However, the low level of formalization of specifications and the uncertainties and opacity of machine learning-based components make it difficult to validate and verify them using most traditional critical systems engineering methods. This raises the question of qualification standards, and therefore of regulations adapted to AI. With the AI Act, the European Commission has laid the foundations for moving forward and building solid approaches to the integration of AI-based applications that are safe, trustworthy and respect European ethical values. The question then becomes "How can we rise to the challenge of certification and propose methods and tools for trusted artificial intelligence?"

Fabio Arnez, Huascar Espinoza, Ansgar Radermacher et al.

Uncertainty quantification methods are required in autonomous systems that include deep learning (DL) components to assess the confidence of their estimations. However, to successfully deploy DL components in safety-critical autonomous systems, they should also handle uncertainty at the input rather than only at the output of the DL components. Considering a probability distribution in the input enables the propagation of uncertainty through different components to provide a representative measure of the overall system uncertainty. In this position paper, we propose a method to account for uncertainty at the input of Bayesian Deep Learning control policies for Aerial Navigation. Our early experiments show that the proposed method improves the robustness of the navigation policy in Out-of-Distribution (OoD) scenarios.

Fabio Arnez, Huascar Espinoza, Ansgar Radermacher et al.

A key factor for ensuring safety in Autonomous Vehicles (AVs) is to avoid any abnormal behaviors under undesirable and unpredicted circumstances. As AVs increasingly rely on Deep Neural Networks (DNNs) to perform safety-critical tasks, different methods for uncertainty quantification have recently been proposed to measure the inevitable source of errors in data and models. However, uncertainty quantification in DNNs is still a challenging task. These methods require a higher computational load, a higher memory footprint, and introduce extra latency, which can be prohibitive in safety-critical applications. In this paper, we provide a brief and comparative survey of methods for uncertainty quantification in DNNs along with existing metrics to evaluate uncertainty predictions. We are particularly interested in understanding the advantages and downsides of each method for specific AV tasks and types of uncertainty sources.

Frédéric Thomas, Jérôme Delatour, François Terrier et al.

One foundation of the model driven engineering (MDE) is to separate the modelling application description from its technological implementation (i.e. platform). Some of them are dedicated to the system execution. Hence, one promise solution of the MDE is to automate transformations from platform independent models to platform specific models. Little work has explicitly described platform characteristics. Yet, an explicit modelling allows taking in account their characteristics more easily (par ex., performances, maintainability,portability). This paper presents both an execution platform modelling state of art and a pattern to describe execution platform modelling framework. It intends to confirm the feasibility and the interests in describing an execution platform metamodel.