Critical embedded systems are deployed in several industrial application sectors such as transportation (avionics, automotive, railways ...), industry 4.0, space, healthcare, energy, smart cities, etc. Such systems are required to deliver trustworthy, resilient and timely services while achieving high levels of performance at the lowest costs. Achieving such requirements requires the development of innovative solutions to address various challenges raised today. We can mention first the need to investigate the impact of the new technological trends that include the advent of multi-core, many core and GPU processor architectures, as well as virtualization and edge computing techniques with the development of new services based on a more advanced distribution of processing and storage on remote infrastructures, using in particular Internet of Things (IoT) and advanced communication technologies (5G, etc.). These evolutions have led to increasing the complexity of the systems, as well as the risks of failures or degradation of performances that result from accidental causes or malicious attacks.

Another major challenge is related to the growing interest in integrating artificial intelligence algorithms into mission-critical applications in order to achieve increasing levels of autonomy, flexibility and adaptation to changes and unforseen events. Although machine learning algorithms have demonstrated high accuracy for specific tasks such as  object recognition and classification, the current state of knowledge does not allow yet to provide the minimum levels of guarantees and justified confidence required for the certification of systems integrating such algorithms for mission critical domains. Indeed, machine learning (ML) algorithms are typically based on the assumption that the training dataset is representative of the inputs that the system will face during its deployment. However, in practice there are a wide variety of unexpected accidental, as well as adversarially-crafted, perturbations on the ML inputs that might lead to violations of this assumption. Furthermore, ML algorithms are often run on special-purpose hardware accelerators, which may themselves be subject to faults. A major goal of current research is to develop innovative solutions  for  the problem of establishing guarantees of reliability, security, safety, robustness and better explainability for systems that incorporate increasingly complex ML models, and for the challenge of determining whether such systems can comply with requirements for safety-critical systems. This includes for example defining new paradigms for the design, verification and systems level assurance of embeddable and certifiable AI architectures as well as the design of new  processor chips to be used for AI critical applications.

Cybersecurity is another challenging area where significant scientific advances are needed to cope with the increase, in number and sophistication, of malicious threats targeting critical infrastructures and systems. Vulnerabilities span all system layers, and increasingly the lower layers and the hardware. The Spectre and Meltdown vulnerabilities identified in Intel processors are recent examples. Hardware assisted protection techniques and trustworthy embedded components are among the promising solutions being explored, together with new paradigms that consider security and safety holistically.  In addition, personal data protection and privacy have also become a real industrial and societal concern due to the massive collection of data and the requirement to demonstrate compliance with the GDPR.

We are looking forward to welcoming you in Toulouse in March 2022 for the 11^th edition of the ERTS Congress to exchange about these topics and your recent contributions!