Towards Functional Safety Compliance of Recurrent Neural Networks Davide Bacciu 1 , Antonio Carta 1 , Daniele Di Sarli 1 , Claudio Gallicchio 1 , Vincenzo Lomonaco 1 , Salvatore Petroni 2 {davide.bacciu, antonio.carta, daniele.disarli, claudio.gallicchio, vincenzo.lomonaco}@unipi.it 1 , salvatore.petroni@marelli.com 2 1 Department of Computer Science, University of Pisa, Pisa, Italy, 2 Legal and Compliance Department, Marelli Europe S.p.A., Turin, Italy Abstract. Deploying Autonomous Driving systems requires facing some novel challenges for the Automotive industry. One of the most critical aspects that can severely compromise their de- ployment is Functional Safety. The ISO 26262 standard provides guidelines to ensure Functional Safety of road vehicles. However, this standard is not suitable to develop Artificial Intelligence based systems such as systems based on Recurrent Neural Networks (RNNs). To address this issue, in this paper we propose a new methodology, composed of three steps. The first step is the robustness evaluation of the RNN against inputs perturbations. Then, a proper set of safety measures must be defined according to the model’s robustness, where less robust models will re- quire stronger mitigation. Finally, the functionality of the entire system must be extensively tested according to Safety Of The Intended Functionality (SOTIF) guidelines, providing quantitative re- sults about the occurrence of unsafe scenarios, and by evaluating appropriate Safety Performance Indicators. Keywords: Functional Safety, Dependability, Recurrent Neural Networks, Autonomous Driv- ing, Safety Performance Indicators. Nowadays artificial intelligence technologies are increasingly in demand for applications in several domains, ranging from gaming, finance, e-commerce, medicine, social media, education, home automation, entertainment, automotive, and others. The type of application in which the in- telligent system is used determines additional properties that the latter must respect. If we consider intelligent systems that perform object recognition on autonomous vehicles, we shall take into ac- count that the system shall comply with the Functional Safety for road vehicles [1] because without safety assurance the system can cause physical injuries or damage to the health of persons. For this reason, it is important to formally define the design process and the building blocks of safe intel- ligent systems, in order to develop such systems according to requirements imposed by the proper functional safety technological standard. Camera-ready version