Connected automated vehicles (CAVs) are—compared conventional vehicles—expected to provide more efficient, accessible, and safer transport solutions in on-road use cases as well as confined areas such as mines, construction sites or harbours. As development of such vehicles has proved more difficult than anticipated, especially when it comes to ensuring safety, more cautious strategies for introduction are now being pursued. An approach where new automated features are initially released with more basic performance to enable successful safety assurance, followed by gradual expansion of performance and number of use-cases using an iterative development process as the confidence in the solution increases, e.g., due to more available field data, improved machine learning algorithms, or improved verification, is highly interesting. Hence a key research question targeted by the SALIENCE4CAV project was: How to ensure the safety of CAVs while enabling frequent updates for automated driving systems with their comprising elements? Today, many of the used methods and practices for safety analysis and safety assurance are not adequate for continuous deployment. In addition, the project has investigated several open questions raised by the predecessor project ESPLANADE and from needs identified by the industry partners; this includes how to handle safety assurance for machine learning components, use of quantitative risk acceptance criteria as a key part of the safety argument, safety for collaborative CAVs including use in mixed traffic environments, the role of minimal risk manoeuvres, and interaction with human operators.

Some key results are: investigation of safety assurance methods and gaps with regards to frequent updates and other challenges for CAV safety assurance; use of safety contracts as an enabler for continuous integration, continuous deployment and DevOps; a method for human interaction safety analysis; application of the principle of precautionary safety for meeting a quantitative risk norm and using field data for continuous improvements; definition of classes of cooperative and collaborative vehicles and their respective characteristics and definition of minimal risk manoeuvre and minimal risk condition strategies for individual, cooperative and collaborative vehicles; use of out-of-distribution detection for safety of machine learning; a simulation-aided approach for evaluating machine learning components; and methods for variational safety using high-dimensional safety contracts.

The SALIENCE4CAV project ran from January 2021 to December 2023 with the partners Agreat, Comentor, Epiroc Rock Drills, KTH Royal Institute of Technology, Qamcom Research and Technology, RISE Research Institutes of Sweden, Semcon Sweden, Veoneer (during the project acquired by Magna) and Zenseact. Coordination was done by RISE.

This final report is a summary of the project results and contains summaries of content from the project deliverables and publications.