The long-term viability of the SAFEXPLAIN approach will be demonstrated by integrating its solutions in a commercial toolset for system testing, and applying its principles to several mixed-criticality case studies relevant for European society. SAFEXPLAIN will focus on state-of-the-art and representative mixed-criticality case studies from the automotive, railway and space domains.
Considering real and representative case studies in each domain will allow the project to capture the real challenges that can emerge from the application of the techniques in industrial-size projects. Selected applications will bring domain-specific challenges and different aspects of mixed-criticality execution, such as scheduling, functional behaviour, isolation properties and timing isolation to be tackled by the SAFEXPLAIN methodology and toolchain.
In all case studies, the results of DL software will be used for CAIS safety-related functions. SAFEXPLAIN selected technical contributions will undergo a review by Certification Experts (CE) in the automotive (ISO 26262, SOTIF), space (ECSS) and railway (EN 5012x) domains. All three case studies will follow the same procedure: (1) stubbing, (2) preparation, porting and integration and (3) evaluation and assessment.
In the context of Automatic Train Operation (ATO), the completely autonomous operation of trains, this case study will check the viability of a safety architectural pattern composed by: DL artificial vision software elements that serve as “sensors” to provide information to safety-related software elements. These safety-related software elements, in conjunction with information coming from the DL software elements and possibly additional sensors, implement the safety function. The case will implement one or more functions such as detection, location and distance estimation of persons and obstacles in the track or around train doors to prevent the train from running people over, colliding with obstacles, or injuring passengers when opening/closing train doors autonomously, etc.
The space case study envisions the use of state-of-the-art mission autonomy and artificial intelligence technologies to enable fully autonomous operations during space missions.
The current approach to space mission operations involves extensive use of ground operators to perform activities such as mission planning, telemetry monitoring, payload data analysis and failure mitigation.
The automotive case study will consider the Apollo open-source project from Baidu. Apollo is an end-to-end industrial and practically implemented project with over 120 industrial partners, many of whom are top-tier AI companies and car manufacturers. Apollo is already deployed on a variety of prototype vehicles (including autonomous trucks) and supports state-of the-art hardware such as latest LIDARs and cameras from Velodyne and other vendors, as well as GPU acceleration. Apollo employs and implements the latest DL-based approaches for its Perception and Prediction modules, which are the most critical modules. Perception is responsible for identifying the surrounding area around the autonomous car, whereas, the Prediction module anticipates the future motion trajectories of perceived obstacles/objects.