This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Machine Learning considerations in the context of Automotive Functional Safety Requirements for Autonomous Vehicles
ISSN: 0148-7191, e-ISSN: 2688-3627
To be published on November 21, 2019 by SAE International in United States
Event: NuGen Summit
We are currently in the age of developing Autonomous Vehicles (AV). Never before in history, the environment has been as conducive as today for these developments to come together to deliver a mass produced autonomous car for use by general public on the roads. Several enhancements in hardware, software, standards and even business models are paving the way for rapid development of AVs, bringing them closer to production reality. Safety is an indispensable consideration when it comes to transportation products, and ground vehicle development is no different. We have several established standards. When it comes to Autonomous Vehicle development, an important consideration is ISO 26262 for, Automotive Functional Safety. Going from generic frameworks such as Failure Mode and Effects Analyses (FMEA) and Hazard and operability study (HAZOP) to Functional Safety, Safety of Intended Functionality, and Automotive Safety Integrity Levels specific is a natural progression. This, in specific to AV development context with a renewed perspective is the need of the hour. The fundamental assumption of a human driver being part of the vehicle, considered in these safety frameworks is now changing with the onset of AV development journey. Machine learning and more specifically deep learning techniques are coming of age, supported by relevant hardware capabilities, and large scale development enabled through skilled workforce coming into the AV development arena. Several of AV functionality such as sensing, perception, planning and control is dependent on Machine Learning and Deep Learning techniques. This study is intended to make a rudimentary assessment of incorporating functional safety measures in the tasks considered, such as sensing, perception and planning for AVs. A specific task in the area of Object and/or Event detection will be considered to present a closed loop recommendation for the Machine Learning Techniques. Results: We will have a recommended best practice outcome from the study that will be Object/Event independent, for the Machine Learning technique to handle with the consideration of Safety Limitations: The study will assume best in class sensor data inputs, and will make minimal considerations for sensor data quality Novelty: The study will give a fundamental thumb rule approach to incorporation of automotive functional safety considerations in Machine Learning techniques that may be extended to other scenarios