Using Agile Development Practices for Safety-Assured Machine Learning Models in Autonomous Systems

What is it about?

Machine learning (ML) has allowed autonomous cyber–physical systems to be applied across a broad range of applications. ML allows these robots and autonomous vehicles to learn, adapt, and operate with no human intervention. However, this autonomous operation poses a problem as we need to prove that autonomous systems are acceptably safe before we use them. Designers and engineers have traditionally used ‘Waterfall’ or V-model development lifecycles to develop safe systems. However, ML engineering requires iteration and adaptation. This requires enhanced lifecycles and methodologies, and needs to systematically integrate rigorous safety assurance practices with the ML development and operation activities. This paper introduce AgileAMLAS, a novel lifecycle, and comprehensive methodology for designing, developing and operating safety-assured autonomous systems which use ML. AgileAMLAS combines Agile software engineering, ML engineering, and a safety engineering framework. It uses iterative and incremental development. AgileAMLAS provides systematic step-by-step guidelines for designing, developing and operating autonomous systems. It combines DevOps and MLOps, with a framework for generating compelling safety cases. We have developed and refined AgileAMLAS on a recent set of projects that developed autonomous robots across a variety of domains.

Featured Image

Why is it important?

AgileAMLAS is adaptable and applicable across domains and applications. It introduces a novel, comprehensive and safety-assured methodology for MLOps. AgileAMLAS provides an end-to-end approach for developing autonomous cyber–physical systems. It tightly integrates Agile software development, safety assurance and ML. Through AgileAMLAS, we provide a guide for system designers, developers and safety experts to work collaboratively when developing autonomous cyber–physical systems.