Design, development, and validation of a software architecture for high performance autonomous robotics

Abstract

Autonomous systems are progressively taking more space in the global tech market. A broad variety of scenarios and challenges have been solved by both the industry and the scientific community. However, the path toward the mass diffusion of fully-autonomous robots still presents several challenges like social acceptance, law regulation, certifications, and so on. One powerful tool to boost the development and demonstrate the capabilities of state-of-the-art autonomous systems is the emerging field of high-performance robotics. With the aim of bringing autonomous platforms to their operational limit, high-performance robotics represents a unique chance to push the development of software and hardware to new levels of performance and reliability.This thesis proposes a software integration framework for high-performance autonomous robotics systems. The aim of the presented solution is to support the development and the evolution of autonomous systems under severely uncertain conditions, with limited development resources, preserving the dependability, evolvability, and performance of the autonomy stack. Besides the abstract specification of the integration framework components, the requirement analysis process that drove the design from high-level software qualities to functional solutions is presented. Also, we present as a case study for the presented solutions a realization of the framework and of an autonomy stack based on it. The studied software stack was adopted by team KAIST as control software for their Dallara AV-21 autonomous racing car in the context of the two robotics competitions Indy Autonomous Challenge 2021 and Autonomous Challenge at CES 2022. The developed system is analyzed from a qualitative and quantitative perspective against the software qualities that have been identified as top priorities for this class of systems.