Summary

This chapter now uses four major technical subchapters. The first establishes software foundations, domain context and standards. The second explains middleware and autonomy architecture. The third addresses lifecycle, configuration management and supply-chain control. The fourth provides a dedicated V&V treatment linked directly to the book's V-model. This organisation reduces fragmentation and makes the chapter easier to read as a coherent contribution to the book on safe autonomous systems.

The main conclusion is that software safety is not demonstrated by final testing alone. It is built through traceable requirements, well-specified interfaces, controlled baselines, verified implementation artefacts, tested integration behaviour, validated operational scenarios and monitored deployment. Middleware deserves special attention because it shapes timing, data freshness, fault propagation, isolation and security. Configuration management is equally critical because the safety argument applies only to the exact software, data, tools and deployment configuration that were verified.

The key V&V challenges are systematic faults, distributed timing behaviour, middleware configuration risk, hardware-software interdependence, open-source and supplier dependencies, AI data and model uncertainty, continuous updates and incomplete scenario coverage. Effective V&V combines reviews, static analysis, unit testing, integration tests, timing analysis, fault injection, simulation, SIL/PIL/HIL, field validation, release audits and operational monitoring. Each method has limitations, so the assurance case must explain how the evidence collectively supports safe operation within the defined operational domain.

The next chapter builds on this software and middleware foundation by turning to the autonomy functions that depend most directly on it: perception, mapping and localisation. These functions transform raw sensor data into an operational understanding of the environment, estimate where the system is within that environment, and provide the spatial context needed for planning and control. The transition is therefore from the software architecture that enables safe data exchange, timing, configuration and integration to the algorithms and data pipelines that allow an autonomous system to sense, interpret and position itself reliably. From a V&V perspective, the next challenge is to show that these perception, mapping and localisation functions remain trustworthy under uncertainty, sensor limitations, environmental variation and real-time operational constraints.