Machine Learning-Augmented Distributed Sensor Fusion for Autonomous Navigation in Confined Spaces

In the UK, much of the infrastructure, including tunnels, railways, bridges, mines, and pipelines, has been in service for over a century . Time and weather have caused this infrastructure to deteriorate. Maintaining this infrastructure requires long-term planning to prevent disruptions that could impact communities and businesses. Many infrastructure inspection tasks currently rely on non-autonomous robot platforms. These robots face challenges when navigating complex environments where GNSS signals are unavailable to support localization. The inertial measurement units (IMUs) used for autonomous localization on small to medium-sized inspection robots often lack the necessary accuracy for reliable positioning in these confined spaces. Additionally, existing robot navigation methods depend on isolated sensor data which can be susceptible to inaccuracies and noise. The sensor fusion techniques currently in use also do not effectively adapt to the changing surroundings which limit their effectiveness in complex infrastructure scenarios. This research aims to address these limitations and fill the gap by developing advanced distributed state space sensor fusion models that can better represent the dynamic behaviour of inspection robots. By using machine learning techniques and enabling data sharing among robot teams, the research seeks to improve the autonomous navigation capabilities of these platforms in confined environments where GNSS is unavailable.