The role of interoception in the assurance of autonomous systems

An automated anomaly detection and fault diagnostic capability onboard an autonomous system reduces the possibility of it becoming a hazard to other systems and the environment. For instance, an underwater glider with power system failure or diving actuator failures can become an obstacle to other sea users, leading to increased collision risks with other systems (e.g. surface vehicles). A robust interoceptive capability onboard an autonomous system limits the possibility of itself becoming a hazard to a minimum by providing an immediate warning signal to the vehicle’s safety and control modules. A remotely operated diagnostic system on the contrary, albeit more energy-efficient, is reliant on processing delayed decimated data transferred over satellite communication before being able to respond with appropriate preventive actions. The role of interoception is particularly significant for systems deployed over long-term remote missions where communication to the base station may be limited and where human intervention is challenging.

Additionally, an onboard system reduces the requirement for extended involvement of human experts. Diagnostics of problems can be limited by the experience of individual operators and the remedies are subject to human error. Therefore, reduced dependence on human expertise provides a higher level of safety and assurance of autonomous systems, provided that the interoception capability is robust in detecting critical failures. In the case of underwater gliders, which are typically deployed for remote missions over a long period of time, the current monitoring guidelines require the pilots’ attention and to be present around the clock. This limits the scale of observational fleets that can be deployed simultaneously. Additionally, less reliance on human presence can significantly reduce the operational costs and enable larger-scale simultaneous deployments of multiple autonomous systems.

Overall, generalised robust interoception systems onboard the autonomous systems can provide increased confidence in new autonomous technologies, increase in outputs, reduction in capital losses and operational costs, and even greater adoption of the new technology. In addition, such systems can also inform and improve the design and management of robotic and autonomous systems. For systems such as underwater gliders that are typically deployed for long remote missions, an automated and generalised onboard diagnosis protocol for adverse system behaviour through a real-time intelligent anomaly detection and fault diagnostics system can achieve a higher level of assurance; such that informed control actions can be made by the system’s control and decision-making modules. Moreover, the records of the interoception system can be applied as a basis to update the system itself in the next design iteration, leading to even higher assurance of the autonomous systems.

Dr Yuanchang Liu
Lecturer in Autonomy
University College London

PI of the ALADDIN project