A General Architecture for Flexible Autonomous Systems

We describe an architectural approach to autonomous systems that not only provides generality and verifiability, but also flexibility through self-awareness and self-reconfiguration. This is an agent-based approach where the core rational agent oversees a range of feedback control systems (within a modular architecture) handling adaptive interaction with the system’s environment. The rational agent makes all high-level decisions and, crucially, is able to provide reasons for its choices. In addition, the agent is self-aware being able to assess activity within the system’s modular architecture, and is able to reconfigure this architecture to cope with changes, either in its hardware or in the environment. This architectural approach has been instantiated in a number of application areas, including autonomous satellites, convoys of driver-less cars, unmanned air vehicles, and autonomous nuclear decommissioning. In several of these, notably the first and last, the flexibility and resilience provided by self-reconfigurability is crucial.