The aerospace research community is working hard at developing UAV control technology thatrequires as little human supervision as possible. “Humans are expensive, let’s give the machines thecapability of making their own decisions - that’ll save us money” are cries heard across the globe.Such autonomy comes at a price - if one gives UAVs the capability of making decisions in a noisy,latent data world then one should expect them to make decisions you might not. If you give free-will why would one not expect it to be executed? (Let’s face it - human pilots don’t always actdeterministic, right?) “But that’s okay”, some muse, ‘we can test it before hand to insure it makesthe right decisions.” Wrong - the proposed complex distributed control architectures using largeblocks of complicated software will have too many paths to exhaustively test (assuming one findsthem all) affordably. If one adds in the non-determinism inherent in multi-intelligent agentnegotiation, biologically inspired control technology, and real-time adaptive systems it’s enough tomake any controls engineer run from the table screaming “Not only don’t I really know what’s in it, Idon’t really know what it will do!”. This paper examines the challenges in allowing UAVs to haveautonomy, while insuring they don't do anything stupid.
展开▼