Long-term autonomy is the problem of keeping a robot working in a changing, partly observable, open world for weeks, months or years, rather than for one run in a fully modelled static environment. Students learn how six areas of AI each change when the time horizon is long: mapping becomes a never-ending process that must decide what to remember and what to forget; perception must separate persistent structure from movable objects and keep learning new categories after deployment; knowledge representation must reason over time and detect the unfamiliar; planning must handle resources, uncertainty and competing goals under an executive that recovers from failure; interaction must build per-user models; and learning must happen during operation, largely without supervision, coupled to exploration. Students also learn to read a fielded robot system as an integration problem, judging deployments by duration and by which AI areas are actually integrated rather than by component benchmarks.
Long-term autonomy (LTA) is a robot system operating in complex, changing, only partly observable real-world settings for weeks, months or years, rather than for single runs in a fully modelled stati…