A blackboard architecture for perception planning in autonomous vehicles

A blackboard architecture, used for the planning of perception actions to be executed by the local navigation system of an autonomous vehicle, is presented. Perception is not used to make a model of the world but rather is used as an aid to the navigation, to get localization clues from the environment and to detect obstacles or the approaching of dangerous terrain, through the detection of alarm beacons. Perception is also used to prepare and serve reflexive navigation and to execute some focusing of attention tasks for image acquisition in specified locations, related to surveillance tasks. A continuous and reactive planning approach, compatible with good real-time performance, is used to implement different navigation and perception strategies and to achieve an adaptive behavior which allows safe navigation in various unpredictable environment conditions. Perception planning is conceptually implemented as a cycle with two phases: a phase during which good perception opportunities, available in the near future, are checked, and an initial perception plan is composed; and an execution phase, during which the plan is monitored, detailed, and altered. The real-time alteration of parts of the plan is done while other parts are being executed, taking into consideration the outcome of previous actions