Reactive Sensing for Autonomous Mobile Robots

This article presents an overview of our ongoing research in reactive sensing. Reac-tive sensing allows sensing processes at both the behavioral and deliberative levels to act as autonomous specialists in maintaining the most robust and eecient perception for the given behavior, state of the robot, and state of the world. Two basic reactive sensing mechanisms are presented. One is a method for identifying and employing a satisfactory recovery strategy after a sensing failure. The recovery-centered method shows an improvement of between 39% and 99.4% over previous methods. The second mechanism is an alternative to Dempster's rule of combination of belief used to op-portunistically reduce sensing without sacriicing certainty. The method dynamically projects whether belief in an object will decay below an unsatisfactory threshold if an observation is missed. These mechanisms were implemented on two mobile robots within the SFX architecture , and demonstrated for indoor, outdoor, and space navigation tasks. SFX is novel in its integration of sensing throughout the reactive and deliberative layers. Sensing is formalized in the reactive layer as perceptual schemas within behaviors. Information agents in the deliberative layer, such as the Cartographer, may also conduct sensing. The output of information agents may be treated as virtual sensors, for use by the perceptual schemas in the reactive layer. A middle agent, the Sensing Manager, allocates sensing resources and is responsible for handling sensor failures.