Dealing with unanticipated events in autonomous navigation

In earlier published work, the authors developed and demonstrated expert system control of an autonomous robot navigating in unknown environmets and dealing with unexpected events such as moving obstacles. The limitations of this earlier work included insufficient on-board computing capability, restrictions on the sensor suite to only sonar transducers, and a geometrically specified goal location. In this paper, we describe advances to our mobile robot series (currently HERMIES-IIB) to include 8 NCUBE processors onboard, (computationally equivalent to 8 Vax 11/780's) operating in parallel, and augmentation of the sensor suite with cameras to facilitate on-board vision analysis and goal finding. The essential capabilities of the expert system described in our earlier paper have also been ported to the on-board HERMIES-IIB camputers thereby eliminating off-board computation. We describe a successful experiment in which a robot is placed in an initial arbitrary location without prior specification of the room contents, successfully discovers and navigates around stationary and occasionally moving obstacles, picks up and moves small obstacles, searches for a control panel, and reads the meters found on the panel. The success of this experiment encourages more complex on-board vision analysis and expert system control in dynamic environments. The experiment will be expanded to include research in robot learning as the robot attempts to adjust the meter reading to a particular value without a priori knowledge of the control panel system dynamics.