Toward Perception-Based Navigation Via EgoSpheres

A method for perception-based egocentric navigation of mobile robots is described. Each robot has a local short-term memory structure called the Sensory EgoSphere (SES), which is indexed by azimuth, elevation, and time. Directional sensory processing modules write information on the SES at the location corresponding to the source direction. Each robot has a partial map of its operational area that it has received a priori. The map is populated with landmarks and is not necessarily metrically accurate. Each robot is given a goal location and a route plan. The route plan is a set of via-points that are not used directly. Instead, a robot uses each point to construct a Landmark EgoSphere (LES) a circular projection of the landmarks from the map onto an EgoSphere centered at the via-point. Under normal circumstances, the LES will be mostly unaffected by slight variations in the via-point location. Thus, the route plan is transformed into a set of via-regions each described by an LES. A robot navigates by comparing the next LES in its route plan to the current contents of its SES. It heads toward the indicated landmarks until its SES matches the LES sufficiently to indicate that the robot is near the suggested via-point. The proposed method is particularly useful for enabling the exchange of robust route information between robots under low data rate communications constraints. An example of such an exchange is given.