Landmark vector model with quantized distance for homing navigation

Inspired by astonishing navigation ability of insects and other animals, many studies observed their behaviors, and considered biomimetic application to robotic systems by investigating mechanisms based on various senses. In this paper, we suggest a new landmark vector model for homing navigation with quantized distance information. The method is highly successful for homing navigation in both perspectives of angular error and success rate. This work has been published in Yu and Kim (2011). Introduction Animals have developed navigation skills based on various senses. Desert ants and honeybees are known to use visual information (Collett, 1996), turtles migrating long distance rely on magnetic compass (Luschi et al., 1996), while other studies have shown that birds use olfactory cues to navigate (Papi, 1990). Many studies have focused on designing bio-inspired navigation algorithms for robotic systems inspired by the excellent performance animals demonstrate. Among them, vision-based homing navigation has been studied through a number of bio-inspired algorithms. One of the simplest method suggested, inspired by desert ants and honeybees, is the ‘snapshot model’ (Cartwright and Collett, 1983). In the snapshot model, currently obtained visual information is compared to that in the snapshot image taken at home location. Several different methods were suggested to process the snapshot images for homing navigation. One of the methods based on the concept is the average landmark vector (ALV) model by Lambrinos et al. (2000). The ALV model is one parameter method where the average landmark vector is obtained by averaging every unitlength landmark vectors perceived in the snapshot. Comparing the average landmark vectors obtained from snapshots taken at the current location and at home location, it is sufficient to guide the agent for the homing direction. The ALV model is based on a simple representation of the environment with powerful performance results in homing navigation, but the model necessarily requires a reference compass information. L C