Bee-Inspired Landmark Recognition In Robotic Navigation

Research has shown that honeybees have a remarkable ability to use landmarks to travel miles to find - and return to - a food source [1, 2]. The long-term goal of this research is to use physical robots, incorporating a variety of image processing approaches, to examine this ability. The results may also improve our ability to develop effective landmark guidance systems for robots.This research examines whether robots using a simple genetic algorithm (GA) and neural network (NN) that are trained to search for a target object adopt the use of landmarks to aid in the search even when landmarks are not explicitly considered. This is a first step towards more complex experiments in less controlled environments. The simplified environment used in these experiments is controllable, making it easier to determine and understand the robot's behavior before moving to a more complex external environment. The robot uses a feed-forward, episodic NN for navigation - there is no inherent use of a memory bank of images. It is trained using Learning from Demonstration (LfD) to follow a simple search pattern independent of landmarks, but visual cues that could act as landmarks are present in the environment. Our results show that even under these conditions visual cues, i.e. landmarks, are incorporated in the learned search.