Bee-Inspired Algorithms 2010

Introduction

Honey bees are truly fascinating creatures. These tiny creatures are responsible for pollinating the majority of fruit and vegetable plants worldwide. However colony collapse disorder was responsible for the collapse of 36% of America’s 2.4 million hives in 2008 [1]. In August of 2009 a research team from Harvard University received a $10 million grant from the National Science Foundation to design, manufacture, and program tiny robotic agents to mimic live bees that scout, forage, and plan [2]. This development may become immensely important in the future if the population of honey bees decreases drastically and flower pollination must be preformed by robotic bees rather than living honey bees.

Modeling honey bee behavior is an active research area with many areas of focus. Bee behavior has inspired algorithms for internet routing, job scheduling, web search, water distribution systems, and more [3]. The research presented in this paper concerns modeling honey-bee foraging behavior in bee-like robots using the Player/Stage robotic simulation environment.

Bee Foraging

Bee foraging typically begins with scout bees exploring an area and looking for nectar sources. Bees have excellent color perception, which is more favored towards ultraviolet, and can differentiate yellow, blue, white, and violet [4]. Bees are unable to differentiate between colors such as black and red since those colors are lower on the ultraviolet color spectrum. The scouts will locate these patches, report back to the hive, and perform the ‘Waggle Dance’ which informs the worker bees where to find the nectar sources [5]. The recruited worker bees then travel to these advertised locations, retrieve pollen, and return to the hive.

Once back at the hive the bees will again perform the ‘Waggle Dance’ after unloading the collected nectar. Each returning bee (referred to as an employed forager) will determine if they should abandon this patch or continue exploiting it. This abandonment probability is determined based on the current patch’s reward. The overall reward is determined by nectar quality vs the distance to the source from the hive. More rewarding patches will have a lower abandonment probability, while less rewarding patches have a higher abandonment probability.

The abandonment probability is the primary factor used for finding the most rewarding nectar patches. Honey bees do not make direct comparisons between nectar sources. They only know of a single source at a time and therefore cannot decide if one nectar source is better than another [6]. Over time the worker bees will all begin to favor the patches with the lowest abandonment probability, which also have the highest reward, but not necessarily the highest nectar concentrations.

On a high level honey bees have the following modes of behavior:

Novice Forager: A bee without any prior experience. A novice forager can an either become a Scout (a random low probability) or will follow a waggle dance. Scout: Wanders randomly in the world until a source is found or when the homing motivation reaches its maximum. Recruit: A novice forager who sees a waggle dance in the hive and goes looking for the advertised source. A recruit only knows the approximate location of the nectar source (angle and distance from the hive). Forager: A former recruit that has found a source (knows the exact location) and is exploiting that source. Several sub-types are foragers are: Employed forager: A forager that is activly exploiting a source. Unemployed forager: A forager that has abandoned its source. Reactivated forager: A forager that has resumed exploiting a previously abandoned source.

For additional details on the categories of foragers see [7].

Approach

Environment / Player/Stage Design

The first step in attempting to simulate honey bee foraging behavior was to create a bee like robot. I decided to build my own custom robot and outfit it with sonar sensors and a blobfinder sensor. Our ‘beebot’ is defined as follows. Note that in our final implementation, the beebot was shrunk to 50% of the original size displayed above in order to fit better in our world.