Supervised Learning in Robotic Swarms: From Training Samples to Emergent Behavior

Abstract

Emergent behavior in swarm robotic systems is key to obtaining complex behavior by a group of relatively simple agents. The question is how to design the individual behaviors of agents in such a way that the desired global behavior emerges. Different approaches have been proposed to solve this problem: from biologically inspired probabilistic behavioral models to evolutionary techniques. In some situations, however, creating a complex probabilistic model of the behavior or developing a proper setup for an evolutionary process can be challenging. In this paper we propose a new method, based on supervised learning on a relatively small number of training samples. We apply our method to the well-known clustering problem and show that this approach yields the desired global clustering behavior.

A review of swarm robotics tasks

“…Quantitative measures of task execution performance often used to assess the effectiveness of an algorithm include the number of clusters being created [95,97,96,91], the average cluster size [96,100], the size of the largest cluster [95,96,100,91,90], and the time for task completion [95]. Task completion can be defined as the achievement of a condition where there is a single cluster per object type (in scenarios where this can be obtained with a significant probability), or the number and size of clusters is "satisfactory" according to given criteria.…”

“…In [100], single-type object clustering is obtained using a simple neural network with 2 inputs and 3 outputs. Robots push objects with a frontal shovel, and are able to detect and count the number of objects in two detection areas, one in immediate proximity and one farther away; the numbers of objects Objects are moved by pushing them.…”

A Reliable Behavioral Model

“…Amongclusteringscenarios,studiedinliterature,therearethosewithbothphysicalandsimulated experiments (Gauci, Chen, Li, Dodd, & Groß, 2014) or only simulated experiments (Barfoot & D'eleuterio, 2005;Hartmann, 2005;Vorobyev, Vardy, & Banzhaf, 2014). These scenarios differ alsointermsofenvironmentsize,numberofrobotsandobjects,perceptioncapacities,andpossible behaviorsofusedrobots.…”

Conformity and Nonconformity in Collective Robotics: A Case Study

“…As a case study, we use the sorting task in the context of multi-agent systems, which is formulated as follows: given a set of objects of different types {x 1 , x 2 , ..., x n }, the group of N agents is to collect them into homogeneous clusters. Swarm robotics offers distributed algorithms for solving this problem (Bonabeau et al, 1999;Deneubourg et al, 1991;Bayindir and Sahin, 2007;Beckers and Holland, 1994;Melhuish and Hoddell, 1998;Wang and Zhang, 2004;Verret et al, 2004;Vorobyev et al, 2012). The distinguishing property of the swarm-based approach is that agents operate and perceive only locally; thus, no global supervision and/or knowledge is required.…”