Combining coordination mechanisms to improve performance in multi-robot teams

 

Coordinating a team of agents such that they collectively achieve a common goal is a complex problem within the field of multiagent systems [13]. Improving coordination in multiagent systems will benefit many application domain including Unmanned Aerial Vehicles (UAV) swarms, search and rescue mission, exploration, and sensor networks [1, 2, 9]. In general, coordination mechanisms can be broken down into two main categories: implicit and explicit coordination mechanisms. Implicit coordination relies solely upon an agent’s observation of its environment to make decisions, while explicit coordination involves direct interaction and exchange of information between two or more agents. Implicit coordination mechanisms tend to be limited by observation restrictions and explicit methods are typically limited by communication restrictions. In many real-world domains agents have access to limited amounts of both observation and communication. In such cases, maximizing the benefit of both types of information by concurrently using implicit and explicit mechanisms is likely to be advantageous. We propose using a combination of explicit and implicit coordination mechanisms to improve coordination and performance over either method individually.

In this work we combine two well-known implicit coordination mechanisms (coupled policy evaluations and stigmergy) with our novel explicit coordination mechanism (communicating agents’ Intended Destination Enhanced Artificial State (IDEAS)) to improve coordination in the Cooperatively Coupled Rover Domain (CCRD). The CCRD is an extension of the Continuous Rover Domain [3], in which a set of rovers must coordinate their actions to collectively optimize coverage over a set of environmental points of interest (POIs) and individual rovers can observe any given POI. The CCRD increases the coordination complexity by requiring teams of agents to observe each POI. Here, agents must not only optimize coverage as a collective, but they must form teams and the teams must coordinate within themselves to optimize coverage of their POI as well. This is difficult because there are two different coordination problems going on concurrently. First, at a high level all agents within the system must coordinate to provide optimal coverage of POIs. Second, agents must co-ordinate amongst themselves to form teams and the agents comprising the teams must coordinate their actions to optimally select and cover a given POI (teams either observe a POI together or not at all). This tight coupling between agents both at a system level and at a team level presents a complex coordination problem.

The key contributions of this paper are as follows:

  •   Introduce a novel explicit coordination mechanism (IDEAS) in the Cooperatively Coupled Rover Domain.
  •   Combine implicit (coupled policy evaluation and stigmergy) and explicit (IDEAS) coordination mechanisms to improve performance of multi-rover teams.

    The remainder of this paper is structured as follows. Section 2 provides background information on implicit coordination, explicit coordination, and multiagent coordination. Section 3 provides an introduction to the Continuous Rover Domain (CRD) used in this work. Section 4 provides an overview of the algorithms, evaluation functions, and methods used in this work. Section 5 contains the experimental results. Finally, Section 6 contains the discussion and conclusions of this work.

For full text: click here

(Author: Ehsan Nasroullahi, Kagan Tumer

Published by Sciedu Press)