An Overview of Swarm Robotics for Search and Rescue Applications

Introduction

The creation of artificial devices with life-like characteristics has been pursued over the last two millenniums, beginning, as so many things in our modern world did, in Ancient Greece. More recently, mainly in the last two decades, a significant progress in applied computing and robotics occurred through the application of principles derived from the study of biology. The navigation of groups of robots, especially swarm robots, has been one of the fields that has benefited from biological inspiration (Bonabeau, Dorigo, & Theraulaz, 1999).

Nevertheless, until recently, swarm robotics applicability has been kept far from real-world missions, such as search and rescue (SaR). To cross the reality gap inherent to MRS and, in particular, swarm robotics, many initiatives^1,2 and worldwide research and development (R&D) efforts have been focusing on the full-scale deployment of such systems in real-world application environments. This chapter essentially revolves around SaR applications as they cover most of the challenges encountered in real-world applications. Those applications are regularly classified by some specific features that pose several problems, especially in terms of inter-robot communication, in which we can highlight the following ones (Murphy, et al., 2008):

• •

  Large: Dimensions of real applications do not stick to laboratorial testbeds, going from hundreds of square meters (e.g., manmade disasters) to hectares of land (e.g., natural disasters);

• •

  Highly Dynamic: Scenarios change over time, either due to agents’ mobility and actions or due to external factors (e.g., collapses, fires, floods, earthquakes, explosions);

• •

  Harsh and Faulty: Abundant presence of abrasive dust, water, chemical substances and a wide range of obstacles in the scenarios;

It is with these real-world problems in mind that swarm robotics is seen as the most promising class of multi-robot systems (MRS). Those techniques offer several major benefits over the conventional search techniques, such as the robustness of the swarm to individual units failure or run-time addition of new units, the scalability of emergent behaviours to swarms of different sizes, the leveraging of self-organization principles of environmental noise and individual differences, and the synergetic effect whereby the work of the swarm is greater than the sum of the work by the individual units (known as superlinearity) (Floreano & Mattiussi, 2008) – a concept shared by other fields, such as complex systems.

On these grounds, other examples of potential applications for swarm robotics include military missions, unmanned space exploration, environmental data collection, and others. However, regardless the application type, all of them require two particular features: a distributed architecture and inter-robot communication.

The first feature is common within swarm robotics context, in which tasks are inherently distributed in space, time, or functionality. Nevertheless, it should be noted that some works still emphasize on centralized architectures (Li, Alvarez, De Pellegrini, Prabhakaran, & Chlamtac, 2007), thus moving away from the fully distributed nature inherent to the principles of collective intelligence. In practice, centralized swarm architectures are computationally expensive and unsuitable as a large number of robots usually generates very dynamic behaviours that a centralized controller cannot handle (Sahin E., 2005). Also, centralized architectures lack robustness as the failure of the centralized entity may compromise the performance of the whole MRS (Parker L. E., 2008a).