|
|
|
| e-Pub |
Previous | 
Home
Section: Application Domains
Mobile Agents
Project-team positioning
CEPAGE has undertaken tasks related to the design of algorithms which control the behavior of so called mobile agents, moving around a network or a geometric environment, with the goal of achieving a specified objective. Objectives of central importance to the study include: exploration of unknown environments, terrain patrolling, network maintenance, and coordination of activities with other agents. Such problems have in recent years been the object of interest of numerous research teams working on Distributed Computing worldwide, in particular, at research centers in Canada (Quebec), Israel (Tel Aviv, Haifa), France (Paris, Marseille), the UK (London, Liverpool), and Switzerland (Zurich). Algorithms for mobile agents in social networking applications are also intensively studied in research labs in the USA (Stanford, Facebook). Papers on mobile agents appear regularly at all of the major conferences in Distributed Computing (PODC, DISC, SPAA), as well as at top venues with a more general algorithmic audience (SODA, ICALP, ESA). Members of CEPAGE actively participate in these events, and are also a recognizable part of the European community focused around mobile agents, helping to animate well-established specialized conferences, such as SIROCCO and OPODIS.
Within Inria, studies of mobile agents are also performed in the GANG project, and to some extent also within MASCOTTE. CEPAGE has active research links with both of these teams.
Scientific achievements
The work of CEPAGE has focused on contributing new decentralized algorithms for controlling mobile entities known as agents, deployed in unknown environments. We mainly considered the network setting, in which agents moving around the nodes of the network graph may be used to analyze the structure of the network and to perform maintenance tasks, such as detecting dynamic faults, improving/monitoring dissemination of information, etc. Our theoretical studies focused on designing new strategies for controlling the behavior of agents and answering crucial questions concerning the feasibility of solving fundamental problems, subject to different model assumptions and constraints on the knowledge and computational power of agents.
One major line of our research focused on the so called anonymous graph model in which an agent is unable to determine the identifier of the node of its current location, but can only see a local ordering of the links around it. Such a study is motivated e.g. by scenarios in which the identifiers of nodes may be too large for the agent to process using its bounded resources, or may change in time. In this model, we studied two of the most fundamental problems: that of traversing all of the nodes of the network (exploration) and of meeting another agent in the network (rendezvous), so as to coordinate with it. Our contributions include a precise characterization of the space requirements for agents solving both of these problems deterministically: exploration in (Trans. Alg. 2008 [84] ) and rendezvous in (Dist. Comp. 2012 [92] ), in a paper presented at the Best Paper Session of PODC 2010. We have also studied fast solutions for specific scenarios of the rendezvous problem (DISC 2010 [60] , DISC 2011 [85] , SPAA 2012 [93] ) and the problem of approximate map construction within an anonymous graph (OPODIS 2010 [82] ). A separate problem, intensively studied in recent years by several research teams, concerns the exploration of a network with pre-configured ports so as to assist the agent. In our work on the topic, our team has proposed several new techniques for graph decomposition, leading in particular to the shortest currently known strategies of periodic exploration for both the case of memoryless (Algorithmica 2012 [112] ) and small-memory agents (SIROCCO 2009 [88] ).
A closely related line of research was devoted to the design of network exploration strategies which guarantee a fast and fair traversal of all the nodes, making use of agents with extremely restricted capabilities. Such strategies were inspired by the random walk, but had the additional advantage of deterministic and desirable behavior in worst-case scenarios. We presented a series of results in the area at notable conferences, involving both the design of new exploration strategies (ICALP 2009 [86] ) and completely new insights into previously known approaches such as the so called “rotor-router model” (DISC 2009 [61] , OPODIS 2009 [62] ). All of the proposed algorithms were shown to be viable alternatives to the random walk, competing in terms of such parameters as cover time, steady-state exploration frequency, and stabilization in the event of faults.
Our efforts have also focused on the theory of coordinating activities of large groups of agents. We have conducted pioneering work in the so called look-compute-move model in networks, in which extremely restricted (asynchronous and oblivious) agents, relying on snapshot views of the system, are nevertheless able to perform useful computational tasks. Our solutions to the problems of collective exploration in trees (Theor. Comp. Sci. 2010 [99] ) and gathering agents on a ring (Theor. Comp. Sci. 2008 [110] and 2010 [109] ) have sparked a long line of follow-up research, accumulating more than 120 citations in total (according to Google Scholar). In a slightly different scenario, we have considered computations with teams of agents whose task is to collaboratively detect and mark potentially dangerous (faulty) links of the network, called “black holes”, which are capable of destroying agents which enter them. We have provided important contributions to the theory of black hole search in both undirected (SIROCCO 2008 [87] , DISC 2008 [100] ) and directed (Theor. Comp. Sci. [113] ) graphs.
It is expected that the mobile agent theme of CEPAGE will give rise to 2 PhD theses. In 2013, Ahmed Wade will defend his thesis on mobile agent protocols for dynamic networks, whereas in 2014 Dominik Pajak will defend his thesis on multi-agent protocols for efficient graph exploration. Our scientific interests also include mobile agent protocols for geometric applications, more remote from the central themes of CEPAGE, but having extensive applications in robotics (providing protocols, e.g., for efficient patrolling and guarding of terrains, traversing terrains using groups of robots, etc.). We have already published several papers in this area (SIROCCO 2010 [90] , SWAT 2010 [91] , ESA 2011 [89] ), building up the theoretical fundamentals of a new field, and already attracting the attention of a wider community of researchers working in robotics and AI.
Perspectives: Our goal is to explore applications of mobile agent techniques in domains of growing importance, namely, social networks and robotics. We are currently discussing applications of our techniques in problems of brand recognition on the web with a local industrial partner (Systonic KeepAlert), and other companies (through our research collaborators in Liverpool). We intend to undertake collaboration with European/American research labs and industrial partners.