Communication of behavorial state in multi-agent retrieval tasks

Abstract

This paper assesses the impact on performance of a society of robots in a foraging task when simple communication is introduced. Results are obtained comparing task achievement in the absence of inter-agent communication relative to performance given the minimal knowledge of the behavioral state of fellow agents. Simple communication can result in significant performance enhancement. Int r o ducti o 11Research is currently being conducted on the use of multiple robots to solve problems in a changing workspace by having them work in parallel, and communicate when necessary. The goal of this research is to create a foundation theory which specifies, for a given task, the most reliable, efficient, and robust, means of interaction between a number of robots. This means that the task will be carried out regardless of robotic failures or communication breakdown. Re 1 at ed Wo r kEthological studies as well as robotic implementations are both relevant to this study. This section briefly reviews some of the progress in both of these areas. EthologyIt is clear through ethological studies that multi-agent societies offer significant advantages in the achievement of global tasks. A wide range of social structures exist: uni-level organizations as found in schooling fish, hierarchical systems as found in baboon societies [l], and caste systems typified by many insect colonies (e.g., bees). The relationships between these agents determine, to a degree, the nature and type of communication that is essential for the social system to prosper. The converse also holds in that the communication abilities determine to a degree the most effective social organizations for a particular class of agents. In [6], we discuss these issues in more detail.For the types of foraging and retrieval task described in this particular study, societies of ants constitute a 1050-4729193 $3.00 0 1993 IEEE 588 reasonably close parallel [17]. Although ants typically communicate through chemical trails left during their foraging activities, an act which this current study does not emulate (see [lo] for such a simulation), the state mechanisms used for foraging, acquiring, and retrieving target objects are in strong parallel with these particular systems. A study by Franks [8] details such a process in army ants. Robotics A significant amount of research on multi-agent robotic systems has begun to emerge. Fukuda's early work on the CEBOT system [9] demonstrates the self-organizing behavior of a group of heterogeneous robotic agents. Beni and Hackwood's research [ll] on swarm robotics demonstrates large scale cooperation in simulation. Work at MIT, by Brooks [7] and Mataric [15], shows the development of subsumption-based multi-agent teams, the latter study involving 20 small robot agents. Many other projects have been reported (e.g., [13, 14, IS]), to the point where an entire conference exists to report the results of such work [12]. ApproachThe controversy between reactive control and hierarchical planning for mobile robot navigation has led to the birth...

Collective Robotics: From Social Insects to Robots

“…For example, a lawn mowed by a team of robots can also be mowed by one robot in a longer period. Other such tasks include sorting 13], searching 18], map making 32], material handling 15], and harvesting 19,2 ] . On the other hand, cooperative tasks cannot be accomplished by a single robot and require the cooperative behavior of several machines working together.…”

Title: An Introduction to Robotic Multi Agent Systems

“…This can be explained by an example of traveling from one point A to point B. A reactive agent might use instruments to devise a suitable route while planning agent might take the help from the diagrammatic representation showing rivers, cities and towns [3]. There may be situations where agent may react independently to a circumstance i.e.…”

Regulation Problems in Explorer Agents

“…It should be stressed that we have adopted as termination point of the experiment, the moment of picking of the last sample rather than the moment of return of the last agent to the base, in order to eliminate the cognitive problem of the previous section (How do they know they have finished? ): in the absence of other collective functionalities, this point is independent of the cognitive properties of the agents, in other words the sweeping of samples is parallel with the operation of the representational=cognitive Our purpose has been to show that in the absence of explicit sociality the performance of the system does not improve in a super-linear manner as is said in (Matarić 1992a) and (Arkin et al 1993). Actually, the dispersion mechanism is not in any sense self-catalytic, consequently the only possible gain is the gain from parallelism, which can be linear at best.…”