Christian Gerber

On the suitability of market-based mechanisms for telematics applications

In this paper we compare the suitability of several market based allo cation mechanisms the Vickrey auction the matrix auction for multiple heterogeneous items and the simulated trading algorithm using the alloca tion of transportation tasks to a eet of trucks as an example domain We distinguish three di erent organizational settings in which the set of vehicles represented by autonomous agents may be coordinated by the examined market based mechanisms in a cooperative setting the truck agents are benevolent and try to reduce transportation cost on behalf of a central coordinator i e an agent that represents the shipping company In a competitive setting the truck agents are self interested and aim at opti mizing their private surplus In the hybrid setting a compromise between the con icting goals cost minimization and surplus maximization has to be found We analyze the communication complexity of the mechanisms on a theoretical basis We empirically examine their scalability and tractability by comparing their processing time and allocative e ciency for order sets of di erent size Thereby the allocative e ciency of the mechanisms is measured in terms of cost surplus and number of trucks The results are rated from the point of view of the di erent organizational settings

An empirical evaluation on the suitability of market-based mechanisms for telematics applications

In this paper, we compare the suitability of several market-based allocation mechanisms, the Vickrey auction, the matrix auction for multiple heterogeneous items and the simulated trading algorithm, using the allocation of transportation tasks to a fleet of trucks as an example domain. We distinguish three different organizational settings in which the set of vehicles, represented by autonomous agents, may be coordinated by the examined market-based mechanisms: in a cooperative setting, the truck agents are benevolent and try to reduce transportation cost on behalf of a central coordinator, i.e. an agent that represents the shipping company. In a competitive setting, the truck agents are self-interested and aim at optimizing their private surplus. In the hybrid setting a compromise between the conflicting goals, cost minimization and surplus maximization has to be found. We analyze the communication complexity of the mechanisms on a theoretical basis. We empirically examine their scalability and tractability by comparing their processing time and allocative efficiency for order sets of different size. Thereby, the allocative efficiency of the mechanisms is measured in terms of cost, surplus, and number of trucks. The results are rated from the point of view of the different organizational settings.

Evolution-based self-adaption as an expression for the autonomy degree in multi-agent societies

This work focuses on the development of a method to allow multi-agent systems (MAS) to configure themselves to any application scale and nature. We describe an evolutionary approach to achieve a dynamic adaption of an artificial agent society to environment changes which makes a former efficient society structure suboptimal. Due to the inherent autonomy property of agents this self-adapting mechanism turns out to be an instrument to restrict the autonomy. Therefore, this mechanism provides an internal representation of the degree of agent autonomy in a multi-agent system.

SIF - the social interaction framework system : description and user's guide to a multi-agent system testbed

We present the Social Interaction Framework SIF and demonstrate how it can be used for social simulation. SIF is a simulation testbed for multi-agent systems. The key design aspects are the ability of rapid-prototyping, a broad implementation platform, the possibility of controling agents by human users and easy access to the internal data of every agent in the simulation. SIF implements the EMS (Effector-Medium-Sensor) paradigm, which provides a generic agent-world interface. In this document we describe the architecture, example applications that have been developed at DFKI and we give an easy to follow ten step guide for creating simulations with SIF.

Bottleneck analysis for self-adaption in multi-agent societies

We describe an approach to achieve dynamic adaption of an artificial agent society to environmental changes. Such changes cause sub-optimalities in the agent society design, leading to overloaded or underloaded agents or agent groups in the society. Our approach provides a mechanism to reveal these sub-optimalities by integrating bottleneck analysis to structural self-adaption in a generic multi-agent framework which allows one to incorporate agents of any architecture.

Adaptivity and learning in intelligent real-time systems

Intelligent monitoring and control is a key technique in modern real-time technology. The abstract resource framework provides an adaptive agent architecture equipped with a scheduler that steers the agent’s computation using internal profiles: Utility-based meta-control trades off the cost of invested resources against the expected performance. As in other approaches to meta-control, the important question of how to incorporate external performance feedback has however not been addressed until now: For situated agents, external profiling is a more natural way of predicting utility, because it grounds its estimations on observational feedback with respect to ultimate goals of the agent This paper presents a control process based on abstract resources that additionally considers confidence into internal profiling. Confidence measures are derived from external stimuli by a multi-method learning algorithm. The approach is exemplified and evaluated at hand of the RoboCup simulation domain that provides a benchmark for adaptivity and learning in intelligent real-time systems.