The Reactive Layer
2.4 The Reactive Layer
The realization of the reactive layer differs from one domain to another. It depends on the actions the agent should execute and on the kind of sensors providing information about the environment. In our transport domain there are only few sensors. The road trains might be equipped with an GPS-System which returns the position of the truck. Upon the actual time and the position of the road train the agent is able to check its plan and if there is time enough to serve all orders correctly. In other domains there are much more sensors thinkable, e.g. if the agent represents a robot, it might have several infrared sensors etc. So this shows that the structure of the abstraction levels really depends on the domain of the agents. In our domain the abstraction hierarchy is divided into three levels. First, there is a generic reactive layer which observes the status of the agent. It has to control whether the communication works and if all sensors and actors are in order. This level is responsible for technical correctness of the hardware and the technical surroundings of the agent.
Generic com. Generic com. layer layer
Transp. com. layer Transp. com. layer
Transp. prot. Transp. prot.
1..n
layer layer
Com. layer Com. layer
Holon Com. layer Holon Com. layer
Ship. Com. layer Ship. Com. layer
Generic planning Generic planning layer layer
Generic KB layer Generic KB layer
Bid computation Bid computation
Driving time Driving time
module module
module module
Transp. planning Transp. planning layer layer
Roadmap module Roadmap module
Planer - container Planer - container
Planer - holon Planer - holon
Planer - driver Planer - driver
Planer - general Planer - general
Transp. KB layer Transp. KB layer
Generic reactive Generic reactive layer layer
Simulation Simulation module module
Transp. reactive Transp. reactive layer layer
planexecution planexecution
plan observing plan observing
layer layer
layer layer
Figure 3.
The object hierarchies within the agent architecture
The next level creates a correlation between reactive and planning Analogous to the planning layer there is another level below which layer. This level of the reactive layer has to check if all entries in
can be seen as a module level. It contains some algorithms for the knowledge base are still okay, and if not, it has to initiate some
quick replanning to get out of critical situations and some modules action which will rectify the knowledge base. However the
for a kind of simulation. Simulation is necessary to make a processing of the plans of the agent differs from the knowledge
visualisation of the activities on the one hand and on the other handling. If a fault in the future of the plan is noticed, say that the
hand it is essential to project the situation to the future to see what execution of the plan would be impossible if the agent does not
can happen. From a software developer’s point of view the life modify its plan, this level has to generate a call to the planning
time of these modules is short. So it is necessary to change this layer of the agent to build a new plan on the base of the current
modules without a lot of work and without changing the whole sensor data and knowledge base.
layer.
The third level of the reactive layer is responsible for the correct execution of the actions which are described in the plans of the agent. Therefore this level has to be able to communicate with the relevant hardware components to execute the tasks at hand. This level is responsible for short time observation whereas the level above is considered to do a long time observation.
Proceedings of the ECAI 2000 Workshop on Agent Technologies and Their Application Scenarios in Logistics
2.5 The Knowledge Base
REFERENCES
To build a generic knowledge base for all kind of scenarios is not BFV99: H.-J. Bürckert, K. Fischer, and G. Vierke. Holonic easy because knowledge base systems are very complex and cannot
Fleet Scheduling with TeleTruck . In Proceedings of the
be divided in independent parts. Hence, we decided to restrict to two abstraction levels.
Second International Conference on Computing Anticipatory Due to the fact that most knowledge bases are build up on
Systems (CASYS'98), 1999
databases, the highest knowledge base level mainly prepares interfaces between database and agent layers. If only a standard
BV99: H.-J. Bürckert and G. Vierke. Simulated Trading SQL-database has to be included, this level should provide some
Mechanismen für Speditionsübergreifende methods which enables the agent to request information from the
Transportplanung . In H. Kopfer and C. Bierwirth, editors, database using the SQL syntax.
In our domain there is no ‚active‘ knowledge base needed which Logistik Management - Intelligente I+K Technologien. has to draw inferences about the knowledge. The knowledge base
Springer-Verlag, 1999.
layer has only to be able to store data about the resources of the agents and it has to save the plan information if the system is going
FF94: T. Finin, R. Fritzson. KQML – A Language and down. From an abstract point of view the knowledge base is just a
Protocol for Knowledge and Information Exchange . data pool for an agent. However, this level includes all interfaces to
Proceedings of the 13 th International Distributed Artificial maybe different database systems and to systems which can hold
persistent data.
Intelligence Workshop, 1994.
In the next lower abstraction level there are algorithms to draw
inferences. This level includes the mechanism which constitutes a FVB99: P. Funk, G. Vierke, and H.-J. Bürckert. A Multi- real knowledge base system.
Agent Systems Perspective on Intermodal Transport Chains. In H. Kopfer and C. Bierwirth, editors, Logistik Management - Intelligente I+K Technologien. Springer-
3 CONCLUSION
Verlag, 1999.
We have described an approach that allows to design and implement multi agent systems in a simple and convenient
GSV99: C. Gerber, J. Siekmann, and G. Vierke. Holonic manner. By abstracting from several existing agent systems
Multi-Agent Systems . Research Report RR-99-03, DFKI, for transportation purposes, we have constructed a generic
1999.
agent model that in turn serves to derive new specific agents in a straightforward manner and that allows to save
Mül96: J. P. Müller. The Design of Intelligent Agents – a redundant implementation effort. The model combines a
Layered Approach . Volume 1177: Lectures notes in sophisticated agent architecture with the holonic agent
artificial intelligence, Springer, 1996. design paradigm. Our future work aims at implementing a computer aided
Sea69: J. R. Searle: Speech Acts. Cambridge University software engineering tool that supports the automated
Press, 1969.
generation of agents independently from programming languages and hardware platforms. We consider our generic
Wei99: G. Weiss: Multiagent Systems – A Modern agent model being an appropriate basis for this purpose.
Approach to Distributed Artificial Intelligence. MIT Press,1999.
Proceedings of the ECAI 2000 Workshop on Agent Technologies and Their Application Scenarios in Logistics
MULTIDIMENSIONAL UTILITY VECTORS IN