B4 - Knowledge Management Supporting Autonomous Logistic Processes

Motivation

In analogy to conventional logistics, autonomous logistic processes are in need of knowledge to perform their task. Data, information, and knowledge are the key resources which ensure the quality of the logistic process. Knowledge management is required to support autonomous logistic processes by context-sensitive provision of knowledge. Furthermore, it has to be considered, that actors in these processes act in a competitive way. Consequently, information and knowledge should be treated as tradable goods which potentially have a high utility for their consumers.

Results Phase 1 (2004-2007)

Starting with a systematic analysis of knowledge-oriented system architectures, the research within this subproject identified requirements for knowledge management in autonomous processes. An approach based on roles and parameters has been developed to enable knowledge management in this domain. The distribution is implemented by flexible agent interaction mechanisms. Roles are complex behaviour patterns which are incorporated by agents. By the help of these roles, complex knowledge management functionalities, e.g., retrieval, distribution, and requests, are enabled. Details of this approach have been presented and published on international workshops and conferences.
Furthermore, an ontology as the formal representation of domain knowledge has been developed, which provides the basic concepts for the specification of transport, information, and manufacturing logistics. The CRC-wide multiagent-based simulation platform PlaSMA has been specified and prototypically implemented. This platform may be used for the analysis of logistic systems and evaluation of autonomous logistic processes by agent-based simulation.


Objectives Phase 2 (2008-2011)

In the second phase of the CRC, the subproject focuses on the subject area of context awareness. To this end, it is initially investigated which pieces of context information bear relevance for autonomous logistic systems in concrete situations where decisions need to be rendered. It is further investigated how, based on a quantification of the aforementioned relevance, a goal-oriented acquisition of required information can be enabled. Moreover, relevant pieces of information are often not immediately available via local sensors or offered by information services. For such cases, the applicability of machine learning, conjointly conducted by the autonomous systems, is subject of further research.

Approach Phase 2 (2008-2011)

The development of a physical world model for the PlaSMA system enabled further experiments regarding the assessment of information by the agents with respect to its value in the decision making process. Consequently, the influence of the accessibility of environmental information on the decision making was quantified in corresponding simulations. Additionally both logic-based and probabilistic forms of knowledge representations were examined in terms of their suitability for dealing with partial and uncertain knowledge. Based on these findings an approach for optimized information acquisition was developed.
To enhance the formal knowledge representations employed Foundational Ontologies were investigated in terms of their suitability for representing context-dependent knowledge. This formed the basis for the reification of concepts and, thus, enabled a situation-based interpretation of environment information.
In close cooperation with the project B10 machine learning methods were examined for local prediction models based on individual experiences for autonomous adaptation. These methods were then adapted for predicting and optimizing individual operative tour planning.


Results Phase 2 (2008-2011)

In the context of the second phase of the CRC, an approach for the quantification of the value of missing information based on the Information Value Theory has been developed. It provides the basis for a methodology for an iterative, utility-oriented information acquisition to support the decision making processes of intelligent agents. In cooperation with the data integration subproject, an architecture for the provision of logistic process data by means of an integration of FIPA-based multiagent systems with the EPCGlobal Framework has been specified. The multiagent-based simulation platform from phase one has been extended substantially with a simulation world model that allows inter alia the explicit modelling of actions of those physical objects controlled by software agents and the generation of environmental events. The ontological modelling of logistic domain knowledge has been continued based on a foundational ontology (DOLCE). Another extension concerns the ontological modelling of multiagent simulation systems themselves.