|University of Ulm, Faculty of Computer Science, Dept. of Artificial Intelligence||up: Research|
PLANET is a co-ordinating organisation for European research and development in the field of Artificial Intelligence Planning and Scheduling. It aims to stimulate innovation and to promote the transfer of this leading-edge technology into European industry. PLANET is a Network of Excellence funded by Esprit, the European Information Technologies Programme.
The network's main activities are the construction of an effective organisational framework and an elaborate information and communication infrastructure. It supports training, exchange, and technology transfer as well as a close co-operation between academia and industry in both the co-ordination of future research and applications and the realisation of appropriate working programmes.
Work in PLANET is concentrating on certain areas of particular interest, so-called Technical Coordination Units (TCUs). These include the application of planning and scheduling technology for workflow management, intelligent manufacturing, and dynamic scheduling, the fields of robot planning, and knowledge engineering for planning and scheduling, as well as the application of the technology in the aerospace and airline sector.
PLANET started in October 1998 and currently consists of 57 nodes from 14 European countries. They represent leading universities, research centres, and industrial companies.
PLANET is an open network and welcomes participants from all over Europe. For applications please contact the network co-ordinator, Susanne Biundo at the University of Ulm, Germany, or email@example.com.We participate in the following TCUs:
|PLANET related Research|
Currently, research activities at the Ulm node are devoted to hierarchical planning and to system support in the construction and maintenance of provably consistent domain models.
|Hierarchical Planning for Autonomous Systems|
The main aim of this project is to develop a hierarchical planning system that integrates deliberative planning, plan execution, and information gathering. The exchange of information between the plan generation and execution components is implemented in such a way that information gathering actions and plans can be parts of the usual domain plans on arbitrary levels of abstraction. Their goal-directed execution is interleaved with planning and the information obtained this way is used to guide the further planning process. In addition execution failures are reported to the planning component, which then initiates a replanning process for the failed task.
The approach relies on the HTN methodology, which it extends towards several directions. The extensions include an increased expressiveness of the underlying planning language. Control structures are introduced and executable programs are integrated as ``bodies'' of basic HTN methods. The planning strategy to be implemented allows for incremental planning and stepwise refinement and for the exploitation of information gathered during planning. Most importantly, the strategy also enables the continuous integration of new tasks into the current planning process.
Furthermore, a formal, logic-based semantics will be provided for this planning method. Based on this semantics, formal properties of plans like correctness, safety, and feasibility can be formulated and proved. This is in particular important in view of safety-critical applications.
The application potential of the approach is manifold. It includes the control of autonomous systems like robots as well as the control of software agents. It is also suitable for applications in the command and control area.
|Systematic Domain Model Construction|
In complex and large-scale planning applications, like planning for autonomous systems, the development and maintenance of the underlying domain models is a crucial task. Consequently, system support in constructing these models and in keeping them consistent is essential when aiming at these applications.
To this end, a concept to assist users in the incremental and modular construction of verified models of planning domains has been developed [BS96,BS97]. It is based on a temporal logic representation formalism and considers domain models as formal structures. Well-defined and safe operations for the union, extension, and refinement allow to build complex domain models out of already existing simpler ones. A deductive component will automatically perform the proofs necessary to guarantee both the consistency of single models and the safety of operations on models. A GUI will be used to keep users from the details of the underlying logical formalism.
While the concept has been completely worked out and a proof-of-concept prototype implementation was completed, a new project currently begins to work on a more sophisticated implementation. The prototype of this system will be available within one year.
In addition to the topics addressed above, research interests and activities of the group are centered around formal methods in planning that allow for the specification and proof of formal properties like the correctness of plans and planning systems, the safety and robustness of plans, the consistency of models, and the reliability of systems.
|The Homepage of PLANET, The European Network of Excellence in AI Planning|
|The European Networks of Excellence Gateway, with links to most of the Networks of Excellence.|
|EuroNets, The Network of European Networks of Excellence|
|Dept. of AI Homepage||Research||Help||Mail to Webmaster||-bs Jul.18,2000|