The PLANET TCU about
On-line Planning and Scheduling

 

 

Look at the call for contributions to the AIPS (International Conference on Artificial Intelligence Planning and Scheduling) workshop on On-line Planning and Scheduling, which will be held in April 2002, in Toulouse, France, and will be at once an AIPS and PLANET workshop.

Whereas the first automatic planning and scheduling systems were designed to run in an off-line setting (planning/scheduling and execution performed in sequence, see Figure 1, often referred to as an open-loop mode), current systems are frequently designed to run in an on-line setting (planning/scheduling and execution running concurrently, see Figure 2, often referred to as an closed-loop mode).

Figure 1: Off-line Planning and Scheduling.

Figure 2: On-line Planning and Scheduling.

Application domains are ubiquitous. They include manufacturing processes, supply chains, workforce organizations, power or communication networks, air or ground transportation, robot control, autonomous spacecraft, and unmanned air vehicles.

New requirements and new technical challenges, that have a strong impact on the planning/scheduling reasoning objective and process, arise in such a setting. Among them:

• the planning/scheduling problem is no longer a static problem; it becomes a dynamic problem, with typically small changes in the definition of the instance to solve between successive calls to the planning/scheduling function;

• new requirements appear, such as robustness and stability in the face of modeling uncertainties and unexpected events that may occur during execution; these requirements must be combined with the usual requirement of producing a plan/schedule of the best possible intrinsic quality;

• hard and soft constraints on the time taken by the planning/scheduling function are introduced, requiring the planning/scheduling module to monitor and control its reasoning process;

• it becomes necessary to take into account inside the planning/scheduling reasoning the time at which the plan/schedule will be provided and executed.

In such a setting, planning/scheduling system designers have to answer important questions related to:

• the precise analysis of the on-line setting and of the new arising requirements;

• the design of methods and algorithms that are able to deal with planning/scheduling as a dynamic problem;

• the design of methods and algorithms for reactive and repair-based planning/scheduling;

• the definition, the combination, and the optimization of criteria like plan/schedule quality, robustness, and stability;

• the definition and the use of various abstraction levels and of various temporal horizons;

• the way of setting and meeting constraints on the time taken by the planning/scheduling function;

• the control of the planning/scheduling reasoning and of the trade-off between quality and time;

• the possible trade-offs between off-line and on-line computing;

• the task sharing and the interactions between the planning, scheduling, and execution functions;

• the validation and assessment of an on-line planning/scheduling system.

All these questions and some other should be dealt with in the Planet II TCU about On-line Planning and Scheduling.

This TCU aims at extending the scope and carrying on with the work of the Planet I TCU about Dynamic Scheduling, which was managed by Simon de Givry and the main result of which was a first characterization of the on-line reasoning process nature.

It is clear that, unlike other TCUs, this TCU is dedicated to no specific application domain. It is a transversal TCU in the Planet network. It is open to any person who wants to spend some time to share with others his/her experience, thoughts, and questions.

The first event of this TCU is the organization of an AIPS (International Conference on Artificial Intelligence Planning and Scheduling) workshop, in April 2002, in Toulouse, France, which will be at once an AIPS and PLANET workshop.

For any question, please contact Gérard Verfaillie, ONERA, Toulouse, France.