A Planning and Scheduling (P&S) system devoted to solve real-world problems must tackle two crucial aspects: a) the plan obtained by the P&S system must be executed in the real world and b) the inherent complexity and uncertainty that govern the real domain applications.

When designing a real-world P&S system, it is necessary to take into account, among others, the following aspects:

  1. a) A complete integration of time and resource management within a practical planning framework, where not only it is sufficient to explicitly handle actions with durations but also the temporal constraints that arise between objects, actions and goals as a consequence of the resource utilization. In addition, the different types of resources (consumable, renewable, etc) may impose different types of constraints.
  2. b) The evolution of a changing and partially observable world and its consequences in the plan execution (non deterministic actions, imperfect perception, non-categorical goals, etc.). The use of appropriate methods for monitoring the plan execution and the adaptation of classical planning techniques for repairing plans when failures are detected allows overcoming these difficulties, thus eliminating the need to resort to highly-reactive planning architectures.
  3. c) Learning to (semi)-automatically acquire knowledge about tasks, adapt to environment changes and improve problem solving and execution. Learning permits: a) generating knowledge to support the design of action models in domain-independent planning, and b) integrating the information acquired during the planning-execution process to improve the overall P&S system performance and the adaptation of plans to successive problems.

The objective of PELEA project is the definition and implementation of an architecture that integrates all the previous techniques as its building components.

We are currently working on improving current basic techniques (for planning, scheduling, execution, monitoring, plan adaptation and machine learning) to take into account their relationship with the rest of components of the architecture.

We have built an implementation of the PELEA architecture that integrates several techniques for each of the components. PELEA is able to use several planners, schedulers, plan adaptation and learning techniques.