Factory Domain - version 1.0

Learning in Planning with Temporally Extended Goals and Uncontrollable Events




Data Files



This web page is intended to make avaliable the first version of the benchmark presented in the article Learning in Planning with Temporally Extended Goals and Uncontrollable Events, submitted to ECAI-08. It represents a practical application to the configuration model presented in the article, in which a sequence of controllable actions must be determined in order to reach a nominal behavior state in a system.

This state is such that, for all possible uncontrollable events sequences generated from it, a specification defined in temporal logic must always hold. Examples of temporal logic include Linear Temporal Logic (LTL), Computational Tree Logic (CTL) and others. Also, this model may be applied to several real-life scenarios, like automated configuration of a composite system such as a power grid or a network of water pipes.

The benchmark domain resembles a factory application, where machines are connected to repositories that supply raw material for production purposes. In this context, the system controllable actions define connections/disconnections between machines and repositories and machine cleaning, while uncontrollable events are related to product generation when mixing raw material, breaking of tanks, and others. As for temporal conditions, liveness properties may represent product demands, while safety specifications can be used to state forbidden product combinations in machines that are not clean.

We strongly encourage tests over the current data set and extensions to the benchmark, such as the inclusion of tankage constraints, production deadlines, quality measurements, and so on.

The authors of the benchmark are Adi Botea and André A. Ciré. The benchmark webpage is created by André A. Ciré.