A planning domain is an abstract model of some target system that we wish to synthesize plans for. In order for our plans to succeed, the planning domain must be accurate in the sense that it precisely describes how the system evolves as a function of the activities that are executed in it.

Real target systems are often very large and complex. Constructing correct planning domains for them can be extremely challenging. The problem is further complicated by the fact that most target systems change over time: some activities and objects are phased out while others are introduced. Adapting to such changes typically require manual correction of the planning domain. In many cases, however, users lack the time or expertise to make these corrections. This leads to a fast degradation of the planning system. It may even be the case that the planning system is useless at time of deployment due to the changes that happened in the target domain between the design of the planning domain and the deployment of the planning system.

The Jaguar project aim at developing automated approaches to adapt a domain model to its target system by learning from execution failures. We consider realistic multi-agent planning problems where agent activities may be uncertain both with respect to duration and effects.

A relief mission planning domain is a good example
of a target system with frequent changes