The goal of the Jaguar project is to develop a planning, execution, and learning framework to adapt planning domain models to the execution environment by learning from execution failures. We consider producing mission plans using a novel planning language called ASynchronous Evolving Tasks (ASET) that can represent non-determinism both with respect to action duration and effects.

Since October 2004, we have been working on developing a prototype learning system. The architecture of this system is shown below. The prototype consists of a ASET planning module connected to an execution simulator. If the plan execution fails, the simulator returns a plan deviation or plan anomaly to a model adaptation module that analyse the failure and correct the domain model.

• Oct. 2004: Design of prototype architecture propsal.
• Nov. 2004: Development of the ASET language.
• Dec. 2004: Implementation of prototype modules
• Jan. 2005: Test of first version of prototype

• (ps,pdf,bib) ASET: Representing Indeterminate Durative Tasks for Symbolic Fault Tolerant Planning in Multi-Agent Domains, R.M. Jensen and M.M. Veloso, Submitted to ICAPS 05, 2004.