Learning Trustworthy Planning Algorithms (LTPA)

This project tackles one of the key challenges in AI: bridging the gap between learning and reasoning. Both approaches have been able to solve challenging problems by themselves, but we need to combine their advantages to make the next leap forward in AI. This project will contribute to this goal by using machine learning to automate the design of trustworthy planning algorithms. Achieving this goal will allow us to build planners that scale to industrial-size tasks and that can be trusted to support humans.

Background

The project is situated in the area of automated planning, one of the original core AI research fields. Automated planning is the task of finding an action sequence that transforms a given initial state into one that satisfies a given goal condition. Traditionally, planning algorithms are domain-independent, that is, they can solve any planning task from any application domain for which there is a model defining the action dynamics. Due to this generality, planners have been successfully used in a wide range of industrial settings, such as robot task planning, space mission planning, computer games, analyzing computer network vulnerabilities, controlling elevators, and automating greenhouses.

PI: Jendrik Seipp

Funding: This project is partially supported by the Zenith research organization from the Faculty of Science and Engineering at Linköping University.