Public defence, Computer Science, MSc Saurabh Fadnis

Title of the thesis: Reductive Approaches to Automated Planning with Partial Observability

Thesis defender: Saurabh Fadnis
Opponent: Senior Associate Professor Jendrik Seipp, Linköping University, Sweden
Custos: Professor Jussi Rintanen, Aalto University School of Science

This dissertation investigates how artificial intelligence systems can plan and make decisions when there is uncertainty about the state of the world and about the outcomes of actions. Such “planning under uncertainty” is a central challenge in the field of automated planning and has applications in areas such as robotics, logistics, and decision‑support systems. 

The work focuses on the partially observable planning problem, where an agent does not see everything about the world, and studies how to make these hard planning problems easier to solve by transforming them into other, better‑understood and simpler problem types. Instead of building new planners from scratch, the thesis develops methods for turning planning problems with uncertainty into forms that existing solver technologies can handle, so that off‑the‑shelf SAT solvers, classical planners, and FOND planners can be reused. 

The main contributions are new ways to represent what a planning system considers possible about the world (its “beliefs”) in a compact form, and practical methods for translating hard partially observable planning into simpler forms. This makes it possible to tackle complex decision‑making problems under uncertainty using general‑purpose planners rather than specialized tools. Overall, the results show that an important problem in AI of planning under uncertainty be addressed effectively by carefully designed transformations that leverage powerful existing solver technology, helping to chip away at the intractability of partially observable planning.

Keywords: Planning, Partial Observability, Satisfiability, Automated Reasoning, Automated Decision-Making

Thesis available for public display 7 days prior to the defence at Aalto University's public display page. 

Contact information:
saurabh.fadnis@aalto.fi