Research in distributed multi-agent planning (DMAP) aiming to formally define, describe, and most importantly prove how much of the private information is really preserved by the planning algorithms.
Distributed Multi-Agent Planning (DMAP) became an established research field in the area of automated planning in recent years. Although most of the literature on DMAP is motivated by privacy preservation of the agents (without it, DMAP does not make much sense), only a fraction of the research formally defines, describes, and most importantly proves how much of the private information is really preserved by the planning algorithms.
In this project, we propose to follow these works and settle the research debt such that it is possible to unambiguously work with the private information in DMAP. In order to get there, we will focus on definition and formalization of private information and the related formulation of DMAP in existing secure computation models as BlindTMs and ORAM. We will identify theoretical sources of privacy leakage both general (for the MA-STRIPS model) and specific (esp. for MAD-A* and MAFS algorithms). Finally, we will create novel planning algorithms and with their help a planner allowing parameterization of preserved privacy, efficiency, and completeness.