This research continues an investigation of multi-agent planning. The theoretical basis of this work, the Group Element Model (GEM), is implemented in a computer system (GEMPLAN). The approach taken is to explicitly subdivide the planning problem into regions of activity. Research is targeted to discovering how planning constraints and properties can be localized within these regions, and how coordination and synchronization can be maintained between regions. The technical approach is through first-order temporal logic for constraint descriptions, and algorithms for local search and constraint satisfaction are developed. The importance of this research is that multi-agent planning systems are useful in a wide variety of applications, such as factory scheduling and assembly by co-ordinated groups of robots. This work develops both the theoretical basis and engineering experience required to expand multi-agent planning technology.
