Controlability and Makespan Issues with Robot Action Planning and Execution

Nowadays, many robotic applications need autonomous decision-making capabilities. Among them, some make intensive use of planning. Yet, planning is an activity whose algorithmic complexity is often incompatible with the reactivity requirement of an exploration rover or a space probe. In past years, some planners have proven their ability to handle complex situations required by autonomous systems. Some of these systems (e.g. RAXPS [Jonsson et al. 2000], CASPER [Chien et al. 2005]) have been deployed. The IxTeT planner1 [Ghallab & Laruelle 1994] was developed to handle such robotic planning problems. It was extended to handle complex resources [Laborie & Ghallab 1995], continuous domains and constraints between both atemporal and temporal variables [Trinquart & Ghallab 2001]. Further work [Lemai 2004] added a temporal executive to IxTeT. Reasoning about time is necessary to address these planning problems. The planner must be able to take into account strict deadlines, temporal windows for some tasks, durative actions, and durative goals. The STN2 [Dechter, Meiri, & Pearl 1991] formalism is often used in temporal planning because the requests on these networks are solved very efficiently by polynomial algorithms. Nowadays, an extension to uncertain constraints has been studied and a polynomial algorithm [Morris, Muscettola, & Vidal 2001] has been proposed. Actual robotic space exploration mission are very expensive, with a high requirement for quality scientific returns. During the MER mission, the use of MapGen has allowed a 25% increase of such returns [Rajan 2004]. In a fully autonomous planner, optimization can be made in two ways: finding directly one good plan or searching through the whole search space several plans to find the optimal one. Due to limited computational capacity, the second approach is often unreasonable. So we have to modify the planner to search for high quality solutions. New issues were raised while experimenting with IxTeT