MAFTEC @ IJCAI-PRICAI
MAFTEC @ IJCAI-PRICAI
IJCAI-PRICAI has been postponed to January 2021 due to the COVID-19 pandemic. Please find updated information about new dates and location in the conference website (https://ijcai20.org/, Covid Update).
The submission deadline of MAFTEC workshop has been extended to september 11, 2020 (firm).
Automated planning is a classic problem which has been studied since the beginnings of Artificial Intelligence. The traditional focus of the planning community is the generation of plans for an individual agent, with a plan being a sequence of physical actions. This basic problem, known as classical planning, assumes that the agent's actions are described by their positive and negative effects in terms of addition and deletion of atomic propositions, and that the planning agent has perfect information (complete knowledge of the current world state). Much progress has been accomplished for these kinds of planning tasks. Fifteen years ago, a standardized language for classical planning tasks was defined, the Planning Domain Definition Language, PDDL. This made it possible to formulate benchmarks and hold an International Planning Competition (IPC) between automated planners. Throughout the last decade, these competitions have witnessed steady progress.
Beyond classical planning, the generation of plans for an individual agent having incomplete information has also been investigated over the last 25 years: the subfield of conformant planning allows for agents with incomplete knowledge of the initial state but without observation actions, while contingent planning also allows for such actions. Temporal planning is an important extension of classical planning, in which actions have durations and may overlap. An important aspect of temporal planning is that, unlike classical planning, it permits the modelling of problems that can only be solved by executing two or more actions in parallel. Building on the success of planners for single-agent tasks, the planning community is also increasingly interested in multi-agent tasks, involving, for example, robot swarms or robot-human interaction. Reasoning about an agent's knowledge and belief (including higher-order beliefs, i.e., beliefs about other agents' knowledge) is the traditional subject of epistemic logic. Since the 1990s, dynamic epistemic logics also take into account the effects of actions on agents' knowledge. Recently some authors have started to consider epistemic planning problems in this framework. It is the aim of this workshop to make these different avenues of research converge.
The goal of this first MAFTEC workshop is to lay the foundations to model and solve complex real-world planning problems in which many agents interact cooperatively and robustly via physical, communication, and sensing actions to attain common goals in a partially-observable environment. In order to be a solution of such a problem, a plan should take into account the beliefs of each agent which can change over time and it should allow simultaneous executions of actions. It should be sufficiently flexible to allow individual agents to make certain choices at execution time and it should be robust to the failure of certain actions and to changes in the environment. This inevitably implies going beyond the restrictive assumptions of classical AI planning in order to obtain the required level of expressive power, while developing efficient algorithms in order to be able to solve real-world planning problems. Such realistic applications require an extended theoretical framework incorporating the following aspects: