Nothing Special   »   [go: up one dir, main page]

 

Hierarchical planning through propositional logic : highly efficient, versatile, and flexible

Loading...
Thumbnail Image

Date

2019-12-23

Authors

Behnke, Gregor

Journal Title

Journal ISSN

Volume Title

Publication Type

Dissertation

Published in

Abstract

AI Planning is a core technology in enabling advanced assistance for human users. When faced with complex problems such as handicraft tasks for household repairs, Do-It-Yourself projects, or difficult assembly tasks, a planning-based assistance system can provide individualised and context-dependent instructions. It thus effectively supports the user in achieving his or her goals. High-quality assistance should in addition conform to the user’s current wishes and preferences to ensure maximum utility. This can be achieved by involving the user into the planning process, i.e. by making planning mixed-initiative. Hierarchical Task Network (HTN) planning is a method particularly well suited for providing user support, as it resembles the means and structures humans use for problem solving. We identified two major challenges that every mixed-initiative HTN planning system must face and show how to address them: generating plans quickly and flexibly altering them according to the user's demands. The main focus of this thesis lies on addressing the first challenge, i.e. on developing a quickly responding and efficient HTN planner. Designing efficient HTN planners is particularly difficult. Their algorithms have to take both dimensions of the problem description - state and hierarchy - into account and have to consider interactions between them. We use a translation into propositional logic, enabling for the first time a uniform view on the whole planning problem. First, we describe a new encoding for totally-ordered HTN planning, before extending it in two consecutive steps to capture general, partially-ordered domains. Second, we introduce Path Decomposition Trees (PDTs) and Solution Order Graphs (SOGs) which enable a compact encoding and alleviate unnecessary reasoning from a SAT solver. They also pave the way for future insights into structural properties of HTN planning problems, which allow for more efficient planning as well as for more advanced user support. Third, we show that our encodings are a significant empirical improvement over the current state of the art in HTN planning. Lastly, we present a fundamental technique for optimal HTN planning - which is especially important in assistance scenarios - namely a method to compute succinct depth bounds for plan-length optimisation. In a mixed-initiative planning environment, users will frequently request the planner to change a currently considered plan. We start solving this second challenge by considering the most basic task involved: to verify that the changed plan is a solution to the planning problem at hand. First, we show that this task is NP-complete. Second, we develop the first plan verifier for HTN planning, which is based on a transformation into propositional logic. Third, we analyse and categorise the requests possibly made by a user and show that the objectives posed by her or him can be suitably represented as formulae in Linear Temporal Logic (LTL). Fourth, we analyse the computational complexity of changing a plan, showing that this task can be between NP-complete and undecidable. Lastly, we use our SAT-based planner to change a plan with respect to a request formulated as an LTL formula. We show that the full spectrum of LTL formulae can be supported efficiently in a propositional encoding. For that, we introduce a new theoretical foundation for reasoning about parallelism in LTL traces. The practical applicability of our techniques has been demonstrated within a joint transfer project with Robert Bosch GmbH. In it, we developed an assistance system that guides novice users through a handicraft Do-It-Yourself project. The underlying hierarchical planning model is highly complex. Currently the only planner that is able to find plans for this model within an acceptable time frame is the SAT-based HTN planner developed as part of this thesis.

Description

Faculties

Fakultät für Ingenieurwissenschaften, Informatik und Psychologie

Institutions

Institut für Künstliche Intelligenz
Institut für Theoretische Informatik

Citation

DFG Project uulm

TRR 62 / Eine Companion-Technologie für kognitive technische Systeme / DFG / 54371073

EU Project THU

Other projects THU

Is version of

Has version

Supplement to

Supplemented by

Has erratum

Erratum to

Has Part

Gregor Behnke, Daniel Höller, and Susanne Biundo. “Finding Optimal Solutions in HTN Planning – A SAT-based Approach”. Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI 2019). IJCAI, 2019, pp. 5500–5508. doi: 10.24963/ijcai.2019/764.
Gregor Behnke, Daniel Höller, and Susanne Biundo. “Bringing Order to Chaos – A Compact Representation of Partial Order in SAT-based HTN Planning”. Proceedings of the 33rd AAAI Conference on Artificial Intelligence (AAAI 2019). AAAI Press, 2019, pp. 7520–7529. doi: 10.1609/aaai.v33i01.33017520.
Gregor Behnke, Marvin Schiller, Matthias Kraus, Pascal Bercher, Mario Schmautz, Michael Dorna, Michael Dambier, Wolfgang Minker, Birte Glimm, and Susanne Biundo. “Alice in DIY-Wonderland or: Instructing novice users on how to use tools in DIY projects”. AI Communications, 32(1), 2019, pp. 31–57. doi: 10.3233/AIC-180604.
Gregor Behnke and Susanne Biundo. “X and more Parallelism: Integrating LTL-Next into SAT-based Planning with Trajectory Constraints While Allowing for Even More Parallelism”. Inteligencia Artificial, 21(62), 2018, pp. 75–90. doi: 10.4114/intartif.vol21iss62pp75-90.
Gregor Behnke, Daniel Höller, and Susanne Biundo. “totSAT – Totally-Ordered Hierarchical Planning through SAT”. Proceedings of the 32nd AAAI Conference on Artificial Intelligence (AAAI 2018). AAAI Press, 2018, pp. 6110–6118.
Gregor Behnke, Daniel Höller, and Susanne Biundo. “Tracking Branches in Trees – A Propositional Encoding for solving Partially-Ordered HTN Planning Problems”. Proceedings of the 30th International Conference on Tools with Artificial Intelligence (ICTAI 2018). IEEE Computer Society, 2018, pp. 73–80. doi: 10.1109/ICTAI.2018.00022.
Gregor Behnke, Daniel Höller, and Susanne Biundo. “This is a solution! (... but is it though?) – Verifying solutions of hierarchical planning problems”. Proceedings of the 27th International Conference on Automated Planning and Scheduling (ICAPS 2017). AAAI Press, 2017, pp. 20–28.
Gregor Behnke, Daniel Höller, Pascal Bercher, and Susanne Biundo. “Change the Plan – How hard can that be?” Proceedings of the 26th International Conference on Automated Planning and Scheduling (ICAPS 2016). AAAI Press, 2016, pp. 38–46.
Gregor Behnke, Daniel Höller, and Susanne Biundo. “On the Complexity of HTN Plan Verification and Its Implications for Plan Recognition”. Proceedings of the 25th International Conference on Automated Planning and Scheduling (ICAPS 2015). AAAI Press, 2015, pp. 25–33.
Gregor Behnke, Denis Ponomaryov, Marvin Schiller, Pascal Bercher, Florian Nothdurft, Birte Glimm, and Susanne Biundo. “Coherence Across Components in Cognitive Systems – One Ontology to Rule Them All”. Proceedings of the 24th International Joint Conference on Artificial Intelligence (IJCAI 2015). AAAI Press, 2015, pp. 1442–1449.

Part of

DOI external

DOI external

Institutions

Periodical

Degree Program

DFG Project THU

item.page.thu.projectEU

item.page.thu.projectOther

Series

Keywords

Planning, HTN Planning, SAT, Mixed-Initiative Planning, Künstliche Intelligenz, Programmierlogik, Artificial intelligence, Operations research, DDC 004 / Data processing & computer science