Marcel Steinmetz
Marcel Steinmetz
Post-Doctoral Researcher
Marcel Steinmetz is a postdoc at LAAS-CNRS and ANITI, University of Toulouse. Prior to that, he did a postdoc in the Foundations of Artificial Intelligence group headed by Jörg Hoffmann at Saarland University, where he also obtained his Ph.D.
He has been working on various aspects gravitating around AI planning, amongst others, conflict-driven learning techniques in AI planning akin to clause learning in SAT; MDP probabilistic planning and quantiative model checking; planning with description logics; explainable AI planning; simulated network penetration testing and web-security analysis via AI planning methods; and most recently on testing and verifying machine-learned policies.
News
07/2024

Our two papers got accepted at ECAI24!
02/2024

We got three papers accepted at ICAPS24!
10/2023

Marcel received the Quality Champion award for reviewing for ECAI23.
10/2023

Marcel’s dissertation was awarded with the Dr. Eduard Martin-Preis. (Official press release).
09/2023

We have a paper on Cartesian abstractions in probabilistic MDP planning at ECAI23.
Publications
2024
  • Safety Verification of Tree-Ensemble Policies via Predicate Abstraction
    Chaahat Jain; Lorenzo Cascioli; Laurens Devos; Marcel Vinzent; Marcel Steinmetz; Jesse Davis; Jörg Hoffmann
    27th European Conference on Artificial Intelligence, ECAI 2024.
    conference
    ECAI
  • Merge-and-Shrink Heuristics for SSPs with Prune Transformations
    Thorsten Klößner; Álvaro Torralba; Marcel Steinmetz; Silvan Sievers
    27th European Conference on Artificial Intelligence, ECAI 2024.
    conference
    ECAI
  • On the Computational Complexity of Stackelberg Planning and Meta-Operator Verification
    Gregor Behnke; Marcel Steinmetz
    Thirty-Fourth International Conference on Automated Planning and Scheduling, ICAPS 2024. 20--24
    PDF Tech. Report
    conference
    ICAPS
    doi: 10.1609/icaps.v34i1.31456
  • Towards Feasible Higher-Dimensional Potential Heuristics
    Daniel Fiser; Marcel Steinmetz
    Thirty-Fourth International Conference on Automated Planning and Scheduling, ICAPS 2024. 210--220
    PDF
    conference
    ICAPS
    doi: 10.1609/icaps.v34i1.31478
  • Explaining the Space of SSP Policies via Policy-Property Dependencies: Complexity, Algorithms, and Relation to Multi-Objective Planning
    Marcel Steinmetz; Sylvie Thiébaux; Daniel Höller; Florent Teichteil-Königsbuch
    Thirty-Fourth International Conference on Automated Planning and Scheduling, ICAPS 2024. 555--564
    PDF
    conference
    ICAPS
    doi: 10.1609/icaps.v34i1.31517
2023
  • Cartesian Abstractions and Saturated Cost Partitioning in Probabilistic Planning.
    Thorsten Klößner; Jendrik Seipp; Marcel Steinmetz
    26th European Conference on Artificial Intelligence, ECAI 2023. 1272-1279
    PDF
    conference
    ECAI
    doi: 10.3233/FAIA230405
  • A Theory of Merge-and-Shrink for Stochastic Shortest Path Problems.
    Thorsten Klößner; Álvaro Torralba; Marcel Steinmetz; Silvan Sievers
    Thirty-Third International Conference on Automated Planning and Scheduling, ICAPS 2023. 203-211
    PDF
    conference
    ICAPS
    doi: 10.1609/ICAPS.V33I1.27196
  • Lifted Stackelberg Planning.
    Philipp Sauer; Marcel Steinmetz; Robert Künnemann; Jörg Hoffmann
    Thirty-Third International Conference on Automated Planning and Scheduling, ICAPS 2023. 370-374
    PDF
    conference
    ICAPS
    doi: 10.1609/ICAPS.V33I1.27215
  • Analyzing neural network behavior through deep statistical model checking.
    Timo P. Gros; Holger Hermanns; Jörg Hoffmann; Michaela Klauck; Marcel Steinmetz
    International Journal on Software Tools for Technology Transfer. 2023, 25, 407-426
    PDF
    journal
    doi: 10.1007/S10009-022-00685-9
2022
  • Glyph-Based Visual Analysis of Q-Leaning Based Action Policy Ensembles on Racetrack.
    David Groß; Michaela Klauck; Timo P. Gros; Marcel Steinmetz; Jörg Hoffmann; Stefan Gumhold
    26th International Conference Information Visualisation, IV 2022. 1-10
    PDF
    conference
    IV
    doi: 10.1109/IV56949.2022.00011
  • Pattern Selection Strategies for Pattern Databases in Probabilistic Planning.
    Thorsten Klößner; Marcel Steinmetz; Álvaro Torralba; Jörg Hoffmann
    Thirty-Second International Conference on Automated Planning and Scheduling, ICAPS 2022. 184-192
    PDF
    conference
    ICAPS
  • Debugging a Policy: Automatic Action-Policy Testing in AI Planning.
    Marcel Steinmetz; Daniel Fiser; Hasan Ferit Eniser; Patrick Ferber; Timo P. Gros; Philippe Heim; Daniel Höller; Xandra Schuler; Valentin Wüstholz; Maria Christakis; Jörg Hoffmann
    Thirty-Second International Conference on Automated Planning and Scheduling, ICAPS 2022. 353-361
    PDF
    conference
    ICAPS
  • Neural Network Action Policy Verification via Predicate Abstraction.
    Marcel Vinzent; Marcel Steinmetz; Jörg Hoffmann
    Thirty-Second International Conference on Automated Planning and Scheduling, ICAPS 2022. 371-379
    PDF
    conference
    ICAPS
  • Expressivity of Planning with Horn Description Logic Ontologies.
    Stefan Borgwardt; Jörg Hoffmann; Alisa Kovtunova; Markus Krötzsch; Bernhard Nebel; Marcel Steinmetz
    Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022. 5503-5511
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V36I5.20489
  • Classical Planning with Avoid Conditions.
    Marcel Steinmetz; Jörg Hoffmann; Alisa Kovtunova; Stefan Borgwardt
    Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022. 9944-9952
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V36I9.21232
2021
  • Making DL-Lite Planning Practical.
    Stefan Borgwardt; Jörg Hoffmann; Alisa Kovtunova; Marcel Steinmetz
    18th International Conference on Principles of Knowledge Representation and Reasoning, KR 2021. 641-645
    PDF
    conference
    KR
    doi: 10.24963/KR.2021/61
  • Why Do I Have to Take Over Control? Evaluating Safe Handovers with Advance Notice and Explanations in HAD.
    Frederik Wiehr; Anke Hirsch; Lukas Schmitz; Nina Knieriemen; Antonio Krüger; Alisa Kovtunova; Stefan Borgwardt; Ernie Chang; Vera Demberg; Marcel Steinmetz; Jörg Hoffmann
    International Conference on Multimodal Interaction, ICMI 2021. 308-317
    PDF
    conference
    ICMI
    doi: 10.1145/3462244.3479884
  • Faster Stackelberg Planning via Symbolic Search and Information Sharing.
    Álvaro Torralba; Patrick Speicher; Robert Künnemann; Marcel Steinmetz; Jörg Hoffmann
    Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021. 11998-12006
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V35I13.17425
  • Pattern Databases for Goal-Probability Maximization in Probabilistic Planning.
    Thorsten Klößner; Jörg Hoffmann; Marcel Steinmetz; Álvaro Torralba
    Thirty-First International Conference on Automated Planning and Scheduling, ICAPS 2021. 201-209
    PDF
    conference
    ICAPS
2020
  • Deep Statistical Model Checking.
    Timo P. Gros; Holger Hermanns; Jörg Hoffmann; Michaela Klauck; Marcel Steinmetz
    40th International Conference on Formal Techniques for Distributed Objects, FORTE 2020. 96-114
    PDF
    conference
    FORTE
    doi: 10.1007/978-3-030-50086-3_6
  • Towards Dynamic Dependable Systems Through Evidence-Based Continuous Certification.
    Rasha Faqeh; Christof Fetzer; Holger Hermanns; Jörg Hoffmann; Michaela Klauck; Maximilian A. Köhl; Marcel Steinmetz; Christoph Weidenbach
    9th International Symposium on Leveraging Applications of Formal Methods, ISoLA 2020. 416-439
    PDF
    conference
    ISoLA
    doi: 10.1007/978-3-030-61470-6_25
  • TraceVis: Towards Visualization for Deep Statistical Model Checking.
    Timo P. Gros; David Groß; Stefan Gumhold; Jörg Hoffmann; Michaela Klauck; Marcel Steinmetz
    9th International Symposium on Leveraging Applications of Formal Methods, ISoLA 2020. 27-46
    PDF
    conference
    ISoLA
    doi: 10.1007/978-3-030-83723-5_3
  • A New Approach to Plan-Space Explanation: Analyzing Plan-Property Dependencies in Oversubscription Planning.
    Rebecca Eifler; Michael Cashmore; Jörg Hoffmann; Daniele Magazzeni; Marcel Steinmetz
    Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020. 9818-9826
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V34I06.6534
  • Let's Learn Their Language? A Case for Planning with Automata-Network Languages from Model Checking.
    Jörg Hoffmann; Holger Hermanns; Michaela Klauck; Marcel Steinmetz; Erez Karpas; Daniele Magazzeni
    Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020. 13569-13575
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V34I09.7083
  • Plan-Space Explanation via Plan-Property Dependencies: Faster Algorithms & More Powerful Properties.
    Rebecca Eifler; Marcel Steinmetz; Álvaro Torralba; Jörg Hoffmann
    Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI 2020. 4091-4097
    PDF
    conference
    IJCAI
    doi: 10.24963/IJCAI.2020/566
  • Bridging the Gap Between Probabilistic Model Checking and Probabilistic Planning: Survey, Compilations, and Empirical Comparison.
    Michaela Klauck; Marcel Steinmetz; Jörg Hoffmann; Holger Hermanns
    Journal of Artificial Intelligence Research. 2020, 68, 247-310
    PDF
    journal
    doi: 10.1613/JAIR.1.11595
2019
  • Towards automated network mitigation analysis.
    Patrick Speicher; Marcel Steinmetz; Jörg Hoffmann; Michael Backes; Robert Künnemann
    34th ACM/SIGAPP Symposium on Applied Computing, SAC 2019. 1971-1978
    PDF
    conference
    SAC
    doi: 10.1145/3297280.3297473
  • The 2019 Comparison of Tools for the Analysis of Quantitative Formal Models - (QComp 2019 Competition Report).
    Ernst Moritz Hahn; Arnd Hartmanns; Christian Hensel; Michaela Klauck; Joachim Klein; Jan Kretínský; David Parker; Tim Quatmann; Enno Ruijters; Marcel Steinmetz
    Tools and Algorithms for the Construction and Analysis of Systems - 25 Years of TACAS: TOOLympics, TACAS 2019. 69-92
    PDF
    conference
    TACAS
    doi: 10.1007/978-3-030-17502-3_5
  • Bridging the Gap between Abstractions and Critical-Path Heuristics via Hypergraphs.
    Marcel Steinmetz; Álvaro Torralba
    Twenty-Ninth International Conference on Automated Planning and Scheduling, ICAPS 2019. 473-481
    PDF
    conference
    ICAPS
2018
  • Formally Reasoning about the Cost and Efficacy of Securing the Email Infrastructure.
    Patrick Speicher; Marcel Steinmetz; Robert Künnemann; Milivoj Simeonovski; Giancarlo Pellegrino; Jörg Hoffmann; Michael Backes
    2018 IEEE European Symposium on Security and Privacy, EuroS&P 2018. 77-91
    PDF
    conference
    EuroS&P
    doi: 10.1109/EUROSP.2018.00014
  • Stackelberg Planning: Towards Effective Leader-Follower State Space Search.
    Patrick Speicher; Marcel Steinmetz; Michael Backes; Jörg Hoffmann; Robert Künnemann
    Thirty-Second AAAI Conference on Artificial Intelligence, AAAI 2018. 6286-6293
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V32I1.12090
  • Compiling Probabilistic Model Checking into Probabilistic Planning.
    Michaela Klauck; Marcel Steinmetz; Jörg Hoffmann; Holger Hermanns
    Twenty-Eighth International Conference on Automated Planning and Scheduling, ICAPS 2018. 150-154
    PDF
    conference
    ICAPS
  • Simulated Penetration Testing as Contingent Planning.
    Dorin Shmaryahu; Guy Shani; Jörg Hoffmann; Marcel Steinmetz
    Twenty-Eighth International Conference on Automated Planning and Scheduling, ICAPS 2018. 241-249
    PDF
    conference
    ICAPS
  • On Stubborn Sets and Planning with Resources.
    Anna Wilhelm; Marcel Steinmetz; Jörg Hoffmann
    Twenty-Eighth International Conference on Automated Planning and Scheduling, ICAPS 2018. 288-297
    PDF
    conference
    ICAPS
  • LP Heuristics over Conjunctions: Compilation, Convergence, Nogood Learning.
    Marcel Steinmetz; Jörg Hoffmann
    Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI 2018. 4837-4843
    PDF
    conference
    IJCAI
    doi: 10.24963/IJCAI.2018/672
2017
  • Beyond Red-Black Planning: Limited-Memory State Variables.
    Patrick Speicher; Marcel Steinmetz; Daniel Gnad; Jörg Hoffmann; Alfonso Gerevini
    Twenty-Seventh International Conference on Automated Planning and Scheduling, ICAPS 2017. 269-273
    PDF
    conference
    ICAPS
  • Critical-Path Dead-End Detection versus NoGoods: Offline Equivalence and Online Learning.
    Marcel Steinmetz; Jörg Hoffmann
    Twenty-Seventh International Conference on Automated Planning and Scheduling, ICAPS 2017. 283-287
    PDF
    conference
    ICAPS
  • Search and Learn: On Dead-End Detectors, the Traps they Set, and Trap Learning.
    Marcel Steinmetz; Jörg Hoffmann
    Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017. 4398-4404
    PDF
    conference
    IJCAI
    doi: 10.24963/IJCAI.2017/614
  • State space search nogood learning: Online refinement of critical-path dead-end detectors in planning.
    Marcel Steinmetz; Jörg Hoffmann
    Artificial Intelligence. 2017, 245, 1-37
    PDF
    journal
    doi: 10.1016/J.ARTINT.2016.12.002
2016
  • Partial Delete Relaxation, Unchained: On Intractable Red-Black Planning and Its Applications.
    Daniel Gnad; Marcel Steinmetz; Mathäus Jany; Jörg Hoffmann; Ivan Serina; Alfonso Gerevini
    Ninth Annual Symposium on Combinatorial Search, SOCS 2016. 45-53
    PDF
    conference
    SOCS
    doi: 10.1609/SOCS.V7I1.18391
  • Towards Clause-Learning State Space Search: Learning to Recognize Dead-Ends.
    Marcel Steinmetz; Jörg Hoffmann
    Thirtieth AAAI Conference on Artificial Intelligence, AAAI 2016. 760-768
    PDF
    conference
    AAAI
    doi: 10.1609/AAAI.V30I1.10062
  • Revisiting Goal Probability Analysis in Probabilistic Planning.
    Marcel Steinmetz; Jörg Hoffmann; Olivier Buffet
    Twenty-Sixth International Conference on Automated Planning and Scheduling, ICAPS 2016. 299-307
    PDF
    conference
    ICAPS
  • Combining the Delete Relaxation with Critical-Path Heuristics: A Direct Characterization.
    Maximilian Fickert; Jörg Hoffmann; Marcel Steinmetz
    Journal of Artificial Intelligence Research. 2016, 56, 269-327
    PDF
    journal
    doi: 10.1613/JAIR.5057
  • Goal Probability Analysis in Probabilistic Planning: Exploring and Enhancing the State of the Art.
    Marcel Steinmetz; Jörg Hoffmann; Olivier Buffet
    Journal of Artificial Intelligence Research. 2016, 57, 229-271
    PDF
    journal
    doi: 10.1613/JAIR.5153
Software

PFD (Probabilistic Fast Downward)
cpp
python

Extension of the Fast Downward planning system by the support of PPDDL encoded MDPs. Features a variety of heuristic search engines (like LAO*, LRTDP, FRET, i-dual) and a variety of heuristics (e.g., based on all-outcomes determinization and the large space of classical planning heuristics that readily ship with Fast Downward; probabilistic pattern databases; and probabilistic occupation-measure heuristics).

Thesis version JAIR16 version

CDL-FD (Conflict-Driven Learning Fast Downward)
cpp
python

Extension of the Fast Downward planning system by conflict-driven learning methods akin to conflict-driven clause learning in SAT. In a nutshell, in general state-space search, conflicts become dead ends, i.e., states without any solution path. CDL-FD ships variants of all common heuristic search algorithm (A*, GBFS) and heuristic-guided depth-first searches that identify such conflicts while search is running; refining a dead-end detector upon each such identification. Multiple dead-end detectors are supported (based on critical-path heuristics; dead-end traps; potential heuristics; and the state-equation heuristic).

Thesis version

ASNet Policy Testing
cpp

The Neural-FD based framework for testing ASNet policies.

ICAPS22 version

PPDDL Benchmark Suite
pddl

Comprehensive set of PPDDL benchmarks composed of benchmarks from the IPPCs, Canadian travaler adaptations of resource-constrained classical planning benchmarks (NoMystery, Rovers, and TPP), and an MDP encoding of simulated network penetration testing. Each domain comes with a limited-budget as well as an unlimited-budget version. The suite consists of a suite of acyclic benchmarks (where the state space is acyclic, allowing particular algorithm enhancements) and cyclic benchmarks.

Thesis version JAIR16 acyclic JAIR16 cyclic
Contact
Marcel Steinmetz
Robotics and InteractionS (RIS) Group
LAAS-CNRS
7 Avenue du Colonel Roche
31031 Toulouse cedex 4, France
marcel.steinmetz@laas.fr
© 2024 Marcel Steinmetz