Project Details
VESCAA: Verifikovatelná a efektivní syntéza kontrolerů
Project Period: 1. 3. 2023 – 31. 12. 2025
Project Type: grant
Code: GA23-06963S
Agency: Czech Science Foundation
Program: Standardní projekty
Decision making under uncertainty; controller design; safety and scalalbility;
inductive synthesis; reinforcement learning, risk-aware learning;
Many modern computing systems can be seen as (semi)-autonomous agents interacting
with their environment. The agent's behaviour is determined by a controller that
necessarily needs to deal with uncertainties including unpredictability of the
environment and the imprecision of data gathered about its current state. There
exists a multitude of approaches to automated controller design, however, they
all tackle the safety-scalability gap: scalability limits the complexity of the
problems that can be handled and safety ensures that agent operates in a safe and
interpretable way. There are two principal approaches: formal methods prioritize
safety and reinforcement learning prioritizes scalability.
The project aims at developing theoretical foundation and synthesis algorithms
that reduce this gap and thus improve their practical applicability. The key idea
is to adapt, further develop and synergically integrate two emerging paradigms:
inductive synthesis improving the scalability of correct-by-construction design
techniques and risk-aware learning improving the safety guarantees.
2024
- ANDRIUSHCHENKO, R.; ČEŠKA, M.; MACÁK, F.; JUNGES, S. Policies Grow on Trees: Model Checking Families of MDPs. Proceeding of 22nd International Symposium on Automated Technology for Verification and Analysis. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Cham: Springer Verlag, 2024.
p. 0-0. Detail
2023
- ANDRIUSHCHENKO, R.; ALEXANDER, B.; ČEŠKA, M.; JUNGES, S.; KATOEN, J.; MACÁK, F. Search and Explore: Symbiotic Policy Synthesis in POMDPs. In Computer Aided Verification. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Cham: Springer Verlag, 2023.
p. 113-135. ISBN: 978-3-031-37708-2. Detail - ANDRIUSHCHENKO, R.; BARTOCCI, E.; ČEŠKA, M.; FRANCESCO, P.; SARAH, S. Deductive Controller Synthesis for Probabilistic Hyperproperties. In Quantitative Evaluation of SysTems. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Cham: Springer Verlag, 2023.
p. 288-306. ISBN: 978-3-031-43834-9. Detail