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

English title
VESCAA: Verifiable and Efficient Synthesis of Agent Controllers
Type
grant
Keywords

Decision making under uncertainty; controller design; safety and scalalbility;
inductive synthesis; reinforcement learning, risk-aware learning;

Abstract

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.

Team members
Publications

2024

2023

Back to top