Specification-guided temporal logic control for stochastic systems: a multi-layered approach

TL;DR

This paper introduces a multi-layered abstraction approach for designing controllers that satisfy temporal logic specifications in stochastic systems, reducing conservatism and improving computational efficiency.

Contribution

It develops a flexible multi-layered framework combining discretization-based and discretization-free abstractions guided by specifications, enhancing controller synthesis for stochastic systems.

Findings

Reduces conservatism in controller design.

Improves computation time and memory usage.

Demonstrates effectiveness on case studies.

Abstract

Designing controllers to satisfy temporal requirements has proven to be challenging for dynamical systems that are affected by uncertainty. This is mainly due to the states evolving in a continuous uncountable space, the stochastic evolution of the states, and infinite-horizon temporal requirements on the system evolution, all of which makes closed-form solutions generally inaccessible. A promising approach for designing provably correct controllers on such systems is to utilize the concept of abstraction, which is based on building simplified abstract models that can be used to approximate optimal controllers with provable closeness guarantees. The available abstraction-based methods are further divided into discretization-based approaches that build a finite abstract model by discretizing the continuous space of the system, and discretization-free approaches that work directly on the continuous state space without the need for building a finite space. To reduce the conservatism in the sub-optimality of the designed controller originating from the abstraction step, this paper develops an approach that naturally has the flexibility to combine different abstraction techniques from the aforementioned classes and to combine the same abstraction technique with different parameters. First, we develop a multi-layered discretization-based approach with variable precision by combining abstraction layers with different precision parameters. Then, we exploit the advantages of both classes of abstraction-based methods by extending this multi-layered approach guided by the specification to combinations of layers with respectively discretization-based and discretization-free abstractions. We achieve an efficient implementation that is less conservative and improves the computation time and memory usage. We illustrate the benefits of the proposed multi-layered approach on several case studies.