Stefan Klikovits

Masaki Waga, Ezequiel Castellano, Sasinee Pruekprasert et al.

It is challenging to use reinforcement learning (RL) in cyber-physical systems due to the lack of safety guarantees during learning. Although there have been various proposals to reduce undesired behaviors during learning, most of these techniques require prior system knowledge, and their applicability is limited. This paper aims to reduce undesired behaviors during learning without requiring any prior system knowledge. We propose dynamic shielding: an extension of a model-based safe RL technique called shielding using automata learning. The dynamic shielding technique constructs an approximate system model in parallel with RL using a variant of the RPNI algorithm and suppresses undesired explorations due to the shield constructed from the learned model. Through this combination, potentially unsafe actions can be foreseen before the agent experiences them. Experiments show that our dynamic shield significantly decreases the number of undesired events during training.

Stefan Klikovits, Paolo Arcaini

This paper presents the first evaluation of k-nearest neighbours-Averaging (kNN-Avg) on a real-world case study. kNN-Avg is a novel technique that tackles the challenges of noisy multi-objective optimisation (MOO). Existing studies suggest the use of repetition to overcome noise. In contrast, kNN-Avg approximates these repetitions and exploits previous executions, thereby avoiding the cost of re-running. We use kNN-Avg for the scenario generation of a real-world autonomous driving system (ADS) and show that it is better than the noisy baseline. Furthermore, we compare it to the repetition-method and outline indicators as to which approach to choose in which situations.

Stefan Klikovits, Paolo Arcaini

Currently, most existing approaches for the design of Automated Driving System (ADS) scenarios focus on the description at one particular abstraction level typically the most detailed one. This practice often removes information at higher levels, such that this data has to be re-synthesized if needed. As the abstraction granularity should be adapted to the task at hand, however, engineers currently have the choice between re-calculating the needed data or operating on the wrong level of abstraction. For instance, the search in a scenario database for a driving scenario with a map of a given road-shape should abstract over the lane markings, adjacent vegetation, or weather situation. Often though, the general road shape has to be synthesized (e.g. interpolated) from the precise GPS information of road boundaries. This paper outlines our vision for multi-level ADS scenario models that facilitate scenario search, generation, and design. Our concept is based on the common modelling philosophy to interact with scenarios at the most appropriate abstraction level. We identify different abstraction levels of ADS scenarios and suggest a template abstraction hierarchy. Our vision enables seamless traversal to such a most suitable granularity level for any given scenario, search and modelling task. We envision that this approach to ADS scenario modelling will have a lasting impact on the way we store, search, design, and generate ADS scenarios, allowing for a more strategic verification of autonomous vehicles in the long run.

Stefan Klikovits, Paolo Arcaini

Multi-objective optimisation is a popular approach for finding solutions to complex problems with large search spaces that reliably yields good optimisation results. However, with the rise of cyber-physical systems, emerges a new challenge of noisy fitness functions, whose objective value for a given configuration is non-deterministic, producing varying results on each execution. This leads to an optimisation process that is based on stochastically sampled information, ultimately favouring solutions with fitness values that have co-incidentally high outlier noise. In turn, the results are unfaithful due to their large discrepancies between sampled and expectable objective values. Motivated by our work on noisy automated driving systems, we present the results of our ongoing research to counteract the effect of noisy fitness functions without requiring repeated executions of each solution. Our method kNN-Avg identifies the k-nearest neighbours of a solution point and uses the weighted average value as a surrogate for its actually sampled fitness. We demonstrate the viability of kNN-Avg on common benchmark problems and show that it produces comparably good solutions whose fitness values are closer to the expected value.

Clovis Eberhart, Akihisa Yamada, Stefan Klikovits et al.

We introduce a new logic named Quantitative Confidence Logic (QCL) that quantifies the level of confidence one has in the conclusion of a proof. By translating a fault tree representing a system's architecture to a proof, we show how to use QCL to give a solution to the test resource allocation problem that takes the given architecture into account. We implemented a tool called Astrahl and compared our results to other testing resource allocation strategies.