Automated Driving System (ADS) is a safety-critical software system responsible for the interpretation of the vehicle’s environment and making decisions accordingly. The unbounded complexity of the driving context, including unforeseeable events, necessitate continuous improvement, often achieved through iterative DevOps processes. However, DevOps processes are themselves complex, making these improvements both time- and resource-intensive. Automation in code generation for ADS using Large Language Models (LLM) is one potential approach to address this challenge. Nevertheless, the development of ADS requires rigorous processes to verify, validate, assess, and qualify the code before it can be deployed in the vehicle and used. In this study, we developed and evaluated a prototype for automatic code generation and assessment using a designed pipeline of a LLM-based agent, simulation model, and rule-based feedback generator in an industrial setup. The LLM-generated code is evaluated automatically in a simulation model against multiple critical traffic scenarios, and an assessment report is provided as feedback to the LLM for modification or bug fixing. We report about the experimental results of the prototype employing Codellama:34b, DeepSeek (r1:32b and Coder:33b), CodeGemma:7b, Mistral:7b, and GPT4 for Adaptive Cruise Control (ACC) and Unsupervised Collision Avoidance by Evasive Manoeuvre (CAEM). We finally assessed the tool with 11 experts at two Original Equipment Manufacturers (OEMs) by conducting an interview study.

While the supervisory control (SC) problem of (maximally permissive) deadlock avoidance for sequential resource allocation systems (RASs) has been extensively studied in the literature, the corresponding results that are able to address potential resource outages are quite limited, both, in terms of their volume and their control capability. This paper leverages the recently developed SC theory for switched discrete event systems (s-DES) in order to provide a novel systematic treatment of this more complicated version of the RAS deadlock avoidance problem. Following the modeling paradigm of s-DES, both the operation of the considered RAS and the corresponding maximally permissive SC policy are decomposed over a number of operational modes that are defined by the running sets of the failing resources. In particular, the target supervisor must be decomposed to a set of “localized predicates,” where each predicate is associated with one of the operational modes. A significant part, and a primary contribution, of this paper concerns the development of these localized predicates that will enable the formal characterization and the effective computation of the sought supervisor. With these predicates available, a distributed representation for the sought supervisor that is appropriate for real-time implementation is eventually obtained through an adaptation of the relevant distributed algorithm that is provided by the current s-DES SC theory.

In order to develop a computationally efficient implementation of the maximally permissive deadlock avoidance policy (DAP) for complex resource allocation systems (RAS), a recent approach focuses on the identification of a set of critical states of the underlying RAS state-space, referred to as minimal boundary unsafe states. The availability of this information enables an expedient one-step-lookahead scheme that prevents the RAS from reaching outside its safe region. The work presented in this paper seeks to develop a symbolic approach, based on binary decision diagrams (BDDs), for efficiently retrieving the (minimal) boundary unsafe states from the underlying RAS state-space. The presented results clearly demonstrate that symbolic computation enables the deployment of the maximally permissive DAP for complex RAS with very large structure and state-spaces with limited time and memory requirements. Furthermore, the involved computational costs are substantially reduced through the pertinent exploitation of the special structure that exists in the considered problem. Note to Practitioners-A key component of the real-time control of many flexibly automated operations is the management of the allocation of a finite set of reusable resources among a set of concurrently executing processes so that this allocation remains deadlock-free. The corresponding problem is known as deadlock avoidance, and its resolution in a way that retains the sought operational flexibilities has been a challenging problem due to: (i) the inability to easily foresee the longer-term implications of an imminent allocation and (ii) the very large sizes of the relevant state spaces that prevent an online assessment of these implications through exhaustive enumeration. A recent methodology has sought to address these complications through the offline identification and storage of a set of critical states in the underlying state space that renders efficient the safety assessment of any given resource allocation. The results presented in this paper further extend and strengthen this methodology by complementing it with techniques borrowed from the area of symbolic computation; these techniques enable a more compressed representation of the underlying state spaces and of the various subsets and operations that are involved in the pursued computation.

A generic state-vector transition (SVT) model is suggested, including a flexible synchronous composition involving both shared variables and events. This model is analyzed, focusing on properties that are important for supervisor synthesis. A synthesis procedure is then developed for the SVT model, where supervisor guards are generated that guarantee a controllable, nonblocking and maximally permissive supervisor. Novel conditions are introduced, such that more flexible specifications can be applied than earlier suggested for related models. Since the SVT model includes automata and (colored) Petri nets, optionally extended with variables, guards and actions, as special cases, the suggested synthesis approach unifies supervisor synthesis for the main discrete event model classes. Finally, the SVT model is naturally represented and efficiently computed based on binary decision diagrams, and the resulting supervisor guards are easily implemented in industrial control systems.

In this paper, we symbolically represent timed discrete-event systems (TDES), which can be used to efficiently compute the supervisor in the supervisory control theory context. We model a TDES based on timed extended finite automata (TEFAs): an augmentation of extended finite automata (EFAs) by incorporating discrete time into the model. EFAs are ordinary automata extended with discrete variables, where conditional expressions and update functions can be attached to the transitions. The symbolic computations are based on binary decision diagrams (BDDs). We show how TEFAs can be represented by BDDs. The main feature of this approach is that the BDD-based fixed point computations are not based on tick models that have been commonly used in this area, leading to better performance in many cases. The approach has been implemented and applied to a simple case study and several large-scale benchmarks.

In supervisory control theory, a supervisor restricts the plant in order to fulfill given specifications. A problem for larger industrial applications is that the resulting supervisor is not easily implemented and comprehensible for the users. To tackle this problem, an efficient method has recently been introduced to characterize a supervisor by tractable logic conditions, referred to as guards. This approach has been developed for a specific type of automata with variables called extended finite automata (EFAs). An extension of this approach to a more general class of models is presented in this paper. It means that classical supervisory control problems for automata and Petri nets are easily and efficiently solved, but also generalized based on the suggested approach. The synthesis procedure is naturally modeled and efficiently computed based on binary decision diagrams.

Supervisory Control Theory (SCT) is a model based framework for automatically synthesizing a supervisor that minimally restricts the behavior of a plant such that a given specification is fulfilled. A problem, which prevents SCT from having a major breakthrough industrially, is that the supervisory synthesis often suffers from the state-space explosion problem. To alleviate this problem, a well-known strategy is to represent and explore the state-space symbolically by using Binary Decision Diagrams. Based on this principle, an efficient symbolic state-space traversal approach, depending on the disjunctive partitioning technique, is presented and the correctness of it is proved. Finally, the efficiency of the presented approach is demonstrated on a set of benchmark examples.