An efficient and versatile Digital Twin model implementation for autonomous connected vehicles based on distributed, low-latency working memory

Abstract

The growing interest in Connected Autonomous Vehicles (CAVs) has led to increased focus on technologies and algorithms that improve their behaviour, comfort, and safety. Central to these advancements is the application of Digital Twin (DT) models, an evolution of Cyber–Physical Systems (CPSs) that has attracted considerable attention in the scientific community. These DTs offer many possibilities by linking real-world activities with their twin counterparts, allowing for the anticipation of scenarios to prevent or improve the handling of different situations. This paper proposes a DT model paradigm for CAVs, supported by a distributed architecture of software agents. This architecture, named CORTEX, forms the core of our DT model and is characterised by its synchronisation capabilities, shared memory, versatility, performance and scalability. The proposed solution combines this distributed architecture with CARLA as its internal simulator. It uses probabilistic models to regulate and select optimal simulations for predicting risky situations during driving, among other capabilities. To validate our proposed DT model, we also present an algorithm that facilitates early detection of potential collisions between autonomous vehicles and pedestrians on their path by generating and simulating traffic scenes unsupervised and applying a particle filter-based methodology to evaluate risk situations. The results show that the proposed DT framework can effectively apply to autonomous driving systems. The DT architecture has been tested by a real electric autonomous vehicle on a university campus, demonstrating its effectiveness in anticipation and safe real-time decision-making.