Periodic Reporting for period 1 - BERTHA (BEhavioural ReplicaTion of Human drivers for CCAM)
Período documentado: 2023-11-01 hasta 2025-04-30
One of the main shortcomings in CCAM development is the absence of a validated and scientifically grounded Driver Behavioural Model (DBM) capable of covering key aspects of human driving performance. Such a model would allow for a safer and more predictable understanding and testing of CCAM interactions with other vehicles, from a human-centred perspective.
The DBM is essential for CCAM component development. It ensures digital validation and, when integrated into ECUs (Electronic Control Units), enables more human-like responses from autonomous vehicles at any automation level, increasing user acceptance.
The primary objective of BERTHA is to develop a scalable and probabilistic DBM, based predominantly on Bayesian Belief Networks (BBNs). This DBM will be implemented in an open-source HUB, which uses a Software-in-the-Loop (SiL) simulation framework, to validate both its technological and practical feasibility in collaboration with industry stakeholders, offering a unique approach for the model’s global scalability. The resulting DBM will also be translated into a simulation platform (CARLA) through diverse demonstrators, enabling the development of new driving models within the platform.
BERTHA will also incorporate a methodology that, due to the HUB, will share the model with the scientific community to ease its growth.
The project includes a set of interrelated demonstrators designed to showcase the DBM approach as a reference for designing human-like, predictable, and widely acceptable behaviours in automated driving functions within mixed traffic environments.
Ultimately, the project will benefit society by enabling safer, more human-like connected and automated vehicles (CAVs) and fostering greater acceptance among all road users.
WP2 made solid progress toward establishing a comprehensive methodology to support the development of the BERTHA Driver Behavioural Model (DBM). Key achievements include the definition of the BERTHA Data Model and the common data acquisition principles, which have aligned metrics, formats, and partner contributions. Experimental plans for both basic and advanced laboratory tests were designed, including scenarios, technologies, and procedures. Preparations for Field Operational Tests (FOTs) were also advanced, detailing participant and equipment, and alignment with use cases. In parallel, a functional prototype for data storage and exchange was implemented.
WP3 established the foundational HUB architecture for a Software-in-the-Loop (SIL) simulation framework. Initial efforts integrated SCENIC with the description of compatible scenarios in CARLA, leading to successful incorporation into the HUB's pipeline. A preliminary data recording and KPI calculation system was also developed.
WP4 successfully completed the design and creation of scenes and scenarios for CARLA simulator, laying the groundwork for future development and testing activities. A thorough evaluation of CARLA’s maps and scenario programming approaches led to the selection and adaptation of both ScenarioRunner and SCENIC tools, with SCENIC proving especially suitable for implementing the defined Use Cases. Maps were partitioned to separate development and testing environments generating ~1,000 parameterized scenarios. Initial efforts began to integrate the DBM with CARLA, including ROS 2 interface design and motor-control module testing
WP5 concentrated on evaluating current validation procedures and identifying new requirements for safety testing in the context of CCAM technologies. An extensive review of existing standards, protocols, and regulations was conducted, resulting in a comprehensive and up-to-date database. Particular focus was given to mixed traffic situations and the alignment with CCAM-related roadmaps, which enabled the identification of standardization gaps potentially addressable by DBM-based approaches. Additionally, Initial work began on defining representative safety tests, including parameter and equipment selection.
Its Driver Behavioural Model (DBM) goes beyond deterministic or black-box methods, describing physical, cognitive, emotional, personal, cultural, and contextual factors. Designed as a probabilistic model based on Bayesian Belief Networks, it offers transparent, explainable reasoning, adapts to new data, and allows modular validation. This will allow the DBM to grow up as new evidence is gathered, and at the same time to quantify the uncertainty associated with the individual drivers, environmental conditions, routes and interactions, etc.
The modular approach of the DBM has been successfully used by UGE’s COSMODRIVE model, based on the perception-cognition-action loop that characterizes information processing workflows, and is the basis of cognitive architectures in many areas of engineering. The modules of BERTHA’s DBM has an enlarged consideration of the aspects involved in the “cognition” block, which includes risk awareness, decision making and affective factors.
BERTHA’s DBM is one of the first cognitive architectures applied to CCAM that embeds emotional dynamics into real-time driving behaviour modelling, and considers a wide spectrum of driver typologies, based on a study that has involved over 4,689 drivers, moving beyond the basic demographic segmentation found in most current applications.
BERTHA’s methodology integrates simulators, lab testing, and Field Operational Tests. As a result, the consortium is developing new standardized safety validation methods that incorporate human behavioural factors (risk perception, distraction, decision-making), and virtual test procedures comparable to on-track homologation, enabling earlier, scalable, and more realistic validation.
BERTHA’s HUB is an open and decentralized platform designed for sharing models, scenarios, and test cases. It supports running simulations, logging results, and benchmarking performance. This enables closer collaboration between stakeholders through remote, collaborative, and scalable simulations, as well as the integration of tests involving persons.
The platform is built as a Software-in-the-Loop (SiL) simulation framework, allowing for validation and assessment of systems. It includes the monitoring of key performance indicators (KPIs) such as safety, human-likeness, and system performance, making it a powerful tool for the development and evaluation.