BERTHA: Advancing Human-Centred Validation for Automated Driving

As Connected, Cooperative and Automated Mobility (CCAM) technologies continue to advance, ensuring that automated vehicles can interact safely and predictably with human road users remains one of the sector’s biggest challenges. The European project BERTHA (BEhavioural ReplicaTion of Human Drivers for CCAM) was launched to address this challenge by placing human behaviour at the centre of automated driving research. Over three years, 14 partners from six European countries worked together to better understand how people perceive, assess risk, make decisions and interact with other road users, and how this knowledge can be integrated into the development and validation of automated driving systems.

 

Understanding Human Behaviour for Safer Automation

A key outcome of the project is the development of a modular and probabilistic Driver Behavioural Model (DBM) capable of representing the diversity of human driving behaviour. Rather than assuming a single “average” driver, a study conducted as part of BERTHA with  4,700 drivers from 5 countries identified eight distinct driver types, characterised by factors such as risk tolerance, attention, efficiency and interaction with other road users. The project also revealed differences in behavioural patterns across European countries, reinforcing the need to account for human diversity when designing and validating automated vehicles.

This human-centred approach enables developers and researchers to better predict how automated vehicles will interact with surrounding road users in real-world traffic situations, ultimately supporting safer and more trustworthy CCAM solutions.

Did you ever wonder which type of driver you are? Take our quick test and find out!

From Real-World Driving to Virtual Validation

To support more realistic testing and validation processes, BERTHA developed innovative perception and simulation tools that connect real-world driving observations with virtual testing environments.

Among its scientific achievements is DriverGaze360, a large-scale omnidirectional driver-attention dataset, together with DriverGaze360-Net, a deep-learning model capable of predicting driver attention across the full driving environment. These developments provide valuable insights into how drivers perceive and react to traffic situations, helping inform the design of automated driving functions that better align with human expectations and behaviour.

The project also developed methods to convert real-world driving observations into simulation-ready assets, enabling the creation of realistic virtual scenarios for testing automated driving systems under diverse traffic conditions.

Figure 1  DriveGaze360-Net Architecture

 

Read the full article here

 

The BERTHA HUB and Scenario Generation

A major project outcome is the BERTHA HUB, an integrated platform that combines behavioural models, scenario creation, simulation execution and data management into a single environment.

Supporting the BERTHA HUB’s data management and simulation capabilities, VEDECOM contributes a sovereign and scalable Big Data infrastructure, leveraging Docker and Kubernetes technologies to enable efficient data storage, processing and management across the platform.

At the heart of this ecosystem lies the ARISE (Adaptive Refinement and Iterative Scenario Engineering) framework, which enables the automatic generation of executable traffic scenarios from natural-language descriptions. This capability significantly streamlines the creation of testing scenarios, supports large-scale simulation campaigns and improves the reproducibility of validation activities.

By integrating behavioural realism into scenario generation and simulation workflows, BERTHA contributes to more robust and representative methods for assessing automated driving systems.

Figure 2 ARISE pipeline.

Read the full article here

 

Validation in Simulation and Real-World Environments

The project’s technologies were evaluated through extensive simulation and demonstration activities. More than 3,100 simulation runs were conducted across multiple safety-critical traffic scenarios, assessing the impact of behavioural diversity on automated vehicle performance. The components of the Driver Behavioural Model were also evaluated against data collected over more than 20 hours and 900 km of real-world driving during field operational trials conducted in urban and highway environments.

Beyond virtual testing, BERTHA’s Driver Behavioural Model was also integrated into a vehicle demonstrator and evaluated in controlled test-track environments. These demonstrations showed how behavioural modelling can improve the development and validation of automated driving functions while supporting safer interactions between automated vehicles and human road users.

Figure 3 BERTHA’s Driver Behavioural Model demonstrator during the final event’s exhibition/networking session.

 

Understanding User Acceptance

Recognising that the successful deployment of automated vehicles depends on public trust, BERTHA also examined how people perceive and accept different automated driving styles.

In acceptance tests with 88 participants at three sites (two driving simulators, where the driving styles were implemented through the DBMs, and one test track, where a professional driver reproduced the driving styles), around 2,800 scenario ratings were collected. In both settings, participants observed the driving behaviour without actively controlling the vehicle. Participants clearly distinguished between driving styles: cautious behaviour was rated considerably better than aggressive behaviour at every site, and pedestrians were more critical of aggressive behaviour than passengers. The results underline the importance of considering human expectations alongside technical performance when developing future CCAM solutions.

Conclusion

By bringing human behaviour into the heart of automated driving development and validation, BERTHA has advanced Europe’s capacity to design CCAM systems that are not only safe and efficient, but also understandable, predictable and accepted by society. Through its Driver Behavioural Model, behavioural datasets, simulation environment and innovative scenario-generation capabilities, the project has demonstrated how realistic representations of human driving behaviour can enhance the testing and assessment of automated vehicles.

Figure 4 Roundtable discussion at the BERTHA Final Event, featuring Espedito Rusciano (EC/JRC), Sergio Fernandes (CCAM), Laura Babio (POLIS Network) and José Solaz (IBV), moderated by Sofia Oliveira (FI Group by EPSA). The session explored key challenges and opportunities for strengthening public trust and acceptance of automated driving technologies in Europe.

 

BERTHA shows that the future of automated mobility depends not only on technological performance, but also on a deep understanding of how people perceive, decide and interact on the road. Its methodologies, tools and scientific results provide a foundation for more reliable virtual validation, more representative testing scenarios and greater public trust in automated driving technologies. As Europe moves towards the large-scale deployment of CCAM solutions, BERTHA’s legacy will continue to support the development of safer, more human-aware and socially accepted mobility systems.

Figure 5 BERTHA project consortium during the final event held at Les Ateliers des Tanneurs (Brussels, Belgium).

 

Find out more about the project at: https://bertha-project.eu/

Subscribe BERTHA newsletter at: https://berthaproject.eu/#newsletter