i4Driving project: Establishing a credible human safety baseline for CCAM evaluation

We are delighted to feature the i4Driving project in the CCAM Partnership’s “Project in the Spotlight” series. This Horizon Europe-funded initiative is advancing the state of the art in virtual assessment and safety benchmarking for Connected, Cooperative, and Automated Mobility (CCAM) systems by developing a robust set of methodologies and tools grounded in realistic human driving behaviour.

Launched on 1 October 2022, i4Driving (Integrated 4D driver modelling under uncertainty) unites a multidisciplinary consortium of academic research teams, industry partners, and experts in traffic engineering, human factors, simulation, and data science from across Europe and beyond. Over its three-year and a half duration, the project aims to lay the foundation for an industry-standard methodology to establish a credible and realistic human road safety baseline that supports the virtual evaluation and benchmarking of CCAM systems.

Addressing uncertainty in human behaviour

A key challenge for safe deployment of automated driving functions is accounting for the inherent uncertainty and diversity of human driving behaviour. Traditional simulation frameworks often lack the complexity needed to reflect real-world driving dynamics, especially under mixed traffic conditions involving both human drivers and automated systems. i4Driving tackles this by combining:

1. A multi-level, modular and extensible simulation library that integrates existing and new driver behaviour models, and
2. A cross-disciplinary methodology to rigorously address uncertainties in both human behaviours and driving scenarios. This ensures that models are not only data-driven but also scientifically grounded and relevant for safety assessment.

By capturing the full spectrum of driving performance — from everyday interactions to safety-critical situations — i4Driving enables more realistic, transparent, and replicable virtual testing of CCAM systems compared to benchmarks based on simplistic or uniform driver models. This approach helps to expose automated systems to a broad range of traffic conditions and behavioural patterns, including those that can trigger edge cases.

Towards a human-centred safety baseline

The outputs of the i4Driving project are designed to serve multiple stakeholders across the CCAM ecosystem. A credible human safety baseline can:

• Inform type approval and regulatory frameworks by providing evidence-based safety criteria against which automated functions can be assessed.
• Guide consumer testing campaigns and industry validation processes by articulating performance standards rooted in realistic driver behaviour.
• Support developers of automated driving systems by offering a rigorous reference for virtual testing and model comparison before and alongside physical trials.

The project’s methodology also includes the development of structured use case definitions, scenario generation processes, and heterogeneous driver behaviour characterisation to reflect a wide and inclusive range of driving styles and conditions, as well as a novel Turing test-inspired approach for subjective evaluation of the realism of driver models.

Collaborative research and impact

i4Driving’s consortium brings together expertise in data collection, advanced simulation, experimental design, and safety assessment from multiple countries and research backgrounds. The project fosters international cooperation — including connections with driving simulation facilities and research partners beyond Europe — ensuring broad scientific and practical relevance.  Active alignment bridge has been established between the i4Driving project and other EU Horizon projects. For example, the alignment discussion with the SUNRISE project enabled the i4Driving project to incorporate the latest harmonised safety assurance framework into the i4Driving application demonstration work.

Ultimately, the i4Driving project contributes to a safer, more robust, and scientifically grounded framework for assessing CCAM technologies, supporting the long-term goal of wider deployment of connected and automated mobility solutions. Its innovative modelling and uncertainty quantification approach paves the way for more dependable virtual validation environments, better regulatory alignment, and increased trust in automated systems among public authorities, industry stakeholders, and road users alike.

Key takeaways

The i4Driving project has developed the following key insights on 4Ddriver modelling under uncertainty:

Behavioural mechanisms
i4Driving augments traditional collision-free “engineering” driver models by integrating human factors and behavioural mechanisms, enabling a more realistic representation of driver behaviour.

Data collection
i4Driving has developed methodologies for designing and conducting data collection experiments to effectively support model testing and validation.

Model development and verification
i4Driving establishes a comprehensive methodology to corroborate models through targeted experiments, based on an iterative process of testing and refinement, supported by large-scale Monte Carlo simulations, calibration, and global sensitivity analysis.

Extending virtual safety ADS assesment
i4Driving models and validation methodology allow extending current virtual safety assessment of ADS:

• enabling context-aware, closed-loop multi-agent simulations with mutually interacting agents.
• supporting population-level benchmarks through transparent metrics based on outcome distributions (e.g. median and upper-quantile targets).
• promoting a Safety II perspective, focusing on system performance across a wide range of conditions rather than only on failure cases.

Discover more

Visit our YouTube channel to discover how these key takeaways—and many more findings—were developed throughout the i4Driving project:

• 6 Feb – Advanced simulation of a human driver model with human factors
• 13 Feb – Experimental data collection using simulators & test tracks
• 20 Feb – A methodology to assess and compare longitudinal driver models
• 27 Feb – How realistic are behavioural driver models?
• 6 Mar – Global Sensitivity analysis for multiverse analysis
• 13 Mar – Human reference models reliability UNR157 / Augmented human driver enhancement
• 20 Mar – Overview and applications of the i4Driving model
• 31 Mar – Key takeaways & relevance for industry and policy

Want to read more about i4Driving? Click here for reports and publications.