Workshop on Computational Models of Human Road User Behavior for Autonomous Vehicle Evaluation

Tuesday, September 30, 2025 | 09:00 - 15:30

Computational road user behavior models play a key role in the evaluation of autonomous vehicles (AVs), both for establishing AV behavior benchmarks and for representing other road users in simulation-based testing. However, the development of road user models that meet the needs for AV evaluation is challenging. Road user behavior is complex and involves a wide range of aspects such as decision-making, adaptation in uncertain environments, visual behavior, attention, social interaction with other road users, and conflict- and collision avoidance responses. Due to this complexity, mechanistic road user behavior models, which explicitly represent underlying mechanisms, tend to be constrained to specific aspects of behavior and/or specific scenarios. More recently, with the advent of generative AI, data-driven road user behavior models have demonstrated remarkable abilities to learn complex behaviors in diverse scenarios from large quantities of data, especially in everyday “normal” traffic. However, it is still unclear to what extent these models can accurately capture long-tail behaviors that are not well represented in the training data, such as the emergence of critical conflicts, human responses to these, and crash outcomes. One potentially promising approach is to integrate mechanistic and data-driven models, but work of this nature is limited so far. For the abovementioned reasons, fully generalizable road user behavior models, able to account for all aspects relevant for AV evaluation, are still lacking. This workshop rests on the assumption that to fully meet the modelling needs, considerable cross-disciplinary exchange will be required. The main objective of the workshop is to facilitate such exchange, by bringing together experts on road user behavior modeling who use a variety of approaches (mechanistic, machine-learned, or combinations of both), as well as current and future users of models (especially in industry and government), to share the latest advances and to discuss open issues to be addressed.

List of topics
Mechanistic modeling; Data-driven modeling; Combined mechanistic/data-driven modeling; Methods for validating that human models are good enough for AV testing (scenarios, datasets, metrics, crash outcome, ...); Human modeling across both routine and critical situations; Human driver models as behavioral reference models for AVs; Human virtual agents in simulated AV testing; Use of human models by different stakeholders

Format
The workshop will be a mix of invited expert talks, panel discussions, and talks and/or posters on submitted contributions.