PODAR: A Collision Risk Model Offering Valid Signals for Vehicular Interactions

Acquiring valid signals for collision risk during driving is essential for the design of safety assistance or self-driving systems. Currently, there is a gap in accessible models that offer rational alerts for diverse driving scenarios. This article introduces a generalized, yet concise framework called the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">potential damage risk</i> (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">PODAR</i>) model for estimating collision risk and issuing warnings of hazard cases. The PODAR model is devised by connecting collision risk with potential collision damage across both spatial and temporal dimensions in a meaningful physical way, which includes four key components: trajectory prediction, damage estimation, spatiotemporal attenuation, and attention simulation. An empirical model featuring six adjustable parameters is presented and calibrated using a publicly available dataset for validation. With predefined fixed risk thresholds, the PODAR model demonstrates the ability to issue valid warnings for potential collisions in both longitudinal and lateral directions, negating the need for manual driving situation classification. The study investigates the performance of the PODAR model in a range of dynamic driving situations through numerical simulations, such as side-pass, car-following, and an unsignaled left turn. Additionally, three demonstrative cases derived from the InD dataset, which features realistic driving scenarios, are also presented. The results reveal that the PODAR model is effective in estimating risk and detecting collisions, providing reliable signals for potential collisions. In addition, the PODAR model, as a generalized framework, has the ability to incorporate existing metrics and exhibits impressive scalability to integrate new advancements in trajectory prediction and human-like risk perception, thanks to its modular structure.