Adaptive Takeover Time Budgets in Conditionally Automated Driving

Abstract

Conditionally automated driving systems issue takeover requests (TORs) in situations that exceed their operational capabilities, requiring drivers to promptly resume manual control and maintain safe vehicle operation. A key factor in ensuring the smoothness of such control transitions is the time budget, i.e., the time offered by automation for control transitions. When the time budget is too short to accommodate the required takeover time (ToT, the time drivers need to regain manual vehicle control after receiving a TOR), the risk of accidents increases as drivers may lack adequate time to perceive, assess, and respond to the situation. Conversely, time budgets that substantially exceed the required ToT may also introduce risks: such TORs can be perceived as false alarms, leading to reduced driver attention and potential dangers, particularly when the out-of-capability situations are not readily perceivable to drivers. Therefore, defining and allocating sufficient time budgets is essential to ensure driving safety and user experience in vehicle control transitions.

This thesis systematically develops an adaptive framework for designing takeover time budgets that account for diverse drivers and situational demands. First, a systematic review synthesises the takeover sequence, identifying factors influencing takeover time and performance, and introduces the concept of the takeover buffer as the safety margin between required and allocated takeover time. Building on this foundation, a driving simulator experiment is conducted to collect behavioural, physiological, operational, and subjective data during takeover situations. Using these data, machine learning models are developed to predict takeover time, revealing that drivers’ perceived Spare Capacity provides substantial predictive power, while extensive driver profiling offers limited additional benefit. The thesis then establishes a multidimensional framework for takeover performance assessment, demonstrating that Situational Awareness primarily influences response efficiency, whereas Spare Capacity has a stronger impact on takeover quality. Finally, these insights are integrated into an adaptive time budget framework that combines predicted takeover time with a preferred takeover buffer to dynamically allocate time budgets.

The proposed framework enables personalised takeover time prediction, multidimensional performance evaluation, and adaptive time budget allocation in conditionally automated driving. In practice, these contributions can support cognition-aware vehicle interfaces, personalised takeover assistance systems, and human-centred automated driving design. Together, they contribute to safer, more reliable, and more comfortable control transitions, supporting the broader deployment and acceptance of automated vehicles.