Answers are grounded in this paper and cite their sources.
In human–machine shared driving, takeover decision-making is affected by multi-source uncertainty from the environment, driver, and system. Existing studies lack integration of uncertainty modeling, evaluation, and the takeover process. This paper presents a focused review of recent literature from four aspects: uncertainty types, modeling methods, takeover decision-making process, and evaluation frameworks. It summarizes four sources of uncertainty: environment and perception, driver state, system and interaction, and situation. Modeling methods are grouped into probabilistic and Bayesian modeling, learning-based uncertainty quantification, cognitive modeling, and decision-theoretic methods. Research progress on pre-takeover risk prediction, takeover request timing, authority allocation, and trust calibration is reviewed. Current limitations include inconsistent evaluation metrics, poor cross-scenario generalization, and constraints on real-time deployment and explainability. Future work should strengthen multi-source fusion and online updating, human-adaptive takeover support, and standardized evaluation to improve the safety and timeliness of takeover decisions.
No takes yet. Share an insight, caveat, or question.
Aobo Shi (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: