To quantify the risk levels and characterize the distribution patterns of takeover maneuvering behaviors during the human-machine control transition, six typical takeover scenarios covering urban roads and expressways as well as three types of non-driving tasks were designed, followed by a series of driving simulator experiments. Considering the safety impacts of dynamic and static road users as well as lane markings on the ego vehicle, a manual takeover risk field model was established to quantitatively grade the takeover risks encountered by drivers. This study further revealed the evolutionary mechanism of characteristic takeover parameters under various non-driving task conditions, and finely profiled high-risk takeover samples from multiple dimensions including driving trajectory, vehicle speed, acceleration, and steering wheel operation. The experimental results indicate that low-risk samples account for the largest proportion (44.8%), followed by medium-risk samples (40.8%) and high-risk samples (14.4%). Takeover scenarios, non-driving task conditions, and driver attributes exert significant effects on the distribution of takeover risk levels. High-risk samples are accompanied by abnormal driving behaviors such as lane crossing and near-collision risks, and are featured with higher average travel speed, larger lateral and longitudinal accelerations, as well as elevated steering wheel turning rate and steering angle entropy during the takeover process. The findings of this research are of great significance for uncovering the evolutionary law of drivers’ takeover maneuvering characteristics under human-machine co-driving conditions and improving the safety and ride comfort of vehicle takeover operations.