A driving authority interaction decision model based on fuzzy cognitive map (FCM) was proposed to address the dynamic transition from autonomous driving to driver takeover in human-machine co-driving scenarios. Firstly, dynamic risk factors influencing driving risks for both the vehicle and the driver were separately analyzed. Relevant influencing indicators such as vehicle spacing, acceleration, steering wheel entropy, driver eye movement information, and electroencephalogram (EEG) signals were selected to construct FCM-based models for dynamic vehicle risk and driver risk. Subsequently, these models were integrated with environmental factors, vehicle conditions, and driver proficiency to establish an FCM-based driver interaction decision-making model. To reduce reliance on expert knowledge, the model was optimized using a real coded genetic algorithm (RCGA) based on real number encoding, thereby enhancing the accuracy of the model's decision-making. The simulation results indicate that the model, after optimization using RCGA, achieves an adaptation fitness value of 0.979. This suggests that the model is capable of effectively addressing dynamic and complex driving environments.