<p>In emergency obstacle avoidance scenarios, drivers often make errors due to panic or misjudgment, which can significantly impact the safety of the driver and surrounding traffic. To address this issue, this paper proposes an online method for identifying erroneous driving behaviors based on vehicle dynamics. This method comprehensively considers the lateral saturation characteristics of tires under varying vertical loads and road adhesion coefficients. By integrating the driver’s input of the steering angle, a controllable domain boundary that changes in real-time with the driver’s actions is constructed. The relationship between the vehicle state and the controllable domain is quickly determined through an analytical calculation method, enabling accurate identification of erroneous driver actions. Multi-condition experiments were conducted on a driver-in-the-loop platform, and the results indicate that, compared to existing identification methods, the proposed method can identify driver errors more quickly and accurately, providing a rational basis for system intervention and effectively avoiding human–machine conflicts. Additionally, this method provides the system with more ample activation time to achieve better control performance, demonstrating good real-time capabilities and proving its effectiveness in real-world driving environments.</p>