<p>To prevent collisions with a preceding vehicle under winter road conditions in cold regions, this study investigates longitudinal braking collision avoidance, lateral lane-change collision avoidance, and a decision-making strategy for collision avoidance modes of autonomous vehicles. Using a pendulum friction tester, the road adhesion coefficient on winter roads in cold regions was measured. Taking this coefficient into account, longitudinal and lateral safety distance models were established, and a collision avoidance mode decision strategy was designed based on the time-to-collision (TTC) model and the safety distance model. For longitudinal braking collision avoidance control, a hierarchical control approach was employed to develop a longitudinal graded braking controller that considers the road adhesion coefficient. For lateral lane-change collision avoidance, an adaptive-horizon model predictive control (MPC) controller was designed, with the horizon varying according to the road adhesion coefficient and the ego vehicle speed. Finally, the proposed collision avoidance control system was simulated and evaluated on a co-simulation platform. In the longitudinal collision avoidance system, scenarios were established including a braking preceding vehicle and a sudden change in adhesion coefficient, under various road adhesion coefficients and vehicle speeds The distances to the preceding vehicle after avoidance were 1.8 m, 1.095 m, and 2.74 m, respectively, confirming the rationality of the graded braking collision avoidance design. When longitudinal braking is insufficient to prevent a collision and the inter-vehicle distance falls between the minimum braking distance and the critical lane-change distance, the system activates the lateral collision avoidance mode. On medium- and low-adhesion road surfaces, the MPC controller accurately tracks the lane-change trajectory to complete the avoidance maneuver while maintaining satisfactory driving stability.</p>