Autonomous driving requires predicting a safe and compliant trajectory to facilitate reliable driving decisions. Previous end-to-end autonomous driving approaches usually research in two directions, one is based on the planned trajectory, and the other is making control decisions directly. In this paper, the proposed autonomous driving model is a two-branch combining architecture. One is waypoint prediction branch, which uses Multilayer Perceptron(MLP) prediction to generate a sequence of future waypoints to obtain a safe and reliable trajectory by using three penalty constraint. The other branch is the Proportion Integration Differentiation(PID) controller, which dynamically adjusts the decision-making control command relying on the vehicle state and trajectory point and optimal waypoints prediction branch. The experimental evaluation on the CARLA simulator shows that the proposed approach achieves the best performance on closed-loop benchmarks over state of the art approaches.