In order to solve the configuration planning problem for intelligent search and rescue robots with assisted flippers to achieve autonomous obstacle crossing, a novel method for planning robot configuration during obstacle crossing was proposed that can be applied to complex terrains. The core of the proposed method is an adaptable and efficient robot pose prediction algorithm. By representing the terrain as a series of discrete point sets, a mathematical model for predicting the one-sided pose of the tracked robot was established; further, a fast solver for this model was proposed, which can predict 1000~1500 poses per second. Based on this, the evaluation metrics of the robot′s state and action in the obstacle-crossing process were established, and an optimization-based real-time flippers action planner was realized by using the dynamic programming algorithm and rolling optimization. The simulation and real-robot experiments show that the proposed approach enables the robot to control the flippers to cross rough terrains autonomously. It performs more smoothly than the reinforcement-learning method and manual operation when crossing obstacles.