An edgewise path selection method for robot obstacle crossing, a chip, and a robot. The method comprises: first, planning edgewise prediction paths for robot obstacle crossing, and selecting, on a preset navigation path, preset inflection points satisfying a guide condition, wherein a navigation path formed by connecting inflection points is preset for the robot; these inflection points are used for guiding the robot to move to the final navigation target point; then according to information of the distance between all the edgewise behavior points on each edgewise prediction path, and preset inflection points satisfying the guide condition on the same navigation path, an edgewise prediction path having the minimum deviation degree relative to the navigation path, so that the robot walks in the edgewise direction of the selected edgewise prediction path after colliding with an obstacle, so as to achieve rapid obstacle crossing. The number of collisions of the robot in the obstacle crossing process is reduced, the robot is also prevented from deviating too far from the navigation path, and the intelligent level of the robot is improved.