Analyzing Generalization in Policy Networks: A Case Study with the Double-Integrator System