Text this: Enhancing generalization in reinforcement learning for robotic control through deep policy similarities