Improving generalization in reinforcement learning through forked agents