The authors propose offline trajectory generalization through World Transformers to improve the generalization capability of offline reinforcement learning methods.
