[2306.13085] Harnessing Mixed Offline Reinforcement Learning Datasets via Trajectory Weighting