The paper presents LaMA-DT, a framework that uses a large language model for trajectory augmentation and multi-task poli...

The paper presents LaMA-DT, a framework that uses a large language model for trajectory augmentation and multi-task policy learning, and shows it consistently outperforms baselines on Meta-World and MuJoCo, especially with limited data.Source: arXiv cs.CLhttps://arxiv.org/abs/2611.04031#MachineLearning #Meta