# Lama-dt outperforms baselines on meta-world and mujoco

Digest AI · Research · published 2026-11-04T05:00:00Z

Canonical: https://digestai.news/story/lama-dt-outperforms-baselines-on-meta-world-and-mujoco

## Summary

The paper proposes LaMA-DT, a unified sequence‑modeling framework that uses a single large language model for trajectory augmentation and policy learning. Experiments on Meta‑World and MuJoCo multi‑task benchmarks show that LaMA‑DT consistently outperforms competitive baselines, especially in low‑data regimes.

The approach leverages natural‑language task descriptions to guide task‑aligned augmentation and includes a scoring‑and‑filtering mechanism to retain high‑quality trajectories. By repurposing the same LLM as a language‑conditioned Decision Transformer, the framework improves cross‑task generalization across heterogeneous offline datasets.

## Key points

- LaMA-DT uses a single LLM for trajectory augmentation and policy learning.
- It leverages natural‑language task descriptions for task‑aligned augmentation.
- Experiments on Meta‑World and MuJoCo show improved cross‑task generalization.

## Why it matters

The approach shows that language guidance can boost offline RL, potentially reducing data needs and improving generalization across tasks, a key challenge in reinforcement learning.

## Sources

1. [Language Model Augmented Decision Transformer for Multi-Task Offline Reinforcement Learning](https://arxiv.org/abs/2611.04031) (arXiv cs.CL, 2026-11-04, primary source)

## Cite

Digest AI, "Lama-dt outperforms baselines on meta-world and mujoco", 4 November 2026, https://digestai.news/story/lama-dt-outperforms-baselines-on-meta-world-and-mujoco

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/lama-dt-outperforms-baselines-on-meta-world-and-mujoco.json