Lama-dt outperforms baselines on meta-world and mujoco
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.
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.
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.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Anthropic's Claude begins 'dreaming' to reorganize memory in tests · 1 src
- John Langford quits Microsoft, declares AI Frontiers over · 1 src
- Anthropic releases study on Claude usage in scientific research · 3 src
- MIT and BASF release CheMeleon for low-data molecular prediction · 2 src
- Anthropic tests Claude agents in EVE Online trading and sabotage detection · 1 src
Comments
via GitHub Discussions