Summary of Embodied Cot Distillation From Llm to Off-the-shelf Agents, by Wonje Choi et al.
Embodied CoT Distillation From LLM To Off-the-shelf Agents
by Wonje Choi, Woo Kyung Kim, Minjong Yoo, Honguk Woo
First submitted to arxiv on: 16 Dec 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Robotics (cs.RO)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents DeDer, a framework for decomposing and distilling large language models (LLMs) into efficient small language model (sLM)-based policies. This allows for utilizing LLMs on capacity-limited devices for complex embodied tasks. The framework consists of a reasoning-policy and planning-policy, where the reasoning-policy is distilled from an LLM’s in-context learning and self-verification, producing effective rationales. The planning-policy generates optimized plans efficiently, guided by these rationales. To enhance rationale quality, the paper introduces the embodied knowledge graph and contrastively prompted attention model. Experimental results on the ALFRED benchmark demonstrate that DeDer outperforms leading language planning and distillation approaches. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about making it possible to use big AI models for tasks that require thinking and acting in the real world, like robots or self-driving cars, without needing super powerful computers. The authors created a new way to break down these big models into smaller ones that can work on regular devices. This helps make it faster and more efficient to make decisions. They also came up with ways to improve the quality of explanations for why these decisions were made. |
Keywords
» Artificial intelligence » Attention » Distillation » Knowledge graph » Language model
