MSI-Agent: Incorporating Multi-Scale Insight into Embodied Agents for Superior Planning and Decision-Making
Dayuan Fu, Biqing Qi, Yihuai Gao, Che Jiang, Guanting Dong, Bowen Zhou
2024-09-30
Summary
This paper introduces MSI-Agent, a new type of agent designed to improve planning and decision-making for Large Language Models (LLMs) by effectively using insights from different scales.
What's the problem?
Agents that rely on insights for decision-making often struggle because they can receive irrelevant information or lack general insights. This can make it difficult for them to make effective decisions, especially in complex situations.
What's the solution?
The MSI-Agent addresses this issue by using a three-part system that includes an experience selector, an insight generator, and an insight selector. This system helps the agent summarize important information and store it in a database. When making decisions, the agent can retrieve relevant insights to guide its choices. The experiments showed that MSI-Agent performs better than other methods when planning tasks using LLMs like GPT-3.5.
Why it matters?
This research is important because it enhances the ability of agents to make better decisions by using relevant insights. By improving how LLMs plan and solve problems, MSI-Agent could lead to more effective applications in fields like robotics, virtual assistants, and any area where intelligent decision-making is required.
Abstract
Long-term memory is significant for agents, in which insights play a crucial role. However, the emergence of irrelevant insight and the lack of general insight can greatly undermine the effectiveness of insight. To solve this problem, in this paper, we introduce Multi-Scale Insight Agent (MSI-Agent), an embodied agent designed to improve LLMs' planning and decision-making ability by summarizing and utilizing insight effectively across different scales. MSI achieves this through the experience selector, insight generator, and insight selector. Leveraging a three-part pipeline, MSI can generate task-specific and high-level insight, store it in a database, and then use relevant insight from it to aid in decision-making. Our experiments show that MSI outperforms another insight strategy when planning by GPT3.5. Moreover, We delve into the strategies for selecting seed experience and insight, aiming to provide LLM with more useful and relevant insight for better decision-making. Our observations also indicate that MSI exhibits better robustness when facing domain-shifting scenarios.