REVECA: Adaptive Planning and Trajectory-based Validation in Cooperative Language Agents using Information Relevance and Relative Proximity

TL;DR

REVECA addresses cooperative planning under partial observability by integrating three components: Relevance Estimation to prune memory, Adaptive Planning to account for collaborator proximity, and Trajectory-based Validation to prevent false plans. The framework uses six modules (Communication, Observation, Memory, Planning, Validation, Execution) to enable efficient information sharing, memory management, and plan execution via a low-level skill book and A* navigation. Across partially observable environments (C-WAH, Noisy-C-WAH, TDW-MAT) and a fully observable game (Overcooked-AI), REVECA outperforms baselines in success rates and efficiency, with ablation analyses highlighting the importance of CoT prompting, proximity, and validation. A user study further indicates higher perceived appropriateness, usefulness, efficiency, and trust when collaborating with REVECA, underscoring its potential for trustworthy human-AI cooperation in complex multi-agent tasks.

Abstract

We address the challenge of multi-agent cooperation, where agents achieve a common goal by cooperating with decentralized agents under complex partial observations. Existing cooperative agent systems often struggle with efficiently processing continuously accumulating information, managing globally suboptimal planning due to lack of consideration of collaborators, and addressing false planning caused by environmental changes introduced by other collaborators. To overcome these challenges, we propose the RElevance, Proximity, and Validation-Enhanced Cooperative Language Agent (REVECA), a novel cognitive architecture powered by GPT-4o-mini. REVECA enables efficient memory management, optimal planning, and cost-effective prevention of false planning by leveraging Relevance Estimation, Adaptive Planning, and Trajectory-based Validation. Extensive experimental results demonstrate REVECA's superiority over existing methods across various benchmarks, while a user study reveals its potential for achieving trustworthy human-AI cooperation.

Figures (25)

• Figure 1: The REVECA process workflow ensures efficient memory management, optimal planning, and cost-effective prevention of false planning through three phases: (a) Observation Time, (b) Planning Time, and (c) Validation Time.
• Figure 2: Comparative results in the Noisy-C-WAH environment with varying dummy objects.
• Figure 3: User study results in C-WAH environment. The mean scores and associated standard errors for responses to four research questions.
• Figure 4: An example scenario demonstrating REVECA’s comprehensive operational flow, highlighting the interaction between various modules, collaborators, and the environment to achieve a common goal.
• Figure 5: The example layouts of the C-WAH and Noisy-C-WAH environments.