Enhanced Overcooked-AI Benchmark
收藏资源简介:
Enhanced Overcooked-AI Benchmark是由南加州大学开发的一个用于评估实体代理在协作场景中实时适应和通信能力的数据集。该数据集包含22个不同复杂度和协作需求的布局,旨在测试代理在动态环境中的反应性适应和有效通信。数据集的创建过程结合了GPT-4模型生成布局,并通过脚本筛选和人工审查确保布局的适用性。该数据集主要应用于评估和提升实体代理在动态协作场景中的适应性和通信效率,旨在解决现有基准在评估实时适应性和动态通信效率方面的不足。
The Enhanced Overcooked-AI Benchmark, developed by the University of Southern California, is a dataset for evaluating the real-time adaptation and communication capabilities of embodied agents in collaborative scenarios. This dataset contains 22 layouts with varying complexity and collaborative requirements, whose core objective is to test agents' reactive adaptation and effective communication in dynamic environments. The development of this dataset leverages GPT-4 to generate layouts, and ensures the suitability of these layouts through script-based filtering and manual review. This dataset is primarily applied to evaluate and improve the adaptability and communication efficiency of embodied agents in dynamic collaborative scenarios, aiming to address the limitations of existing benchmarks in assessing real-time adaptability and dynamic communication efficiency.

- 1Benchmark Real-time Adaptation and Communication Capabilities of Embodied Agent in Collaborative Scenarios南加州大学 · 2024年



