Atari Game Performance Dataset

该数据集汇总了多种多模态大型语言模型（LLMs）在多个Atari环境中的游戏性能数据，包括它们在1000步内的累积奖励。此外，数据集还包含了将LLMs与人类玩家、DQN智能体以及随机智能体的性能进行归一化比较的指标，同时具体评估了视觉和空间推理能力。该数据集的性能数据跨越了七个Atari环境，其任务是进行游戏性能评估。

This dataset aggregates game performance data of various multimodal Large Language Models (LLMs) across multiple Atari environments, including their cumulative rewards within 1000 steps. Additionally, the dataset contains normalized comparison metrics that benchmark the performance of LLMs against human players, DQN agents, and random agents, while specifically evaluating visual and spatial reasoning capabilities. The performance data contained in this dataset spans seven distinct Atari environments, and the dataset is designed for game performance assessment.