PLAICraft
收藏资源简介:
PLAICraft是一个大规模的时间对齐的多模态数据集,旨在支持具身人工智能的研究。该数据集捕捉了多人Minecraft互动中的视频、游戏输出音频、麦克风输入音频、鼠标和键盘动作等五种时间对齐模态。数据集由超过10,000名全球参与者提供,总时长超过10,000小时。PLAICraft通过毫秒级时间精度记录每个模态,使得同步的具身行为研究成为可能。数据集还包括一个评估套件,用于基准测试模型在物体识别、空间感知、语言接地和长期记忆方面的能力。PLAICraft为训练和评估在实时中流畅、有目的地行动的智能体开辟了道路,为真正具身的人工智能铺平了道路。
PLAICraft is a large-scale temporally aligned multimodal dataset designed to support embodied artificial intelligence research. This dataset captures five temporally aligned modalities from multiplayer Minecraft interactions, including video, in-game audio output, microphone input audio, mouse and keyboard actions, and other relevant modalities. It is contributed by over 10,000 global participants, with a total duration exceeding 10,000 hours. PLAICraft records each modality with millisecond-level temporal precision, enabling research on synchronized embodied behaviors. The dataset also includes an evaluation suite for benchmarking model capabilities in object recognition, spatial perception, language grounding, and long-term memory. PLAICraft opens up new avenues for training and evaluating agents that act fluently and purposefully in real time, laying the foundation for truly embodied artificial intelligence.

- 1PLAICraft: Large-Scale Time-Aligned Vision-Speech-Action Dataset for Embodied AI不列颠哥伦比亚大学 · 2025年



