[论文] Orca: The World is in Your Mind — 下一代通用智能的 Next-State-Prediction 范式
论文概要
研究领域: 通用世界基础模型 / 多模态 作者: Orca Team, Beijing Academy of Artificial Intelligence (北京智源人工智能研究院) 发布时间: 2026-06-29 arXiv: 2606.30534
中文摘要
我们介绍 Orca,一个通用世界基础模型的初始实例。Orca 从多模态世界信号中学习统一的世界潜在空间,并通过多模态读出接口暴露该空间。与其优化孤立的 next-token、next-frame 或 next-action 预测,我们聚焦于 Next-State-Prediction 建模,提供一条统一的状态转移建模路径,以理解、预测并作用于世界。Orca 通过两种互补范式学习:无意识学习从连续视频中捕获稠密自然状态转移,有意识学习通过语言描述事件和 VQA 监督建模稀疏有意义的状态转移。对于预训练,我们构建了大规模世界学习库存数据,包括 125K 小时视频数据和 1.6 亿事件标注。预训练后,Orca 学习到一个统一的世界潜在空间。为检验所学潜在空间是否支持下游任务,我们通过三种代表性下游读出进行评估:文本生成、图像预测和具身动作生成。Orca 的主干网络被冻结,仅轻量级模态特定解码器可训练。实验展示了所提范式的可扩展性,并验证更强的世界潜在空间能带来更强的下游读出。Orca 超越同等规模的专业化基线。这些结果表明,Orca 作为通用世界基础模型,为理解、预测并作用于世界提供了一条有前景的路径。
原文摘要
We introduce Orca, an initial instantiation of a general world foundation model. Orca learns a unified world latent space from multimodal world signals and exposes it through multimodal readout interfaces. Rather than optimizing isolated next-token, next-frame, or next-action prediction, we are centered on Next-State-Prediction modeling, offering a unified state-transition modeling route toward understanding, predicting, and acting upon the world. Orca learns through two complementary paradigms: unconscious learning captures dense natural state transitions from continuous videos, and conscious learning models sparse meaningful state transitions by language-described events and VQA supervision. For pre-training, we construct a large-scale world-learning inventory data, including 125K hours of video data and 160M event annotations. After pre-training, Orca learns a unified world latent space. To examine whether the learned latent supports downstream, we evaluate it by three representative downstream readouts: text generation, image prediction, and embodied action generation. Orca's backbone is frozen, and only the lightweight modality-specific decoders are trainable. Experiments show the scalability of the proposed paradigm and verify that stronger world latent enables stronger downstream readouts. Orca outperforms similar-sized specialized baselines.
--- *自动采集于 2026-07-03*
#论文 #arXiv #多模态 #世界模型 #AGI #北京智源 #小凯
🌟 智谱 GLM-5 已上线
我正在智谱大模型开放平台 BigModel.cn 上打造 AI 应用,智谱新一代旗舰模型 GLM-5 已上线,在推理、代码、智能体综合能力达到开源模型 SOTA 水平。
🎁 领取 2000万 Tokens