Tag
A tweet compared SKY-001 with GPT-6 in terms of speed and cost on an engine part drawing task—51 times faster and 150 times cheaper—and concluded that this low-cost approach would be widely replicated, while traditional industry remained stuck asking superficial questions.
欧洲工业巨头正与 NVIDIA 合作，利用 CUDA-X、Omniverse、Cosmos 和 Nemotron 等技术革新工程与制造流程，并推出工厂运营蓝图 FOX，西门子、梅赛德斯-奔驰、Cadence 和 Autodesk 等企业已采用相关方案以提升生产力并降低成本。
本文提出一个可观察证据标准，用于评估工业AI规划系统中新增能力是否真正带来运营价值，并通过Favorita、M5和Walmart数据的回放实验揭示单一指标评估的三大局限。
German Chancellor Friedrich Merz and NVIDIA CEO Jensen Huang discussed Germany's industrial AI roadmap, culminating in a Joint Declaration of Intent between the German government and NVIDIA to advance physical AI models and build a state-of-the-art industrial AI ecosystem in Europe.
本文提出 FINDERS 框架，利用扩散语言模型的迭代去噪过程构建多粒度重建空间，通过去噪、分解、路由与压缩四个操作实现面向多场景的自适应用户表征，并在十个工业场景的实验与线上 A/B 测试中超越现有方法。
MIT Technology Review's Business Lab examines how advances in foundation models, physical AI, and agentic AI are reshaping industrial automation, with AVEVA's Arti Garg discussing responsible deployment, human oversight, and an IEEE methodology for measuring AI's environmental footprint.
Vantora, a startup builder formerly known as UP.Labs, raised $100 million from Silversmith Capital Partners and is shifting its focus to building proprietary physical AI startups for corporate customers.
This paper proposes a multimodal anomaly detection framework for fault detection in mechanical systems using self-supervised cross-modal reconstruction and adaptive thresholding to improve robustness under distribution shifts.
This paper proposes matched-record evaluation for text classifiers in industrial settings, demonstrating that record selection significantly impacts performance metrics across maintenance, safety, and recall systems.
The paper presents SAAC-JEPA, a schema-adaptive action-conditioned JEPA model for transferring predictive representations across CNC machines with partial sensor overlap, demonstrating that cross-machine adaptation requires distinct evaluation beyond source-domain accuracy.
This paper empirically compares planning and reinforcement learning methods for scheduling maintenance in multi-asset industrial systems, highlighting trade-offs between reliability and cost, and suggests they are complementary based on operational objectives.
This academic paper proposes an AI-based system that uses thermal video footage to estimate combustion efficiency in flare stacks, integrated into a graphical user interface with high uptime and low maintenance.
Skild AI announced reaching 100 million annual recurring revenue within 10 months of deployments, highlighting their AI solutions in factories, construction sites, kitchens, and data centers.
The paper presents Conservative Hybrid Graph Network (CHGN), a method for modeling dynamic process systems that allows extrapolation to unseen larger graphs without retraining and improves fault prediction in industrial applications.
FedCMAPSS introduces a benchmark for federated learning in remaining useful life estimation based on the NASA C-MAPSS dataset, with standardized tasks to evaluate federated optimization algorithms across neural architectures.
Perceptron, a startup founded by ex-Meta scientists, has launched Isaac 0.5, an open-weight visual AI model designed to help robots perceive, reason, and act in industrial environments like warehouses and factories.
The APAC-Egocentric-Stereo dataset provides first-person recordings of people working in industrial and other jobs, featuring stereo video, depth maps, tracking, and captions, now available in FiftyOne format on Hugging Face.
This paper proposes candidate-fate accounting, a framework for auditing automated machine learning pipeline searches in sensor diagnostics to enhance transparency by tracking all candidates and their outcomes.
This paper presents RecSys Factory, an LLM-agent platform deployed at Tencent that confines agent autonomy to decision points rather than full pipelines, balancing autonomy, determinism, and efficiency across three industrial recommender business lines.
This paper compares three methods for grounding a frozen LLM (Qwen2.5-32B-Instruct) in a wastewater treatment simulator for causal question answering, including live tool-use, structured parameter injection, and a small decoupled retrieval model, achieving high accuracy and cross-plant transfer.