1 Insights into Scaling Virtual Power Plants Real-World Findings for Successful Deployment January 2025 Angela Long (Rockcress Consulting) Ryan Long (Rockcress Consulting) PacifiCorp Stacy Miller, U.S. DOE Tolu Omotoso, NRECA Tomas Smith, Puget Sound Energy 5 5 REAL-WORLD FINDINGS FOR SUCCESSFUL DEPLOYMENT Insights into Scaling Virtual Power Plants Executive Summary locationality to their VPP). PGE is enhancing operations at the grid edge for flexibility. This involves real-time computation and communication at the device level to manage customer usage patterns dynamically

18 技能泛化：组合泛化实现未知技能的学习 n RA-P (IROS 2025, NeurIPS 2024 OWA): composable generalizable agents in real world n Decompose complicated tasks into fine-grained primitive skills, generalizable to new physical 2025 技能泛化：组合泛化实现未知技能的学习 n RA-P (IROS 2025, NeurIPS 2024 OWA): composable generalizable agents in real world n Decompose complicated tasks into fine-grained primitive skills, generalizable to new physical 2025 技能泛化：组合泛化实现未知技能的学习 n RA-P (IROS 2025, NeurIPS 2024 OWA): composable generalizable agents in real world n n Decompose complicated tasks into fine-grained primitive skills, generalizable to new

lines, processes, and equipment. Employing advanced diagnostic tools, such as thermal imaging and real-time energy monitoring, enhances the accuracy and depth of the audit. These tools help uncover hidden controls. Variable frequency drives (VFDs) can further reduce energy waste by adjusting air output to real-time demand. Installing local boosters for equipment requiring higher pressure helps avoid raising overall energy use. ii All payback period data and estimates are based on practical experience from real-world cases. In individual cases, the actual payback time can vary significantly, as it depends on

Public administration Manufacturing Agriculture Accommodation and food services Construction and real estate $2,000 $1,500 $1,000 $500 $0 2024 2025 2026 2027 2028 2029 2030 10% 6% 11% 13% 18% which offers 3 Gbps downlink and 200 Mbps uplink speeds, alongside AI- enhanced features such as real-time translation and cloud computing services. The operator aims to have more than 20 million 5G-Advanced logistics. In smart manufacturing, China Mobile has deployed AI-driven industrial IoT platforms that use real-time data from 5G-connected sensors to optimise production lines. In a partnership with a major automotive

Systems (MES) track and document the transforma- tion of raw materials into finished goods, providing real-time production manage- ment to drive enterprise-wide compliance, quality and efficiency. Distributed Monitoring provides seamless connectivity to machines on the plant floor, delivering transparent, real-time operational KPIs like Overall Equipment Effectiveness (OEE). Design & Visualization tools that acquire, process and monitor operating data. Smart Manufacturing is the intelligent, real-time orchestration and optimization of business, physical and digital processes within factories

具身智能目前最⼤的瓶颈在于数据，各层面均处于早期阶段，如何权衡 真实数据和模拟数据，已成为⾏业内发展路径差异的关键分歧 18 信息来源：量⼦位智库，1）Sim2Real（Simulation to Reality）是指将⼈⼯智能模型从模拟环境（simulation）训练中获得的知识和能⼒转移到现实世界（real world）中应用的过程 数据构成 精简度 • 数据规模小：具身智能的数 据积累需要从零做起，数据 稀少，尤其端到端的⾼质量 误差积累 • 成本⾼速度慢， 需雇佣⼤量数采 ⼈员，搭建采集 环境，以⼈/天 为单位采集数据 • 基于虚拟场景模 拟现实中的物理 规则或特定任务 环境（如⼯厂、 家庭） • 中，Sim2Real1的 分布偏移问题难 以解决 • 低成本且更敏捷， 仿真数据的成本、 采集速度远低于 真实数据 • 通过训练其他⼈ 类或者本体的任 务操作类视频进 ⾏学习 • 低，数据质量参 差不齐，需要⼤ 视频 数据 • 真实数据是最佳 选择，仿真数据 主要是出于数采 成本过⾼的现实 选择，⼤部分具 身智能厂商都在 同时使用真实数 据和仿真数据， 但侧重不同 • 仿真路线的核⼼ 是解决Sim2Real 中间的差距，也 需要结合少量⾼ 质量的真实数据 做数据增强 数据积累 规模 数据闭环 成熟度 空间智能概览 自动驾驶 3D⽣成 具身智能 扩展现实（XR） 世界模型 扩展现实（XR）

digital twin configuration in figure 1 could have included the deployed finished product, for example. Real-time information about how a product actually performs in the field could prove useful to future Holdowsky et al., Inside the Internet of Things (IoT). 16. Data can be analyzed either as a batch or in real time or as a combination (lambda architecture). Batch analytics is an efficient way of processing over the day. In comparison, real-time analytics involves generating insights about a physical asset through continuous processing of incoming data pertaining to the asset. Real-time analytics might be used

platform has functions such as real-time collection, monitoring, analysis, and management of environmental data. Scientific applications can achieve intelligent control and real-time monitoring of environmental

Suite – Kernel Real-Time Application 32/64 bit Kernel DLL MDF/PCAPng Kithara RealTime Suite – API User Mode Control Layer and External Libraries Communication and Fieldbus Layer Real-Time Layer Hardware Servers 边缘服务器 确定性实时控制 工厂 OT/IT 网络 边缘工作节点 [n] m g m t Real Time HW/SW General Purpose Compute AI Control HMI m g m t Real Time HW/SW General Purpose Compute AI Control HMI 边缘工作节点[1] 械臂的整体运动，从而整体实现使 用自然语言对机械臂的操控。 LLM RTmotion EtherCAT/CANopen ASR Code Generator Real-Time Non-Real-Time CPU P & E Cores iGPU CPU E Cores NPU Intel® Core™ Ultra ARL-H Audio Prompts

process to prompt the generator to generate images progressively resembling the distribu- tion of real data [7, 8]. Moreover, the diffusion models stand out as the most revolutionary technologies that SketchyGAN [37] focuses on the sketch-to-image generation task and aims to achieve more diversity and real- ism. Currently, ControlNet [38] can control diffusion mod- els by adding extra conditions. The sketch-to-image mindsets and explore new design spaces. Simul- taneously, diffusion probabilistic models can generate real- istic architectural rendering images suitable for Virtual Re- ality (VR) or Augmented Reality (AR)