您正在使用IE低版浏览器，为了您的雷峰网账号安全和更好的产品体验，强烈建议使用更快更安全的浏览器
此为临时链接，仅用于文章预览，将在时失效
0
导语：从底层芯片、神经形态计算，到机器人自主部署、极端环境物理智能和地表物理环境建模，人工智能前沿技术正沿着多条不同路径推进产业应用落地。如何让实验室里的原创成果跨越
从底层芯片、神经形态计算，到机器人自主部署、极端环境物理智能和地表物理环境建模，人工智能前沿技术正沿着多条不同路径推进产业应用落地。如何让实验室里的原创成果跨越工程验证与商业化之间的距离，成为科研创新走向产业实践的重要课题。
10月26日，北京智源人工智能研究院发布的“源创计划”首期路演在中关村国际技术交易中心举办。临界核科技、灵枢启智、张量场科技、苍穹极智、潜向量科技五个项目集中亮相，分别聚焦3D自选通V-RRAM芯片、神经形态具身智能、哈密顿世界模型、极端环境的可靠物理智能、地表物理环境统一世界模型等方向。活动还设置了项目与基金洽谈环节，推动创业团队与投资机构交流对接。
从“源头创新”到“产业验证”，源创计划要打通的不只是资金
活动现场，智源研究院副院长杨洋介绍了“源创计划”的定位与支持机制。作为主要面向智源学者群体的原创技术创业孵化计划，源创计划聚焦人工智能领域的原创科研课题，遵循“源头创新、快速决策、小步试行、全流程合规转化”的原则，整合技术、人才、资本与产业资源，为科研团队从实验室走向产业提供支持。计划不仅关注资金对接，更着力构建连接原创科研、创业团队与产业市场的协同机制，推动科研成果从技术发现走向工程验证与商业化应用。
源创计划的推出，也回应了人工智能产业发展与科技成果转化的现实需求。当前，AI正处于从技术突破迈向产业应用的关键阶段，原创技术的工程化验证、场景适配与商业化转化日益受到关注。与此同时，科技成果转化相关政策与机制持续完善，为科研机构探索原创成果转化提供了新的条件。源创计划希望把握这一窗口期，进一步打通科研创新与产业需求之间的连接，支持更多有潜力的原创技术走向市场。
五个项目，呈现AI原创技术的多元探索
1. 临界核科技：从原创器件到软硬件平台
临界核科技聚焦3DV-RRAM芯片，布局底层自选通RRAM器件、芯片、FlagOS软件生态与系统应用，致力于打造软硬件协同的全栈产品体系。项目以存储级内存和存算一体为切入点，探索新型存储与AI算力协同发展的新路径，旨在突破现有计算机体系结构面临的存储与算力瓶颈。团队由清华大学国家级青年人才领衔，曾获省部级自然科学一等奖(2025年)，兼具基础研究和工程技术攻关优势。
2. 灵枢启智：以神经形态计算重构具身智能感知架构
灵枢启智聚焦跨尺度多模态感知与神经形态具身智能，致力于为具身智能构建物理感知与类脑计算基础设施。项目从底层光学器件、感算一体芯片延伸至模型算法，探索从感知到决策的全栈技术架构。
其核心目标不仅是让智能体感知外部环境，还包括理解物质属性与人类意图，并在此基础上实现毫秒级自主决策。通过器件、芯片与算法的协同设计，项目探索提升具身智能环境理解与实时决策能力的新路径。
3. 张量场科技：以哈密顿世界模型推动机器人自主部署
张量场科技面向人因风险高、工程师难以反复调试的高价值场景，打造由哈密顿世界模型驱动的机器人部署引擎，提供环境建模、策略生成、端侧部署与持续进化等全链条服务。
团队提出哈密顿世界模型架构，通过物理先验约束提升模型输出的物理一致性。据项目介绍，其预训练数据效率较主流方案提升3-4个数量级。模型可适配多种机器人本体，项目还提出“部署即服务”模式，计划与机器人本体厂商、硬件供应商及云计算平台协同发展。团队曾获清华大学校长杯第三名，已完成天使轮融资；公司于2026年8月在北京海淀注册成立。
4. 苍穹极智：极端环境的可靠物理智能
面向深空、深地等极端环境，苍穹极智聚焦物理智能基础设施建设，试图解决极端场景下数据难以获取、模型难以训练、智能系统难以部署等问题。
团队提出AI4π（AI for Physical AI）技术范式，通过物理内嵌仿真引擎、自进化分层脑训推系统与泛本体操作系统，构建覆盖智能体设计、训练、测试与改进的技术闭环，探索突破物理智能研发中的数据与工程瓶颈。
团队由国家级领军人才领衔，依托北航与智源的科研资源，已获得顶会杰出论文、世界人形机器人运动会全球季军等十余项国际奖项。目前，项目通过仿真数据订阅与系统授权探索商业化，已在园区维保、商业航天等领域获得千万元级订单。
5. 潜向量科技：以地表物理环境统一世界模型连接虚拟与现实
潜向量科技聚焦地表物理环境统一世界模型，其团队自主研发EWM（Earth World Model），将大气强迫、太阳辐射等外部驱动因素，土壤、植被、水体、山地、碳循环等地表过程，以及电站、农田、工业与基础设施等人类系统纳入统一潜空间进行表征，探索构建可干预、可反演、可证伪的通用物理基座模型。
项目采用按Token计费的模型服务模式，新能源场景已在5个国家的12个站点完成实证验证。在光伏场景中，团队以自研Boson系列清洁与运维机器人为执行终端，与云端模型共同构建“感知—决策—执行—反馈”闭环，探索地表物理环境建模与实体智能系统协同落地的路径。
连接原创技术与资本，推动科研成果走向产业
从原创芯片到世界模型，从类脑具身到机器人自主部署和极端环境物理智能，五个项目呈现了人工智能原创技术在底层硬件、计算架构、模型研发与实体应用等方向的多元探索。它们的共同特点，是尝试将前沿科研能力转化为可验证、可部署、可持续发展的技术与产品。
对科研团队创办的硬科技企业而言，实验室里的技术突破只是起点。后续的工程验证、产品开发、场景适配与商业模式探索，都需要持续的资源投入和产业协同。如何有效连接科研创新、创业团队、资本与市场，是原创技术实现产业转化的重要环节。
此次源创计划路演通过项目展示与基金洽谈，为创业团队和投资机构搭建了交流对接的平台，推动原创技术与资本、产业资源进一步连接。活动展示了智源学者在不同技术方向上的原创探索，也呈现了从科研成果挖掘到创业项目培育的多元路径。
从实验室到产业一线，原创技术的转化既需要持续的科研突破，也需要工程能力、应用场景与产业资源的共同支撑。源创计划将继续围绕人工智能前沿原创成果，推动科研创新与产业需求有效衔接，支持更多有潜力的原创技术走向产业实践。
智源学者：从“梦之队”到“产业新势力”
智源学者计划是智源研究院打造高层次前沿基础研究人才队伍的引领性工程，旨在遴选并培养具有国际影响力的学术领军人才和具有发展潜力的青年学术英才，建立面向重大攻关任务和未来前沿的协同创新团队，引领未来人工智能技术发展方向，构建人工智能产业创新生态。七年来，智源学者计划历经自由探索、任务牵引、创新引领三个阶段迭代，已经成为吸引、培养人工智能领域顶尖人才的重要载体，智源学者数量已超过180人，建成了全球人工智能“梦之队”。
更重要的是，这支“梦之队”正在成为AI创业的重要源头。近年来，智源研究院已累计孵化智谱AI、面壁智能、银河通用、星源智、逆矩阵、子午线光年等20余家具有核心技术能力的创新企业。从大模型到具身智能，从底层视觉到安全治理，一批具有代表性的AI创业公司背后，都能看到智源学者和智源生态的身影。源创计划首期路演的落地，正是这一人才势能向产业动能转化的最新注脚。
为了您的账户安全，请验证邮箱
请验证您的邮箱
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
You are using a low-end IE browser and strongly recommend a faster and safer browser for your Thunder Peak account security and better product experience.
This is a temporary link, only for article preview, which will expire at the time
0
Introduction: From bottom chips, neuromorphological calculations, to robotic autonomous deployments, extreme environmental physical intelligence and geophysical modelling, artificial intelligence front-line technology is driving industrial applications down along multiple paths. How to get the originals across the lab.
From bottom chips, neurological calculations, to robotic autonomous deployments, extreme environmental physical intelligence and geophysical modelling, artificial intelligence frontier technologies are moving industrial applications down along multiple paths. How to move the original results from the laboratory from engineering to commercialization is an important topic in the evolution of scientific innovation to industrial practice.
On 26 October, the inaugural road show of the Generating Initiative of publish of the Beijing Institute of Intellectual and Intellectual Intelligence was held at the International Technology Exchange Centre in Nakawan. Critical nuclear technology, schizophrenia, chronology, accelerosis, and potential vector technology were highlighted in five projects focusing on the direction of the 3D self-selected V-RRAM chip, neuromorphologically intelligent, Hamilton World Model, reliable physical intelligence in extreme environments, and integrated world models of geophysics. The event also established a project-fund interface to facilitate the interface between the business team and the investment agency.
From "source innovation" to "industry certification," it's not just finance that's going to make it work.
At the event, Yang Yang, Vice-President of the Institute, presented the initiative's positioning and support mechanism. As an incubation programme for original technology, which is primarily aimed at intellectual-source scholars, it focuses on original scientific research in the area of artificial intelligence, following the principles of “source innovation, rapid decision-making, small-step piloting, full-process conversion”, integrating technology, talent, capital and industrial resources, and supporting research teams from laboratories to industry. The plan focuses not only on financial convergence, but also on building synergies between original research, entrepreneurship teams and industry markets, as well as promoting research from technological discovery to engineering validation and commercialization.
The launch of the Initiative also responds to the real needs for the development of the artificial intelligent industry and the transformation of scientific and technological results. At present, AI is at a critical stage in moving from a technological breakthrough to an industrial application, and there is growing interest in the engineering validation, adaptation and commercialization of original technologies. At the same time, policies and mechanisms for the conversion of scientific and technological results continue to improve, creating new conditions for scientific institutions to explore the transformation of original results. The Initiative hopes to take advantage of this window to further bridge the link between scientific innovation and industrial demand and to support more promising original technologies to market.
Five projects, multi-dimensional exploration of AI original technologies
Critical nuclear technology: from originals to soft and hardware platforms
Critical nuclear technology focuses on 3DV-RRAM chips, with the layout bottom-selected RRAM devices, chips, FlagOS software eco-system applications, and works to create a system of all-storey products that combines software and hardware. The project uses storage-level memory and computing as entry points to explore new pathways for the development of new types of storage in synergy with AI algorithms, with a view to overcoming storage and computing bottlenecks facing existing computer systems. The team was led by young people at the national level at the University of Qinghua and received a first prize in natural sciences from the provincial ministry (see annex). In 2025, with the advantage of both basic research and engineering techniques.
2. Spiritual inspiration: re-engineered body-smart sensory structure in neuromorphics
The core focus multimodular sensory and neuromorphological form is smart and is committed to building physical sensory and brain-based computing infrastructure for physical intelligence. The project extends from a bottom optical device, a sensory integrated chip, to a model algorithm to explore a whole-house technical architecture from perception to decision-making.
Its core objective is not only to give AI agent a sense of the external environment, but also to understand the physical properties and human intentions on which to base its decision-making in milliseconds. Through the synergistic design of the device, chip and algorithm, the project explores new pathways to enhance the ability to understand and make real-time decisions in a smart environment.
3. Zoom Field Technology: Auto-deployment of robots using the Hamilton World Model
Zoomfield technology is oriented towards high-value scenarios that people are exposed to because of the high risks and it is difficult for engineers to adapt over and over again, building robotic deployment engines driven by the Hamilton world model, providing full-chain services such as environmental modelling, strategy generation, end-side deployment and continuous evolution.
The team presented the Hamilton world model structure to improve the physical consistency of model outputs through a priori physical constraints. According to the project, its pre-training data are 3-4 orders of magnitude higher than the mainstream programme. Models can be adapted to a variety of robotics, and the project proposes a “deployment is service” model that plans to develop in collaboration with robotics owners, hardware suppliers and cloud computing platforms. The team received third place in the Cup of Rectors of Qinghua University and completed the Angel Wheel Finance; the company was registered in Haiti, Beijing, in August 2026.
4. Elevator intelligence: reliable physical intelligence in extreme environments
In extreme environments such as deep space and deep ground, accelerator-focused physical intelligence infrastructure is being built to try to address the problems of hard access to data in extreme settings, poor training of models and the difficulty of deployment of intelligent systems.
The team proposed a technical paradigm for AAI for Physical AI, which would build a technical loop covering AI agent design, training, testing and improvement through a physical emulation engine, a self-evolutionary stratification brain-motion system and a panthematic operating system, and explore data and engineering bottlenecks in the development of physical intelligence.
The team, which is led by national military leaders, has received more than a dozen international awards, based on the scientific resources of Norte and Iwigen, such as the Distinguished Distinction Paper, the World Human Robotic Games and the Global Seasons. Currently, the project explores commercialization through analogue data subscriptions and systems authorization, and has received millions of meta-orders in areas such as park protection and commercial space.
5. Potential vector technology: Connecting virtual and realistic world models with a unified surface physical environment
Potential technology focuses on integrated world models of the surface physical environment, where teams independently develop EWM (Earth World Model), integrate external drivers such as atmospheric coercion, solar radiation, surface processes such as soil, vegetation, water bodies, mountains and carbon cycles, as well as human systems such as power stations, agricultural land, industry and infrastructure into integrated subspaces, and explore the development of interventionable, reversible and perjurable generic physical base models.
The project uses a model service model based on Token, and the new energy scene has been validated empirically at 12 sites in 5 countries. In a light-voltage scenario, the team works with cloud-based models to construct a "sensor-decision-implementation-result" closed loop for the implementation of terminals by studying the Bson series of cleaning and transport robots, and explores the path of geophysical modelling in synergy with physical intelligence systems.
Connecting original technologies and capital to drive research to industry
From original chips to world models, from brain type to robotic autonomous deployment and extreme environmental physics, five projects present multiple explorations of artificially intelligent original technologies in the direction of bottom hardware, computing architecture, model development and physical applications. Their common feature is an attempt to transform forward-looking scientific capabilities into proven, deployable and sustainable technologies and products.
For hard-tech enterprises created by scientific teams, technological breakthroughs in laboratories are only the starting point. Subsequent engineering validation, product development, mapping and business model exploration require sustained resource inputs and industry synergies. How to effectively link scientific innovation, entrepreneurial teams, capital and markets is an important link in the transformation of original technologies.
The initiative led to a project demonstration with the Fund, which created a platform for entrepreneurship and investment agencies to interact and promote greater connectivity between original technologies and capital and industrial resources. The event demonstrated the original exploration of intellectuals in different technical directions, as well as the multifaceted path from the extraction of scientific results to the development of entrepreneurship projects.
From the laboratory to the industry, the transformation of original technologies requires both continuous scientific breakthroughs and a combination of engineering capabilities, applications and industrial resources. The Initiative will continue to focus on the cutting edges of artificial intelligence, promote an effective interface between scientific and technological innovation and industrial demand, and support more promising original technologies for industrial practice.
Genius Scholar: From Dream Team to New Power in Industry
The Genius Scholars Programme is the leading project of the Institute of Intellectual Resources to create a high-level front-line cadre of basic research personnel. It aims to select and develop academic leaders with international influence and young academic excellence with development potential, to create collaborative and innovative teams for critical tasks and future frontier, to guide future development of artificial intelligence and to build an innovative environment for the artificial intelligence industry. Over the past seven years, the intellectuals programme has been a three-staged process of free exploration, mission-led, innovative, and has become an important vehicle for attracting and developing the best talent in the field of artificial intelligence, with a total of over 180 intellectuals and a global artificial intelligence “dream team”.
More importantly, this “dream team” is becoming an important source of AI entrepreneurship. In recent years, the Institute of Intellectual Resources has accumulated over 20 innovative enterprises with core technical capabilities, including Zhipu AI, face-wall intelligence, Galaxy universal, star-sourced intelligence, a counter-matrix, and the Year of Zion Light. From LLM to smartness, from bottom vision to safe governance, the presence of intellectual scholars and intellectual ecology can be seen behind a representative group of AI entrepreneurship companies. The landing of the first phase of the Genesis is the most recent sign that this talent can transform into an industrial dynamic.
For your account security, please verify the mailbox.
Please verify your mailbox.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.