点击下方 卡片 ，关注" 具身智能 之心 "公众号 RLinf通过标准化的接口，支持主流VLA模型及基于CPU与GPU的模拟器，并率先实现了对π0和π0.5模型系列的强化学习微调，欢迎大家star&follow。 | | | Evaluation results on the four LIBERO task groups | | | | | | --- | --- | --- | --- | --- | --- | --- | | Model | | | | LIBERO | | | | | Spatial | Object | Goal | Long | Avg. | A Avg. | | Full Dataset SFT | | | | | | | | Octo | 78.9% | 85.7% | 84.6% | 51.1% | 75.1% | - | | OpenVLA | 84.7% | 88.4% | 79.2% | 53.7% | 76.5% | - | | Tlfast | 96.4% | 96.8% | 88.6% | 60.2% | 85.5% | - | | OpenVLA-OFT | ...

Domestic News - Meituan officially released and open-sourced LongCat-Flash-Chat, featuring an innovative Mixture-of-Experts architecture with a total of 560 billion parameters and an activation parameter range of 18.6 billion to 31.3 billion [2] - Tsinghua University and other institutions open-sourced RLinf, a large-scale reinforcement learning framework for embodied intelligence, achieving over 120% system speedup compared to other frameworks [3] - A batch of national standards related to AI-generated content identification and safety measures for electric bicycles will be implemented starting September 1, aimed at promoting healthy development in emerging industries [4] - Beijing Data Group is expected to be officially listed soon, with a registered capital of 3 billion yuan, focusing on big data services and AI public service platform technology consulting [5][6] - Sanwei Xinan is actively laying out Web3.0 applications, focusing on stablecoins and RWA, and has established itself as a vice-chairman unit of the Hong Kong Web3.0 Standardization Association [7] - Alibaba launched the AgentScope 1.0 framework for multi-agent development, providing a comprehensive solution for the entire lifecycle of intelligent applications [8] - Tencent announced the open-sourcing of the Youtu-Agent framework, which does not require additional model training and is based entirely on the open-source ecosystem [9] - Tencent released the HunyuanWorld-Voyager model, the first to support native 3D reconstruction for virtual reality and gaming applications [10] - Digital China is expanding into the robotics industry and has formed partnerships with leading companies in the field [11] - ByteDance plans to issue stock options to its Seed department, focusing on large model technology personnel [12] - Douyin established a new company focused on AI applications in healthcare, with a registered capital of 100,000 yuan [13] Financing and Mergers - Obita completed over 10 million USD in angel round financing, with funds aimed at core system development and market expansion [25] - New Unisplendour Group established a high-tech company with a registered capital of 10 million yuan, focusing on integrated circuit design and sales [26] - Ant Group's subsidiary invested in Xinyuan Semiconductor, increasing its registered capital from approximately 46.5 million yuan to about 50.3 million yuan [27] - AI company Anthropic raised 13 billion USD in a new funding round, increasing its valuation to 183 billion USD [28] - OpenAI agreed to acquire product testing startup Statsig for 1.1 billion USD in stock, marking one of its largest acquisitions [29] Policies and Trends - The National Development and Reform Commission plans to issue "AI vouchers" to promote the use of intelligent terminals and reduce R&D costs [33] - The Ministry of Industry and Information Technology announced plans to support the development of high-performance AI training and inference chips [44] - The Ministry of Industry and Information Technology reported that software business revenue reached 83.246 billion yuan in the first seven months, a year-on-year increase of 12.3% [41] - The Ministry of Industry and Information Technology indicated that internet enterprises achieved a total profit of 93.88 billion yuan in the first seven months, a year-on-year decrease of 1.8% [42] - Shanghai is organizing the 2025 "AI+" action project application, focusing on enhancing AI capabilities and promoting industry development [45] - The National Standards Committee plans to revise over 4,000 national standards in fields such as AI and IoT [46]

（文章来源：北京商报） 据了解，RLinf名称中"inf"既代表"infrastructure"（基础设施），也寓意"infinite"（无限扩展），核心解 决当前框架对具身智能支持受限的问题。相比纯推理大模型，具身智能需兼顾"大脑"（推理规划） 与"小脑"（执行操作），且存在"渲训推一体化"特性，对算力、显存及框架灵活性要求更高。RLinf通 过六大层级（用户层、任务层、执行层、调度层、通信层、硬件层）设计，针对性突破技术难点。 北京商报讯2025年9月1日，无问芯穹官方公众号发布消息称，公司联合清华大学、北京中关村学院，并 携手北京大学、加州大学伯克利分校等机构，正式开源首个面向具身智能的"渲训推一体化"大规模强化 学习框架RLinf，为人工智能从"感知"向"行动"跨越提供关键技术支撑。 ...

Core Insights - The company announced the launch of the RLinf framework, a large-scale reinforcement learning framework aimed at embodied intelligence, in collaboration with several prestigious institutions [1] Group 1: Framework Overview - RLinf stands for "reinforcement learning infrastructure" and symbolizes "infinite" scalability, addressing limitations in current frameworks for supporting embodied intelligence [1] - The framework integrates "perception" and "action," requiring a balance between reasoning (brain) and execution (cerebellum), which imposes higher demands on computing power, memory, and framework flexibility compared to pure inference models [1] Group 2: Technical Design - RLinf is designed with six layers: user layer, task layer, execution layer, scheduling layer, communication layer, and hardware layer, specifically targeting technical challenges in the field [1]

Core Viewpoint - The article discusses the launch of RLinf, a large-scale reinforcement learning framework aimed at embodied intelligence, highlighting its innovative design and capabilities in enhancing AI's transition from perception to action [2][5]. Group 1: Framework Overview - RLinf is a flexible and scalable framework designed for embodied intelligence, integrating various components to optimize performance [5]. - The framework's name "inf" signifies both "infrastructure" and "infinite" scaling, emphasizing its adaptable system design [7]. - RLinf features a hybrid execution model that achieves over 120% system speedup compared to traditional frameworks, with VLA model performance improvements of 40%-60% [7][12]. Group 2: Execution Modes - RLinf supports three execution modes: Collocated, Disaggregated, and Hybrid, allowing users to configure components based on their needs [17][15]. - The hybrid mode combines the advantages of both shared and separated execution, minimizing system idle time and enhancing efficiency [12][15]. Group 3: Communication and Scheduling - The framework includes an adaptive communication library designed for reinforcement learning, optimizing data exchange between components [19][22]. - RLinf features an automated scheduling module that minimizes resource idleness and dynamically adjusts to user training flows, achieving rapid scaling capabilities [23][24]. Group 4: Performance Metrics - RLinf has demonstrated significant performance improvements in embodied intelligence tasks, achieving success rates of 80%-90% in specific scenarios, compared to 30%-50% in previous models [24][26]. - The framework has also achieved state-of-the-art (SOTA) performance in mathematical reasoning tasks across multiple datasets, showcasing its versatility [29][30]. Group 5: Documentation and Community Engagement - Comprehensive documentation and API support are provided to enhance user experience and facilitate understanding of the framework [32][34]. - The RLinf team encourages collaboration and invites users to explore the framework, highlighting ongoing recruitment for various research and engineering positions [33][34].

Core Viewpoint - The article discusses the launch of RLinf, a large-scale reinforcement learning framework designed for embodied intelligence, emphasizing its flexible and scalable architecture that integrates training, rendering, and inference processes [5][7]. Group 1: Development of RL Framework - The transition in artificial intelligence from "perception" to "action" highlights the importance of embodied intelligence, which is gaining attention in both academia and industry [2][4]. - RLinf is developed collaboratively by Tsinghua University, Beijing Zhongguancun College, and Wuwenchin, aiming to address the limitations of existing frameworks in supporting embodied intelligence [5][7]. Group 2: Features of RLinf - RLinf's architecture consists of six layers: user layer, task layer, execution layer, scheduling layer, communication layer, and hardware layer, allowing for a hybrid execution mode that achieves over 120% system speedup [7][12]. - The framework introduces a Macro-to-Micro Flow (M2Flow) mechanism, enabling flexible construction of training processes while maintaining high programming flexibility and debugging ease [14][15]. Group 3: Execution Modes - RLinf supports three execution modes: Collocated Mode, Disaggregated Mode, and Hybrid Mode, allowing users to configure components for optimal resource utilization [19][20]. - The framework integrates low-intrusion multi-backend solutions to cater to the diverse needs of researchers in the embodied intelligence field [16][20]. Group 4: Communication and Scheduling - RLinf features an adaptive communication library designed for reinforcement learning, optimizing data exchange between components to enhance system efficiency [22][28]. - An automated scheduling module minimizes resource idling by analyzing component performance and selecting the best execution mode, significantly improving training stability [24][25]. Group 5: Performance Metrics - RLinf demonstrates superior performance in embodied intelligence tasks, achieving over 120% efficiency improvement compared to existing frameworks in specific tests [27][33]. - The framework has shown significant success rate improvements in various tasks, with models achieving up to 97.3% success rates in specific scenarios [31][35]. Group 6: Future Development and Community Engagement - The RLinf team emphasizes open-source principles, providing comprehensive documentation and support to enhance user experience and facilitate collaboration [40][41]. - The team is actively recruiting for various positions to further develop and maintain the RLinf framework, inviting community engagement and feedback [42][43].