Core Insights - TwinRL introduces a digital twin-driven reinforcement learning framework that enhances the exploration capabilities of robots in real-world tasks, achieving a 100% success rate in various operations within approximately 20 minutes, while reducing human intervention by over 50% [1][22][36]. Group 1: Technology and Framework - TwinRL is not a simulator but an exploration amplifier and guide, designed to expand the exploration space for robots beyond the limitations of traditional methods [16][15]. - The framework consists of three main components: exploration space expansion, parallel online reinforcement learning in the digital twin, and sim-to-real guided exploration [32][36]. - The exploration space expansion strategy utilizes high-fidelity digital twin environments to generate synthetic trajectories that exceed human demonstration coverage [25][32]. Group 2: Performance and Efficiency - TwinRL demonstrates a significant improvement in exploration efficiency, achieving at least a 30% acceleration in convergence time compared to existing real-world reinforcement learning methods [22][39]. - In experiments, TwinRL maintained a near 100% success rate in both in-distribution and out-of-distribution areas, showcasing its robustness against environmental changes [39][46]. - The framework effectively bridges the gap between offline training and online learning, allowing for a smoother transition and reducing performance degradation during the learning process [39][34]. Group 3: Research Background and Observations - The research highlights that the effective exploration space in real-world VLA reinforcement learning is heavily constrained by the distribution of supervised fine-tuning (SFT) data [27][30]. - The study reveals that traditional reinforcement learning methods struggle with exploration deadlock in out-of-distribution scenarios, emphasizing the need for a broader exploration strategy [30][31]. - TwinRL addresses these challenges by moving the exploration process to a controllable and expandable digital twin environment, allowing for more effective learning [15][36].

Investment Rating - The report maintains a "Market Weight" rating for the mechanical equipment industry, indicating that the industry index is expected to perform within ±10% of the market index over the next six months [45]. Core Insights - The demand for humanoid robots is expected to accelerate in 2026 due to advancements in embodied intelligence, supported by policy backing and rapid technological iterations. The industry is transitioning from concept to mass production, presenting opportunities for investment in the robot sector [3][41]. - The engineering machinery sector shows signs of recovery, with significant growth in excavator sales in January 2026. Factors such as the accelerated issuance of special bonds and global interest rate cuts are expected to boost demand for engineering machinery [3][41]. - The mechanical equipment sector has outperformed the CSI 300 index, with a bi-weekly increase of 4.29%, ranking 4th among 31 sectors [4][12]. Summary by Sections Market Review - As of February 12, 2026, the mechanical equipment industry has seen a bi-weekly increase of 4.29%, outperforming the CSI 300 index by 5.01 percentage points [12]. - The engineering machinery sector recorded the highest bi-weekly increase of 7.80% among the five sub-sectors [15][17]. Valuation Situation - The current PE ratio for the mechanical equipment sector is 36.68, with sub-sectors showing varying valuations: general equipment at 53.27, specialized equipment at 37.85, and engineering machinery at 25.73 [21]. Data Updates - The report highlights significant growth in excavator sales, with domestic sales showing a year-on-year increase of 23.90% in January 2026 [40]. Industry News - The report includes updates on various companies, such as OptoTech planning to raise funds for expanding its industrial automation capabilities [39]. Weekly Insights - The report suggests focusing on companies like Huichuan Technology and Greentec Harmonic for potential investment opportunities in the robot sector, and Sany Heavy Industry and Hengli Hydraulic in the engineering machinery sector [41][43].

午后机器人（核心股）概念再度走强，双林股份涨超10%，此前百达精工、天奇股份涨停，精锻科技、 宏昌科技、五洲新春、锋龙股份跟涨。消息面上，宇树科技创始人、CEO 王兴兴在接受央视采访时表 示具身智能未来热度或远超移动互联网。如果未来几年，有真正大规模应用的具身智能 AI 模型和机器 人技术突破，那时候热度可能会比现在至少高 100 倍，甚至 1000 倍。 PCB概念午后震荡回升，铜冠铜箔涨超10%，续创历史新高，惠柏新材、江南新材、强达电路、天津普 林、华正新材跟涨。消息面上，日本半导体（核心股）材料厂Resonac宣布自3月1日起调涨铜箔基板 （CCL）、黏合胶片等印刷电路板（PCB）材料售价，涨幅达30%以上。 2月14日，A股蛇年收官，三大指数全天震荡下跌。截至收盘，沪指跌1.26%，报4082.07点，深成指跌 1.28%，报14100.19点，创业板指跌1.57%，报3275.96点，科创50指数跌0.72%，报1470.33点。沪深两市 合计成交额19826.8亿元，个股跌多涨少，沪深京三市约3900股飘绿。 板块题材上，军工（核心股）装备、影视（核心股）院线、造纸（核心股）、半导体设备、智能座 ...

Investment Rating - The industry investment rating is "Positive" (maintained) [4] Core Insights - Magic Atom has become the strategic partner for intelligent robots at the 2026 Spring Festival Gala, indicating a significant milestone in its market presence [1] - The collaboration with various companies such as Green Harmony, Lingyi Intelligent Manufacturing, and Huichuan Technology highlights the strategic partnerships aimed at enhancing the capabilities and production of humanoid robots [2][3] - The strategic financing of several hundred million yuan from ChipLink Integration's CVC platform indicates strong investor confidence and potential for growth in the humanoid robotics sector [3] Summary by Relevant Sections Company Progress - Magic Atom has set high-quality standards in collaboration with Green Harmony, ensuring precision inspection of key components [2] - A formal strategic partnership with Lingyi Intelligent Manufacturing focuses on embodied intelligence, robot manufacturing services, and overseas business expansion [2] - Huichuan Technology provides joint drive and motion control solutions that meet international standards for Magic Atom's complex movement requirements [2] - Other collaborations include partnerships with Mingzhi Electric and Hongchang Technology, indicating a broad network of support for Magic Atom's initiatives [2] Industry Outlook - The humanoid robotics industry is expected to grow, supported by strategic partnerships and technological advancements [4] - The report emphasizes the importance of continued development and collaboration within the industry to meet market demands and expectations [4]

联系人：李智 河南资本市场月报 分析师：李济生 登记编码：S0730522100002 lijs@ccnew.com 分析师：郑婷 登记编码：S0730524110001 zhengting@ccnew.com 相关报告 《河南资本市场月报（2026 年第 1 期）》 2026-1-15 《河南资本市场月报（2025 年第 11 期）》 2025-12-9 （2026 年第 2 期） 证券研究报告 发布日期：2026 年 2 月 13 日 投资要点： 风险提示：全球贸易和产业管制政策带来的不确定性风险；经济增长和 业绩不及预期；市场流动性风险。 本报告版权属于中原证券研究所股份有限公司 www.ccnew.com 请阅读最后一页各项声明 第 1页/共 20页 电话： 0371-65585629 地址： 郑州郑东新区商务外环路10号18楼 地址： 上海浦东新区世纪大道 1788 号 T1 座 22 楼 经济运行情况：2025 年，我国国内生产总值达到 1401879 亿元，比 上年增长 5.0%，经济增速目标圆满完成；2025 年河南省 GDP 达到 66632.79亿元，总规模位居全国第六、中部第一，GDP同 ...

Core Insights - Robo.ai Inc. has announced the establishment of a joint venture in Dubai focused on embodied intelligence data, marking a significant step in its strategic initiatives [1] - The company has secured its first order to provide 30,000 hours of embodied intelligence robot training data to DaBoss.AI Inc., with a delivery timeline of 12 months [1][3] Order Details and Financial Impact - The order represents a substantial initiation of Robo.ai's "embodied intelligence data port" strategy, confirming the company's business model viability [3] - The contract specifies the provision of 30,000 hours of effective data duration, which includes high-value multimodal raw data such as visual RGB-D, six-degree-of-freedom motion trajectories, and force-controlled tactile feedback, aimed at training L4 and above embodied intelligence models [3] - According to U.S. GAAP standards, the revenue from this business will be consolidated into Robo.ai's financial statements through the joint venture, contributing to the company's cash flow for the fiscal year 2026 [3] Global Strategic Collaboration - The project signifies the official launch of a global strategic collaboration between DaBoss.AI and Robo.ai, integrating GCC regional operations into the joint venture platform [4] - This partnership establishes Robo.ai as a core node in the global embodied intelligence data supply chain, aligning technological standards with regional resources [4] - Aiden Zhu, co-founder and CEO of DaBoss, highlighted the successful transition from a "heavy asset operation" to a "flexible cloud service" model, enabling immediate compliance with high-fidelity physical data for Silicon Valley clients [4] Company Profiles - DaBoss.AI Inc. is a leading AI model data service provider based in Silicon Valley, specializing in high-quality, real-world human interaction datasets and is a strategic partner to several top global AI laboratories [5] - Robo.ai Inc. (NASDAQ: AIIO) aims to build a leading global artificial intelligence machine economy platform, integrating AI software, intelligent hardware, and smart assets to create a unified AI operating system and blockchain-enabled ecosystem [6]

Core Viewpoint - The stock price of UBTECH (09880.HK) experienced a significant increase of 3.54% on February 13, 2026, closing at HKD 137.50, driven by new product launches and multiple favorable factors [1] Product Development Progress - On February 12, UBTECH launched the full-size research humanoid robot "Tiangong Walker DEX" in collaboration with the Beijing Humanoid Robot Innovation Center, showcasing significant improvements in stability, motion control, and embodied intelligence collaboration, while also providing a secondary development toolchain aimed at research and education [2] Financial and Technical Aspects - On the same day, the net inflow of main funds was HKD 40.83 million, with institutional buying accounting for approximately 12%, indicating active participation from institutional investors. The stock price broke through the 5-day moving average of approximately HKD 132.8, reaching a peak of HKD 139.00 during the day, with a volume ratio of 2.18, reflecting strong short-term buying activity [3] Industry Policy Status - Humanoid robots were featured in the 2026 Spring Festival Gala rehearsal, increasing public attention; the preheating of Tesla's Optimus V3 and positive statements from Elon Musk have boosted global interest in the humanoid robot sector. By the end of 2025, the Ministry of Industry and Information Technology established a standardization technical committee for humanoid robots and embodied intelligence, with UBTECH leading the formulation of several national standards, enhancing its industry influence [4] Business Progress - In 2025, the total order amount for humanoid robots exceeded RMB 1.4 billion, with the industrial version Walker S2 entering the supply chains of international giants like Airbus, and plans to increase production capacity to 10,000 units in 2026. The mid-year report for 2025 indicated an overall loss of RMB 414 million, but the first batch of Walker S2 was successfully delivered, marking a significant step in commercialization from "0 to 1" [5] Future Development - Attention is needed as the company's price-to-earnings ratio (TTM) is negative at -61.83, indicating that profitability has not yet materialized. Challenges remain in terms of cost (humanoid robots are still 5-8 times more expensive than AGVs), reliability (production lines require high uptime), and delivery (complexity of the supply chain) for scaling up operations [6]

Core Viewpoint - UBTECH Robotics (09880) has seen a stock increase of over 4%, currently trading at 137.7 HKD with a transaction volume of 857 million HKD, following the launch of its new humanoid robot "Tian Gong Hang Zhe DEX" in collaboration with the Beijing Humanoid Robot Innovation Center [1] Group 1: Product Launch - The "Tian Gong Hang Zhe DEX" humanoid robot features significant improvements in stability, motion control, and autonomous operation, aimed at providing tailored solutions for research and education sectors [1] - The robot is designed to cater to four core groups in research and education, enhancing their capabilities in various scenarios [1] Group 2: Technological Advancements - On February 1, UBTECH officially open-sourced its self-developed embodied intelligent model "Thinker," which emphasizes "small parameters, high performance, and full open-source" [1] - The "Thinker" model achieved first place in nine global benchmark evaluations related to spatial understanding, task planning, visual reasoning, and temporal decision-making, marking it as the first lightweight open-source intelligent base in the industrial humanoid robot field [1]

Core Viewpoint - UBTECH Robotics (09880) has seen a stock increase of over 4%, currently trading at 137.7 HKD with a transaction volume of 857 million HKD, following the launch of its new humanoid robot "Tian Gong Hang Zhe DEX" [1] Group 1: Product Launch - UBTECH has announced the release of "Tian Gong Hang Zhe DEX," a full-size, versatile research humanoid robot developed in collaboration with the Beijing Humanoid Robot Innovation Center [1] - The new robot features significant improvements in stability, motion control, and autonomous operation, providing tailored solutions for research and education across various scenarios [1] Group 2: Technological Advancements - On February 1, UBTECH officially open-sourced its self-developed embodied intelligent model, Thinker, which is characterized by "small parameters, high performance, and full open-source" [1] - Thinker achieved first place in nine global benchmark evaluations related to spatial understanding, task planning, visual reasoning, and temporal decision-making, marking it as the first lightweight, full-stack open-source intelligent base in the industrial humanoid robot sector [1]

《控制理论与应用》 "具身智能与人形机器人" 专刊征文通知 具身智能的核心在于智能体通过与物理环境的实时交互来学习、进化并执行任务. 人形 机器人凭借其类人的形态, 能够适应人类设计的非结构化环境, 但其高维度、强非线性及欠 驱动特性, 使得"感知决策控制"的闭环实现极具挑战. 当前, 学术界正致力于构建"大脑(认知 规划)+小脑(运动控制)"的协同架构, 试图解决SimtoReal(虚实迁移)鸿沟、灵巧操作的泛化性 以及人机共融的安全性等关键问题. 为了深入探讨具身智能大模型如何赋能人形机器人的运动控制与交互能力, 推动控制科 学、人工智能与机器人学的交叉融合, 分享最新的研究成果与实践经验, 本刊特此发起"具身 智能与人形机器人"专刊征稿活动, 诚邀广大专家学者贡献智慧与洞见. 征稿方向包括但不限于 1) 人形机器人动力学建模与控制; 2) 具身多模态感知与环境理解; 3) 具身灵巧操作与运动技能学习; 4) SimtoReal虚实迁移与泛化技术; 5) "大脑小脑"协同的分层控制架构; 6) 基于视觉语言动作(VLA)模型的端到端控制; 7) 基于世界模型(world models)的预测与规划; 随着 ...