Workflow
机器人技术
icon
Search documents
边缘设备上高效运行!NanoVLA :保留 VLA 模型的精度与泛化能力,推理速度提升 52 倍
具身智能之心· 2025-11-01 16:03
点击下方 卡片 ,关注" 具身智能 之心 "公众号 作者丨 Jiahong Chen等 编辑丨具身智能之心 本文只做学术分享,如有侵权,联系删文 >> 点击进入→ 具身智能之心 技术交流群 更多干货,欢迎加入国内首个具身智能全栈学习社区 : 具身智能之心知识星球 (戳我) , 这里包含所有你想要的。 在机器人操控领域," 通用化 " 与 " 轻量化 " 的矛盾长期制约着技术落地——现有视觉-语言-动作(VLA)模型虽能实现复杂任务推理,但因参数量庞大、计算需求 高,难以部署在移动机器人、嵌入式系统(如 Jetson Orin Nano)等资源受限设备上。 而由英属哥伦比亚大学、阿尔伯塔大学与小米汽车团队联合提出的 NanoVLA ,用 " 视觉-语言解耦融合+长短动作分块+动态路由 " 的创新架构,彻底打破这一困 境:既保留通用 VLA 模型的任务精度与泛化能力,又将推理速度提升 52 倍、参数量压缩 98%,首次实现 "在边缘设备上高效运行通用机器人策略" 的目标。 为什么要重构 VLA 模型的边缘部署逻辑? 当前主流 VLA 模型陷入 "性能与效率不可兼得" 的困境:为实现跨任务泛化,模型通常依赖数十亿参 ...
“进一步退两步”,特朗普制造业回流目标正在被自身关税和移民政策绊倒
Guan Cha Zhe Wang· 2025-10-19 07:35
【文/观察者网 柳白】"我们真不希望这是个进一步却退两步的方案。"对于美国总统特朗普让制造业回 流的雄心,美国电气制造商协会高管斯宾塞·彼得森颇为无奈。 毕竟无论是关税还是移民等一系列政策,特朗普政府的做法似乎都在背道而驰。 美国"政客新闻网"在10月18日发表的评论文章中称,特朗普政府的制造业回流政策陷入自我矛盾的困 境,关税推高了美国制造商扩张所需的原材料和进口设备成本,新的签证和移民政策可能缩减人才供 给,加剧制造业技能工人短缺问题,而白宫推动的开支削减则威胁到企业回流所需的补贴,导致补贴环 境不稳定,甚至特朗普政府内部都存在左右互搏的政策,导致资金项目被暂停审查。以至于不少企业抱 怨,他们压根不知道应该围绕哪种"美国优先"政策进行规划。 "只想着征税,忽略了供给侧障碍" 文章直言,这些政策共同凸显了特朗普即兴式执政方式的局限性,也暴露了其关税制度更多是为了报复 而非抢占先机,结果导致了一场代价高昂的平衡博弈,商界领袖表示这一局面越来越难以应对。 例如,关税提高了进口设备的价格,其中包括用于社区大学实验室的训练机器人,使得培训美国本地工 人变得更贵,而这些工人正是政府希望填补新岗位的人才来源。 代表30 ...
东土科技:公司目前并不直接生产人形机器人本体
Zheng Quan Ri Bao Wang· 2025-09-24 09:10
Core Viewpoint - Dongtu Technology (300353) does not directly produce humanoid robots but focuses on providing foundational technology support in the field of embodied robots [1] Group 1: Company Overview - The company's Hongdao operating system is involved in national major projects and possesses fully autonomous and controllable attributes [1] - Dongtu Technology's AUTBUS bus network is the only domestic bus technology that has received international standards certification [1] Group 2: Technological Capabilities - The combination of the Hongdao operating system and AUTBUS bus network can create a robotic electronic architecture with time-deterministic networks and deterministic computing capabilities [1] - This architecture supports the high safety, high performance, and high-quality development of embodied robots, providing autonomous and controllable foundational technology assurance [1]
劳动者自己当代言人,北京十大劳务品牌盯准“北京特色”
转自:北京日报客户端 山西的"吕梁山护工"、甘肃的"岷归药工"、湖北的"孝感米酒师"……这些年各地涌现出一大批具有地域 特色的劳务品牌。这些劳务品牌为稳就业保就业、推动乡村振兴提供了有力支撑。 众所周知,北京是个劳务输入城市,那北京有自己的劳务品牌吗?答案让很多人出乎意料——不光有, 还不少!现在,第三届全国劳务品牌形象代言人征集展示活动正在进行中,北京推出的10个劳务品牌精 彩亮相——怀柔宝山大姐、密云文物医生、 京顺小蜜蜂、房山智源工匠、京虹无忧堵漏师、京城消控 卫士、北京智虫匠才、北京温暖侬嫂、北京惠心相伴医疗护理员、通州绢人唐娃娃制作师。 记者注意到,北京的劳务品牌同样极具地方特色,针对劳务输入城市的特点,为外来务工群体提供全方 位的职业发展支持,通过系统化的就业指导和专业技能培训,帮助务工人员提升职业素养,掌握实用技 能,从而实现稳定就业。 另外,在这次征集展示活动中,北京各劳务品牌不找明星,不找网红,10位深耕各自领域的劳动者亲自 代言,用他们的亲身经历和专业实力,向全国展示北京劳务品牌的独特魅力。 劳务品牌:北京惠心相伴医疗护理员 代言人:吴波 4000余名惠心相伴医疗护理员分布在北京大大小小7 ...
印媒:印度应与中国携手向前
Huan Qiu Wang Zi Xun· 2025-09-15 23:10
Group 1 - China's electricity generation capacity is 2.5 times that of the United States, with plans to add an amount equivalent to Germany's total generation capacity each year [1] - China is leading in clean energy production, with a significant position in the global battery supply chain due to low commercial electricity prices and strong manufacturing capabilities [1] - In the electric vehicle sector, China dominates, accounting for nearly two-thirds of global electric vehicle sales in 2024, with six out of the ten best-selling electric vehicle brands being Chinese [1] Group 2 - India’s growth strategy relies on large-scale energy production, a vast domestic market, and opportunities for acquiring cutting-edge technology, all of which China currently leads [2] - India should focus on developing its solar energy and storage industries, closely linked to China's supply chain, and collaborate with Chinese capital for local production [2] - The development of artificial intelligence in India will thrive where computing costs are low, data is abundant, and regulatory support is present, aligning with China's strategy of using clean energy to power open-source AI [2] Group 3 - India's long-standing strategy has been characterized by "multilateral alliances," but it should now make clearer choices to collaborate with China in areas that advance its own goals [3]
杭州六小龙融资近20亿,群核科技两度IPO失效,宇树强脑竞速上市
Jin Rong Jie· 2025-08-14 23:15
Group 1: Core Insights - The "Hangzhou Six Little Dragons" are emerging as a focal point in the capital market, showcasing strong capabilities in cutting-edge technologies such as artificial intelligence, robotics, and brain-computer interfaces [1] - Qunhe Technology, the first among the "Hangzhou Six Little Dragons" to pursue an IPO, submitted its prospectus to the Hong Kong Stock Exchange on February 14, but it automatically lapsed on August 14 due to not completing the listing process within the six-month validity period [3] - Despite impressive gross margins of 72.7%, 76.8%, and 80.4% for 2022, 2023, and the first three quarters of 2024 respectively, Qunhe Technology remains in a loss-making position with adjusted net losses of 338 million, 242 million, and 93.61 million yuan [3] Group 2: Investment Trends - Investment institutions are increasingly focused on the "Hangzhou Six Little Dragons," with Qiangna Technology reportedly negotiating an IPO pre-financing of approximately 100 million USD at a valuation exceeding 1.3 billion USD [5] - Yushutech has initiated its A-share IPO process, signing a counseling agreement with CITIC Securities on July 7, with plans for a comprehensive evaluation of listing conditions by October [4] - The influx of venture capital not only provides essential funding for these companies but also brings valuable industry resources and management experience, reflecting the growth potential of the Hangzhou tech industry [5] Group 3: Government Support - Government-guided funds play a crucial role in nurturing tech enterprises, with many newly established funds having a duration of over 10 years, some even extending to 20 years, which supports a more patient capital approach [6] - The Shenzhen Futian guiding fund has set a benchmark by extending the duration of its managed sub-funds by 2 years, demonstrating a commitment to long-term investment strategies [6]
2021版“散户暴打空头”重演?这是美股最被做空的小微盘名单
Hua Er Jie Jian Wen· 2025-07-23 01:06
Group 1 - The current market is experiencing a resurgence of retail investor enthusiasm reminiscent of 2021, with a focus on "Meme stocks" and a significant increase in options trading [1][5] - Retail investors are targeting small-cap stocks with high short interest, aiming to replicate the previous success of forcing short sellers to cover their positions [4][8] - The proportion of call options in the market has surged to 70%, the highest level since the "Meme stock" phenomenon began in 2021, indicating a rise in speculative sentiment [5][8] Group 2 - Small-cap stocks with market capitalizations between $10 million and $1.5 billion and short interest exceeding 25% are becoming the new targets for retail investors [4][6] - Notable high short interest stocks include BEELINE HOLDINGS with 166.77% short interest and a market cap of $18 million, and NEOVOLTA with 81.22% short interest and a market cap of $160.4 million [4][7] - Analysts highlight that many of these companies have poor fundamentals, but high short interest provides ample fuel for potential short squeezes, creating a self-reinforcing cycle of price increases [6][9]
分析了102个VLA模型、26个数据集和12个仿真平台
自动驾驶之心· 2025-07-22 02:18
Core Viewpoint - The article discusses the transformative breakthrough of Visual-Language-Action (VLA) models in robotics, emphasizing their integration of visual perception, natural language understanding, and embodied control within a unified learning framework. It highlights the development and evaluation of 102 VLA models, 26 foundational datasets, and 12 simulation platforms, identifying current challenges and future directions for enhancing robotic autonomy and adaptability [3][4][6]. Group 1: VLA Models and Framework - VLA models represent a new frontier in robotic intelligence, enabling robots to perceive visual environments, understand natural language commands, and execute meaningful actions, bridging the semantic gap between various modalities [7][9]. - The architecture of VLA models integrates visual, language, and proprioceptive encoders into a diffusion backbone network to generate control commands, facilitating end-to-end processing of multimodal inputs [11][12]. - The development of effective VLA models relies on large-scale, diverse multimodal datasets and realistic simulation platforms, which are crucial for training models to robustly understand language instructions and perceive visual environments [5][30]. Group 2: Datasets and Evaluation - The article outlines the evolution of VLA datasets, noting that early datasets focused on discrete decision-making in constrained environments, while recent datasets incorporate richer sensory streams and longer task durations, addressing the need for complex multimodal control challenges [21][22][29]. - A comprehensive benchmarking strategy is proposed to evaluate datasets based on task complexity and modality richness, highlighting the need for new datasets that integrate high task difficulty with extensive multimodal inputs [24][28]. - The analysis reveals a gap in current VLA benchmarks, particularly in combining long-duration, multi-skill control with diverse multimodal integration, indicating a promising direction for future dataset development [29][43]. Group 3: Simulation Tools - Simulation environments are critical for VLA research, enabling the generation of large-scale, repeatable, and richly annotated data that surpasses physical world limitations [30][31]. - Various advanced simulation platforms, such as AI2-THOR and NVIDIA Isaac Sim, provide high-fidelity physical effects and customizable multimodal sensors, essential for developing robust VLA models [32][33]. - The integration of simulation tools with VLA datasets accelerates the collaborative development of control algorithms and benchmark datasets, ensuring advancements in multimodal perception are effectively evaluated before deployment in real robotic platforms [30][33]. Group 4: Applications and Challenges - VLA models are categorized into six broad application areas, including manipulation and task generalization, autonomous mobility, human assistance, and interaction, showcasing their versatility across various robotic tasks [34][35]. - The article identifies key challenges in VLA model architecture, such as tokenization and vocabulary alignment, modality fusion, and cross-entity generalization, which need to be addressed to enhance model performance and adaptability [39][40][41]. - Data challenges are also highlighted, including task diversity, modality imbalance, annotation quality, and the trade-off between realism and scale in datasets, which hinder the development of robust general-purpose VLA models [42][43].
利亚德(300296) - 2025年7月4日投资者关系活动记录表
2025-07-07 01:08
Group 1: AI and Spatial Computing Business - The company focuses on AI and spatial computing, leveraging global leading optical motion capture technology for product development and sales, applicable in various fields such as film, robotics, healthcare, sports, and industrial simulation [1] - In 2024, the company entered the embodied intelligence sector, emphasizing "data, services, algorithms, and hardware" as core capabilities, and established a joint laboratory with industry partners [2] Group 2: Data and Service Capabilities - The company has a rich accumulation of motion data and can continuously generate new data based on motion capture technology, creating high-quality, high-precision robot motion datasets [2] - Services have expanded from hardware provision to include data collection, robot training services, and simulation training, tailored to customer needs [2] Group 3: Investor Questions and Responses - The company offers comprehensive solutions, including hardware, software, and data services, based on varying customer requirements [3] - A high-quality motion database is available, allowing robot manufacturers to extract data directly, with ongoing data collection based on specific needs [3] - The amount of data required for training robots varies; for example, training a robot to run may require around 100 effective data points, while more complex actions necessitate more data [4] Group 4: Data Processing and Technology - Data collected from motion capture must be cleaned to eliminate noise and distortion before being optimized for robot use [5] - The company employs various technical routes, including optical, inertial, and hybrid systems, with optical technology currently deemed the most precise for robot training [5]
镁伽科技冲刺港交所:国内收入最大的自主智能体供应商,在手订单15亿元
IPO早知道· 2025-06-26 00:39
Core Viewpoint - Magnesium Technology Co., Ltd. is positioned as a leading provider of autonomous intelligent agents in China, focusing on enhancing productivity and innovation in smart laboratories and intelligent manufacturing through proprietary robotic and AI technologies [2][3]. Group 1: Company Overview - Established in 2016, Magnesium Technology has expanded its business from life sciences to various sectors including consumer integrated circuits, food service, new energy, and agriculture [2]. - The company aims to facilitate intelligent transformation for enterprises by allowing human resources to focus on high-value tasks [2]. Group 2: Market Position and Growth - As of June 21, 2025, Magnesium Technology ranks first among domestic autonomous intelligent agent suppliers in China, with the broadest application coverage in robotic solutions [3]. - The global market for autonomous intelligent robotics technology is projected to grow from approximately 31.8 billion in 2020 to about 114.3 billion by 2024, with a compound annual growth rate (CAGR) of 37.7% [5]. Group 3: Technological Innovations - Magnesium Technology has developed several industry-first technologies, including the Labillion and LibraX operating systems, and the Megalaxy Lab for intelligent life sciences [4]. - The company has introduced the Auflo AI-driven liquid handling workstation and the CellVue high-content imaging analysis system, achieving high pixel resolution [4]. Group 4: Financial Performance - Revenue for Magnesium Technology is projected to grow from 455 million in 2022 to 930 million in 2024, reflecting a CAGR of 43.0% [5]. - The gross profit margins for the same period are reported at 28.1%, 23.9%, and 29.0% respectively [6]. Group 5: Client Base and Retention - As of December 31, 2024, Magnesium Technology has served over 880 clients, including major companies like Agilent Technologies and WuXi AppTec, with a customer retention rate of 74% and a revenue retention rate of 115% for new clients acquired in 2022 [5]. Group 6: Future Plans - The funds raised from the IPO will primarily be allocated for technology and product development, capacity expansion, sales network growth, strategic partnerships, and general corporate purposes [6].