1. 人形機器人元年或將到來，斯坦福ALOHA機械臂硬件成功布局
Mobile ALOHA機器人走紅到來，AI賦能加速人形通用機器人量產落地。“模仿學習+ACT強化學習算法”讓真人在機器人面前演示動作或操作機器人執行動作，機器人觀察後，直接遷移到自己本體上完成任務，形成人形機器人的行爲邏輯，可低成本訓練靈巧機器手，助力ALOHA機械臂快速理解並精確模仿人類動作；多場景及新舊任務協同訓練將助力機器人在靜態ALOHA數據集基礎上，自主尋找不同任務的對應最優解，提高操作的精準性與靈敏性，爲人形機器人運用在更多場景上創造可能；“關鍵技術自研+硬件外購”的開發路徑，掌握核心技術，採購市場上成熟產品組裝機器人，降低製造成本，便於後期維護、模塊化更新。
Mobile Aloha: AI robotic arms project brings humanoid robots a step closer
When Stanford University and Google DeepMind recently unveiled their open-source Mobile Aloha robotic arms system to a warm public reception, the artificial intelligence (AI) empowerment it represents moves the momentum toward a future of mass-produced general-purpose humanoid robots. The robot’s learning algorithm combines imitation learning with ACT reinforcement learning, where a human makes a set of actions to be mimicked by the robot. (ACT refers to “action chunking with transformers” and is about sequential chunks of actions rather than single actions.) The robot observes an action chunk and absorbs the information directly to form its behavioral logic and imitates the action chunk. The system trains up robotic arm dexterity at low cost and allows the robot to quickly understand and accurately mirror human movements. Based on static Mobile Aloha data sets, collaborative training across multiple application scenarios involving both new and old tasks helps the robot independently find optimal solutions to a variety of tasks. Operational accuracy and sensitivity would improve over time and expand potential scenarios for humanoid robots. The developmental path of independent research and hardware outsourcing entails mastering key technologies and purchasing mature products on the market to assemble the robot, thereby reducing manufacturing cost, while facilitating subsequent maintenance and module upgrades.
2. AI底層平臺搭建完畢，英偉達Isaac AMR系統助力軟件研發
Issac AMR平臺實現機器人全地形移動功能，幫助研發者專注軟件開發。基於“NVDIA工具包”，Isaac AMR平臺通過“虛擬環境仿真”訓練方法，將“人類示範最優動作”改爲虛擬環境中“數據自行擬合最優動作”，基於大量行動和環境數據，加快訓練速度，大幅提升機器人在實際場景中應用部署的效率，增強其自主運動控制能力與移動規劃能力，形成人形機器人的“小腦”與“大腦”；Nova Orin系統控制平臺高度集成，提供傳感器+算力一體化的通用範式，降低機器人硬件研發門檻，實現差異化發展，融入模塊化設計，加速人形機器人大規模生產實現。
Isaac: with underlying AI platform, Nvidia’s AMR system aids software R&D
Nvidia’s Isaac autonomous mobile robot (AMR) platform coordinates movement for a fleet of robots across all terrains and aids developers in the software development. Based on the Nvidia Toolkit, the Isaac platform uses virtual environment simulation training methodology, where it replaces human-modeled optimal movements with data-driven self-fitting optimal movements in a virtual environment. The training pace is accelerated due to the large amounts of action and environmental data, which sharply improves the efficiency of robots’ application and deployability in real-life scenarios. Its enhanced autonomous motion control and mobile planning capability respectively form the “cerebellum” and “brain” for the robots. The Nova Orin system is a highly integrated control platform for the universal paradigm of sensors and computing power. It lowers the threshold for robotic hardware R&D, thereby driving differentiated development and collating modular design, thus accelerating mass production of humanoid robots.
3. 人工智能重塑行業競爭格局，核心製造企業追趕爭先
AI成爲新的增長點，核心互聯網及工業等企業積極擁抱人工智能浪潮。工業5.0革命下，如何更好應用AI成爲時代競爭主題，微軟、IBM、日立、松下、軟銀等都在積極探索人工智能高適應性的公司轉型之路，不斷嘗試將人工智能與其核心業務相結合，AI改造傳統產品；並利用自身雄厚資金和技術儲備，做好前瞻技術佈局，橫向擴展AI使用場景，亮點案例如下：
1）Amazon：利用AI建立超75萬個倉庫機器人組成物流智能分揀系統，紅杉系統（Sequoia）與Digit 人形機器人開創未來無人化高效物流格局。
2）西門子：藉助AI算法，開發ACUSON Sequoia 超聲系統，其具備實時可視化成像功能，提升醫生診斷效率，爲患者提供定製化的醫療方案。
AI redrawing competitive landscape: key manufacturers catching up with leaders
Key internet and industrial companies are embracing AI to harness its ability as a transformative growth catalyst. The competitive thrust in Industry 5.0 era, where humans coexist with robots, is how to use AI better. Microsoft, IBM, Hitachi, Panasonic and SoftBank, for instance, are undergoing corporate transformation with high AI adaptability. They continually try to integrate AI into their core businesses to revamp their traditional products, using their own strong capital and technology reserves to construct a forward-looking technological framework and expand AI application scenarios horizontally. We highlight several cases in point:
Amazon: it is building a smart logistics sorting system with more than 750,000 warehouse robots, where its Sequoia robotics and Agility Robotics’ Digit humanoid robots flesh out an unmanned future of logistical efficiency.
Siemens: its Acuson Sequoia ultrasound system uses real-time visual imaging to raise doctors’ diagnostic efficiency and provide patients with customized medical solutions.
4. 初創團隊引領下游硬件多點開花，重量級硬件應用涌現可期
“人工智能技術+互聯網平臺”給傳統行業帶去創新動力，AI 下流應用層出不窮，未來有可能出現顛覆性的應用。我們緊密跟蹤新使用場景的打開，梳理出有趣且有潛力的募資產品項目，在多功能機器人、智能消費電子、智能家居、工業場景應用等領域有可能出現重量級產品。
1）MOFLIN ：具有情感功能的人工智能寵物機器人，通過人機互動學習不斷提升情感分析能力，形成獨特個性，向人類迴應和表達自己的情感。
2）Heisenberg H1：帶人工智能視覺功能的多合一機器人割草機，利用機器人視覺、雙人工智能芯片和深度學習算法，爲整個草坪創建 3D 導航，實現完美的邊緣和均勻的修剪。
Downstream hardware startup projects: we sniff out some emerging applications
Traditional industries are using AI technology combined with an internet platform to increase their innovative ability. AI applications are showing up in emerging downstream markets one after another, heightening the prospects for disruptive applications. These downstream industries include multi-functional robots, smart consumer electronics and smart homes, as well as industrial application scenarios. We have been tracking new application scenarios closely and picked out a few of the interesting ones with fundraising product project potential:
Moflin is an AI pet robot imbued with emotional functionalities and with emotional analysis capability that keeps improving via interactive human-to-computer learning. Each Moflin evolves into a unique personality and is able to respond to human emotions as well as express its own emotions.
Heisenberg H1 is an all-in-one lawn mower robot with robotic vision. Its dual AI chips and deep learning algorithms map out 3D navigation for an entire lawn to enable perfect edges and evenly mowed surfaces.
