物理AI：加速时刻

Contents Building the future: Physical AI comes of age Why now? The convergence that changes everythingA global race: With governance catching upPAI in the real world Where PAI scales value: Industrial robotics The proving groundThe PAI technology stackGetting past bottlenecks Realising the value: The dual maturity lens Technology application maturity: Four stagesOperational maturity and transformation readinessGovernance: The emerging dimensionNavigating the journey: Three failure modes to avoid The three questions every operationsleader should be asking now04 Figure 1: Physical AI explained Physical AI refers to AI systems embedded in or controlling physical hardware – robots, autonomous vehicles, drones, and smartmanufacturing systems – that interact with and act upon the real world. Unlike purely digital AI, PAI perceives its environment throughsensors, makes real-time decisions, and executes actions with tangible consequences. It bridges three technology domains to mergedigital intelligence and physical reality. Building the future:Physical AI comes of age Today, just 5 per cent of firms say Physical AI (PAI) is transformingtheir organisation. Within three years, 41 per cent expect it will.That gap, between current impact and future expectations, definesthe story of this paper. And it has real urgency: only 3 per cent offirms have PAI extensively integrated into operations today, yetthis is forecast to reach 18 per cent within two years.1Those whomove now will not just gain an operational edge – they will build theorganisational learning that shapes competitive advantage fora decade. Around the world, 2025 may prove to be the year when Physical AI(PAI) – the merger of physical systems with AI – definitively movedout of the realms of science fiction and into mainstream businessconsciousness. What has long been regarded as something for thefuture is now emerging as a practical reality, driven by cheaper andmore capable hardware, and by software that learns how to learn. AI-driven machines bridge digital intelligence and formatsto bring physical intelligence to bear in a wide-range of differentcapabilities and environments How can we use AI to enable robotics? Advanced Artificial Intelligence (Cont’d) Generative AI Robots can be programmed to carry out physical actions,navigate environments and manipulate objects in real world Create new and original content based on algorithmstrained on large datasets Computer Vision Edge Services Hardware at the network’s entry points that locally processdata reducing latency Key to enabling real-time perception, train ofrobots/machines by observing human behaviours, “map”physical spaces for simulations etc. Advanced Artificial Intelligence Machine LearningAlgorithm enable computers to learn and improve Agentic AIAgentic AI systems can fast-track decision-making without from experience requiring constant human input Deep learningNeural networks model complex data patterns to detect anomalies This paper does two things. First, it explains why 2025–26 marksa genuine inflection point: why the technology has reached athreshold of practical viability, and why the competitive andgovernance environment is accelerating adoption. Second, andmore importantly, it provides a structured, practical framework forbusiness leaders navigating PAI adoption – where to start, how to sequence investment, and what organisational foundations must bein place for technology to deliver its potential. The focus throughoutis on PAI and industrial robotics: the sector where value is beingproven today and where the hard lessons of implementation arebeing learned in real time. AI-Enabled RobotAI-enabled robot combines physical AI with robotics and is a powerful tool with cross-industry applications Key Form Factors Autonomous Mobile Robots (AMRs)AMRs utilize intellignet navigation to freely move HumanoidsHumanoids robot replicate human characteristics and fucntionality to augment the workforce and increase output through environments, operating across diverse settings Autonomous VehiclesAutonomous vehicles are similar to other robots in bothdesign and intelligence Task SpecificTask specific robots have a less prescriptive form factor and are more used for repeatable process and tasks QuadrupedsQuadrupeds harness stability, agility, and adaptability to DronesDrones are flying machines that can enable autonomous navigate challenging environments modes through physical AI and other technologies PAI in the real world A global race: With governancecatching up Why now? The convergence that changes everything PAI’s move from theoretical potential to commercial reality is theresult of several simultaneous advances that have reached a criticalthreshold together. For the public, perceptions of PAI have been shaped less byfactories than by fiction. More recently, everyday encounters withsemi-autonomous robotic vacuum cleaners and lawnmowersalongside news stories a