物理人工智能，第二部分：人形机器人

Physical AI, part 2: Humanoid robots 12 March 2026 Key takeaways •AI is redefining robotics, enabling a new generation of humanoid robots that can perceive, learn and act autonomously in thephysical world. After decades of incremental progress, the field has reached a turning point, with breakthroughs in generative AIand large language models accelerating the rise of "embodied intelligence" and reshaping expectations for what service robotscan accomplish. •The humanoid robot population is poised for take-off. BofA Global Research projects that the number of humanoid units inoperation could reach three billion by 2060, surpassing cars on a per‑capita basis. While industrial and service applications areexpected to lead adoption in the near term, household humanoids are projected to accelerate quickly and ultimately account forthe largest share - 62% - by 2060. •In part 2 of our physical AI series, we outline what you need to know about humanoid robots - including how they work, howthey're being used today and key challenges the industry faces as it moves toward large-scale deployment. Humanoid robots 101 Humanoid robots are a type of service robot designed to move, perceive and interact in ways that resemble human beings.As atype of physical AI, they embody artificial intelligence in a dynamic, mechanical form capable of acting on the world (for readersnew to this topic,Physical AI, part 1: The basicsprovides a helpful foundation).This enables them to operate in complex,unstructured environments, rather than controlled settings. Unlike traditional service robots, which are built for narrow,repetitive tasks, humanoids are required to navigate real-world spaces, interpret human behavior and respond through physicalaction–placing far greater demands on AI for sensing, motion control and natural-language interaction. After decades of incremental progress, the field reached a turning point in 2023, as breakthroughs in generative AI and largelanguage models (LLMs) accelerated the emergence of“embodied intelligence.”By giving AI a physical form, humanoid robotscan learn directly from the world, translate language into action and adapt to people in real time–reshaping expectations ofwhat robots can do. Humanoid adoption is accelerating – but why?According to BofA Global Research, humanoid robot adoption is speeding up (Exhibit 1), mainly because of: 1.AI maturity: Model breakthroughs are now enabling more general-purpose learning, better long-term reasoning and morereliable AI operation in unstructured, real-world human environments.2.Hardware economics: Rapid cost declines and improvements across key humanoid components (e.g., batteries, motors,sensors, actuators and onboard AI compute) have lowered the required bill of materials (hardware costs) and boostedoverall feasibility of humanoids.3.Industrial overlap: Humanoid robots can tap into existing supply chains, manufacturing methods and capital investmentfrom electric vehicles (EVs), autonomous tech and AI hardware sectors, helping them scale faster and reduce executionrisk.4.Demographics: Ongoing labor shortages, aging workforces, wage inflation and high turnover make flexible robotic laboreconomically appealing across industries, such as manufacturing, logistics and services, even before humanoids fullymatch human abilities.5.Deployment readiness: Humanoid form factors fit with existing tools, buildings, and workflows, which cuts down ondeployment barriers. Together, improved safety, autonomy and self-learning reduce integration, regulatory and operatingfriction. Exhibit1:Tech breakthroughs, economicsand demographicsarekey driversbehindhumanoid adoptionInfographic illustrating factors accelerating humanoid robot adoption Scaling-up from pilot to production The robot rush: Increased investment is happening nowThe commercial race to develop humanoids is heating up as the tech matures and prices decline. Investments have risen sharply –from $0.7 billion in 2018 to $4.3 billion in 2025.1And as of January 2026, over 50 companies were in the process of buildingtheir own humanoids,2with 150 commercial launches to date.3So far, the development landscape has bifurcated betweencompanies vertically integrating robots in their own factories and those selling or renting humanoids to their customers. A robot population boom is on the horizonBofA Global Research forecasts annual humanoid robot shipments will surge from 20,000 units in 2025 to 10 million by 2035, implying an 86% compound annual growth rate (CAGR) (Exhibit 2). In 2026 alone, sales are expected to reach 90,000, as leadingcompanies ramp up production capacity and scale deployments. Exhibit3:The humanoidrobot “population” could reachthreebillion by 2060E Exhibit2:Annual humanoid robot shipments are projected to reach1.2 million in 2030 and 10 million by 2035Annualglobalhumanoid robot shipments(thousands (k)) Long-term forecast of humanoid robot units in operation (UIO) byapplication (millions (mn