Humanoid Robots

A humanoid robot is a robot built with a body plan resembling that of a human being — typically a head, a torso, two arms with articulated hands and two legs used for walking — so that it can operate in workplaces, homes and public spaces built for people, rather than in purpose-built industrial enclosures. Long a research curiosity, the field moved toward commercial deployment in the mid-2020s as artificial intelligence made it possible for robots to learn general tasks from experience rather than follow pre-programmed scripts.^[1][2]

History

Humanoid research dates to the 1970s, when university laboratories in Japan built the first walking machines, and to iconic demonstrations such as Honda's ASIMO, unveiled in 2000, which could walk, climb stairs and greet people but remained a costly showpiece. For decades, control software limited such robots to rehearsed routines in controlled settings.^[1]

The modern wave began in the early 2020s. Tesla announced its Optimus programme in 2022 and framed humanoid manufacturing as an extension of its electric-vehicle production expertise; startup Figure AI, founded in 2022, began targeting warehouse and factory work; and China's Unitree launched low-cost research humanoids such as the G1. AGIBOT, founded in 2023, focused on mass production of embodied robots, and Boston Dynamics unveiled a fully electric version of its Atlas robot in April 2024, later paired with Google DeepMind's Gemini Robotics foundation models under a partnership announced at CES 2026.^[1][2][3]

By late 2025 the first volume shipments had begun. AGIBOT reported rolling out its 5,000th mass-produced humanoid and was ranked first globally in 2025 humanoid shipments and market share by the analyst firm Omdia, with more than 5,100 units and 39 per cent of the market. Unitree said it shipped more than 5,500 units during 2025 — more than all other humanoid makers combined, by its count — and listed its G1 robot on Amazon in February 2026 at US$17,990.^[4][3]

Key Concepts and Technology

Modern humanoid robots pair mechanical hardware — electric actuators, dexterous hands and battery packs typically good for one to two hours of active work — with AI models that map vision and language instructions directly to movement. These vision-language-action (VLA) models allow a robot to be told what to do in natural language rather than being reprogrammed for each task, and they are trained with reinforcement learning in simulation and with motion data collected through teleoperation, a process known as sim-to-real transfer.^[1][6]

Capability varies widely. Tesla's Optimus Gen 3 redesign centres on 22-degree-of-freedom hands with 50 actuators, aimed at fine manipulation in factories, while Google DeepMind's Gemini Robotics 2 targets long-horizon planning — chaining multi-step tasks in unfamiliar environments. Pricing spans research devices such as Unitree's US$16,000 base G1 to enterprise fleets priced per hour of work.^[3][4][6]

Applications and Impact

Manufacturing and logistics are the first commercial battlegrounds: Tesla deploys Optimus inside its own plants, while Figure and Apptronik have run paid pilots with industrial and logistics customers such as Mercedes and GXO. Hospitality, tourism, retail reception, security patrol and education are earlier-stage markets, and home use is emerging through subscriptions such as 1X's NEO, offered at about US$20,000 or US$499 per month for early adopters willing to help train the robot. A robots-as-a-service (RaaS) model, in which customers pay for robot labour rather than buy hardware, is spreading alongside outright sales.^[2][3][5]

The technology raises familiar questions about job displacement in warehousing and manufacturing, safety around people, and the reliability of marketing claims versus deployed capability; analysts note that most 2026 deployments remain pilots, supervised trials or showroom installations rather than unattended general-purpose labour.^[3]

>See Also

• Physical AI
• Autonomous Vehicles
• Reinforcement Learning
• Computer Vision
• AI in Malaysian Manufacturing
• NVIDIA