A humanoid robot is a machine designed to resemble and move like a human: two legs, two arms, a torso, and a head. Unlike factory arms bolted to the floor or wheeled delivery bots, humanoids are built to navigate the same spaces humans use — walking down aisles, picking objects off shelves, and fitting into infrastructure designed for human bodies.
For decades they were expensive lab curiosities. What changed is AI. Large language models and advanced computer vision now give robots the ability to understand context, plan multi-step tasks, and adapt in real time to unpredictable environments — capabilities that simply didn’t exist before 2022.
How They Work
A modern humanoid combines three systems: mechanical hardware (joints, actuators, force sensors), onboard AI (vision, reasoning, motion planning), and control software that ties everything into a continuous feedback loop.
The hardest part is bipedal locomotion — walking upright on two legs. Unlike wheels, two-legged movement requires the robot to constantly recalculate its balance. Add dexterous manipulation (picking a specific bolt, plugging a cable connector under uncertainty) and real-time environment perception, and you have a system where every component depends on the others. A sensor delay ripples into an imbalance, which throws off the arm, which drops the part.
Today’s models run AI inference directly on the robot’s onboard hardware — no cloud connection needed — so they can respond in real time. Most use camera-based vision, force sensors in the hands, and neural networks trained first in simulation and then fine-tuned on real-world data.
What They Can Do Today
The most mature deployment is factory work. At BMW’s Spartanburg plant in South Carolina, Figure AI’s Figure 03 performs component sequencing on the X3 assembly line — picking specific parts from bins and placing them in precise order. Agility Robotics’ Digit has moved more than 100,000 totes at a GXO Logistics warehouse and is also deployed at Amazon.
Beyond automotive and warehousing, early pilots are running in hospitals (medication transport, room sanitization) and in outdoor inspection tasks. In every case, the robot handles a narrow, well-defined job in a controlled section of a larger facility — not acting as a free-roaming general assistant.
Who’s Building Them
Several companies have working robots in or near active deployment:
- Figure AI (Figure 03): automotive logistics — BMW
- Agility Robotics (Digit): warehouse tote handling — Amazon, GXO, Toyota
- Boston Dynamics (Atlas): industrial material handling; first commercial units shipping mid-2026
- Tesla (Optimus): internal factory use for learning data collection; first commercial units expected late 2026
- Unitree Robotics (G1, H1): lower-cost models available today; widely used for research
- 1X Technologies (NEO): targeting the consumer home market from 2026
Prices range from about $16,000 for research-grade units (Unitree G1) to $250,000 or more for industrial deployments.
What They Can’t Yet Do
Today’s humanoids are narrow specialists, not general assistants. Most excel at two or three specific tasks in structured environments. Unexpected variations — a new box shape, a wet floor, a repositioned shelf — can stop them entirely.
Battery life limits a typical robot to two to four hours of work before returning to a charging dock. Most walk at one to two meters per second, roughly a slow human pace, and none can run, climb a ladder, or handle highly varied tasks without retraining.
Cost remains a barrier for all but the largest manufacturers. At $150,000–$250,000 per unit, the return on investment only works for high-volume, repetitive tasks where labor is scarce or hazardous.
What Comes Next
Industry analysts expect humanoid robots to remain primarily an industrial tool through the early 2030s, with the consumer home market opening later — when prices are expected to fall below $20,000 and reliability has been proven at scale. The trajectory resembles early factory automation: slow adoption at first, then rapid expansion once ROI is demonstrated.
One distinctive feature of the current moment: AI software is advancing faster than the mechanical hardware. Software updates alone can meaningfully expand what a deployed robot can do — meaning the units already in factories today will become more capable over time, without a hardware replacement.
For those tracking the technology, IEEE Spectrum’s robotics coverage is the most reliable source for technical depth and deployment news.
In the News
Two recent stories mark milestones for the industry. Figure AI’s Figure 03 began active logistics work at BMW’s US plant, making it one of the first humanoids in production on an automotive assembly line. Separately, Agility Robotics announced a SPAC merger valuing the company at $2.5 billion — a signal that investors see a viable commercial future for the technology.
FAQ
Why build a robot shaped like a human?
Because the world — factories, warehouses, offices, homes — is built for human bodies. A robot with two arms and two legs can use existing infrastructure without costly retrofitting.
How is a humanoid robot different from a chatbot?
A chatbot generates text; a humanoid robot takes physical actions in the world. Many use similar underlying AI (language and vision models), but the output is movement rather than words.
Are they safe to work near?
Current industrial deployments use physical barriers and speed limits around humans. Standards for unprotected human-robot collaboration in shared spaces are still being developed by bodies such as ISO and OSHA.
Will they replace jobs?
They are being deployed first in roles that are physically demanding, repetitive, or hazardous. Most economists expect them to shift the type of tasks humans do rather than eliminate work outright — consistent with previous waves of industrial automation.
Sources: IEEE Spectrum, BMW Group press release, Agility Robotics SPAC announcement, Wikipedia (Humanoid robot).