Humanoid robots have moved from research labs and science fiction into investor presentations and factory pilots. Companies such as Boston Dynamics, Figure, Agility Robotics, Tesla, Unitree, Apptronik and many others are developing machines with two legs, two arms and a head. Why the human shape — and how close are these robots to being useful?

Why build robots shaped like people?

The main argument is practical: the world is built for humans. Stairs, doors, tools, shelves, vehicles and workstations all assume human proportions. A robot with a human form could, in principle, work in existing environments without redesigning them, and use the same tools people use.

A second argument concerns learning. Large amounts of video show humans doing tasks; a humanoid body may make it easier to learn from that data.

Why now?

Several trends converged:

  • AI foundation models that understand language and images give robots better perception and a way to follow instructions.
  • Robot learning advances, including large shared datasets such as the Open X-Embodiment collaboration, which pooled data from many robot types.
  • Simulation lets robots practice millions of times virtually before transferring skills to reality.
  • Cheaper hardware: motors, batteries, sensors and compute have improved and fallen in price.
  • Labor shortages in manufacturing, logistics and care in many countries create demand.

What they can do today

Current humanoids are most capable at:

  • walking, balancing and recovering from pushes;
  • picking up and moving objects such as boxes and totes in structured settings;
  • repetitive tasks in pilot deployments at factories and warehouses;
  • following simple spoken instructions in demonstrations.

Many impressive videos show carefully prepared scenarios, sometimes with teleoperation (a human remotely controlling the robot). Separating autonomous performance from remote control is one of the first things to check in any demo.

The hard problems

  • Dexterity: human hands are extraordinarily capable; robotic hands remain fragile and limited.
  • Reliability: factories need near-continuous operation; robots must work for thousands of hours without failure.
  • Generalization: handling new objects, layouts and surprises remains difficult.
  • Safety: heavy, powerful machines working near people require rigorous safeguards and standards.
  • Battery life: a few hours of operation before recharging limits usefulness.
  • Cost: prices must fall far enough to compete with human labor and simpler automation.

Humanoids vs. specialized robots

For many tasks, wheeled robots, robotic arms or conveyor automation are cheaper and more reliable. Humanoids make sense where flexibility in human environments outweighs their complexity. Expect the first large deployments in industrial settings, long before household robots become common.

Common misconceptions

  • Every demo shows autonomy. Some demonstrations use teleoperation or scripted motions; check the details.
  • Humanoids will replace all factory automation. Specialized machines remain more efficient for many repetitive tasks.
  • Home robots are imminent. Homes are the most varied, unstructured environments; industrial uses will come first.

Why the humanoid shape?

Human environments — stairs, doors, tools, shelves — are designed for human bodies, so a humanoid could in principle work anywhere people do without redesigning spaces.

What to watch next

  • Paid deployments, not pilots. The milestone that matters is robots doing productive work for paying customers over long periods.
  • Autonomy disclosure. Expect more scrutiny of how much demos rely on teleoperation.
  • Hand hardware. Breakthroughs in durable, dexterous hands would unlock many more tasks.
  • Safety standards. Formal standards for humanoids working near people will determine where they can be deployed.
  • Unit costs. Price declines through mass manufacturing will decide whether humanoids compete with human labor and simpler automation.

Key takeaways

  • Humanoids promise to work in environments and with tools designed for people.
  • Foundation models, shared robot datasets, simulation and cheaper hardware drive the current wave.
  • Today they mainly walk, balance and move objects in structured pilot settings.
  • Dexterity, reliability, generalization, safety, battery life and cost are the main hurdles.
  • Industrial deployments are expected well before household robots.

Frequently asked questions

Are humanoid robots available to buy?

Some research and developer models are sold, and companies run commercial pilots, but general-purpose humanoids for consumers are not yet a mainstream product.

How do humanoid robots learn tasks?

Through combinations of human demonstrations, teleoperation data, simulation and reinforcement learning, increasingly guided by vision-language models.

Are they safe to work near people?

Safety is a major focus, with standards and safeguards being developed. Most current deployments separate robots from workers or operate under strict controls.