More Than 200 Robot Makers, One Big Question: Are Humanoids Actually Ready?

A new industry report cuts through the excitement around humanoid robots to ask what works today, what is still years away, and what it all means for workers in factories and warehouses.

AI2Day Newsdesk3 min read
Photoreal news-editorial style, 16:9 framing, full-frame edge-to-edge composition
Share

Key points

  • Over 200 verified humanoid robot companies are now active worldwide, according to a new report from The Robot Report.
  • Semi-humanoid and non-humanoid robots that move around and grab things are already operating in real factories and warehouses.
  • Key engineering problems still unsolved include battery life, affordable all-around sensing, and teaching robots new skills at scale.
  • A company called Brightpick combines a mobile robot base with a gripping arm for warehouse tasks, and its CEO says specialised robots still beat general-purpose ones for most jobs.
  • Researchers and companies are experimenting with recording tasks from a first-person point of view to build richer training data for robot AI.

Humanoid robots, machines built with a torso, two arms and two legs to move through human spaces, have attracted billions of dollars in investment over the past year. Headlines have followed every unveiling. But a report published by The Robot Report asks the harder question: which of these machines can actually do a job today?

The short answer is that most humanoids are still in trials or development. The robots quietly arriving on factory floors right now look nothing like science fiction. They are boxy, wheeled platforms with one or two arms bolted on, built for one task done reliably rather than for looking human.

What problems are engineers still trying to solve?

Three challenges keep coming up. First, power: a robot that runs flat after two hours is useless on a warehouse shift. Second, perception, meaning the cameras and sensors that let a robot understand its surroundings. Full 360-degree awareness is possible but expensive, and bringing that cost down is an active area of work. Third, skills training: teaching a robot to handle a new object or a new task still takes far more time and data than anyone would like.

Chip maker Lattice Semiconductor, in a section of the report aimed at hardware developers, flags a fourth concern specific to robots that work near the public: security. A humanoid in a hotel lobby or a hospital corridor is a potential target for tampering, and the report recommends building safety checks into the hardware from the start, not patching them in later.

What does this mean for workers?

Right now, most of these machines are handling repetitive pick-and-place tasks in warehouses, the kind of work that is physically hard and often causes injury. Jan Zizka, chief executive of mobile robot company Brightpick, argues in the report that a robot designed for one job will outperform a general-purpose humanoid for years to come. That is a meaningful claim for anyone wondering whether a warehouse robot will replace a whole team or just take over the heaviest, most tedious parts of the shift.

The report also covers Palm Garden AI, which is training humanoid robots for hospitality and therapeutic care settings, places where how a robot moves and speaks matters as much as whether it can carry a tray. The company is developing what it calls a "Coherence Guard," a software layer intended to keep a robot's social behaviour predictable and appropriate.

For skills training, a company called Unidata is collecting video and sensor data from a first-person viewpoint, essentially recording tasks through the robot's own eyes, to build better AI training sets.

What happens next?

The honest picture is incremental. Specialist mobile robots are already earning their place in logistics. True humanoids, the kind that can move between jobs the way a human worker can, remain a work in progress. The industry experts interviewed for the report draw a comparison to the early days of automotive manufacturing: enormous potential, but the path from prototype to reliable production line took decades, not months.

© 2026 AI2Day