Humanoid robots in 2023: where the industry actually stands
A sober look at the state of humanoid robotics by the end of 2023: what is genuinely ready, what remains demonstration, and what to track.
2023 was a busy year for humanoid robotics announcements. Figure AI, Agility Robotics, Tesla with Optimus, Boston Dynamics - companies were showing new versions, demonstrating work in industrial settings, announcing partnerships. The media coverage creates a sense that the revolution is close.
I prefer to look at this a little differently. Not to diminish the progress - it is real. But to understand where we actually are and what this means for business in the near term.
What has genuinely changed in recent years
The hardware has become significantly better. Humanoids have learned to walk on uneven surfaces, recover from falls, manipulate objects with sufficient precision for a range of tasks. This is the result of years of investment in mechanics, actuators, and balance control systems.
An important shift is happening on the software side. Applying machine learning methods to robot control - reinforcement learning, imitation learning - has substantially accelerated the development of new capabilities. Previously every movement had to be programmed manually. Now a robot can "watch" a human and repeat the movement.
Where the large gaps remain
Despite the progress, the gap between demonstration videos and reliable real-world operation remains significant.
Demonstrations typically show specially prepared scenarios. A specific task, controlled conditions, a few hours of tuning before filming. Real production is variability: different lighting, different object positions, unpredictable situations. Reliable operation under such conditions requires far more robustness than demos show.
Speed and throughput - another gap. Existing humanoids are significantly slower than humans at most manual tasks. For some tasks this is acceptable, but for many operational applications it is critical.
Cost and maintenance - these are still systems priced at hundreds of thousands of dollars with non-trivial support requirements.
What to track over the next year or two
I look at a few things as signals of genuine maturity.
Announcements of commercial deployments. Not pilots and partnerships, but actually working systems with measurable economics. BMW and Amazon have begun working with Agility Robotics - an interesting signal, but the results are still to come.
Tasks, not general-purpose application. Humanoids will most likely find their first productive applications in narrow, well-defined tasks - moving boxes of a specific size, assembling specific products in stable conditions - rather than as universal workers.
Progress in reliability. Mean time between failures in real conditions is the number that separates a pilot from industrial application.
A practical position for a leader
If you manage manufacturing or logistics - it is worth tracking this space, but not expecting commercially mature solutions within the next year.
A reasonable position: observe the development, understand which specific tasks in your operational context could become candidates for this technology, and consider joining early pilots in one to two years - with real expectations, not with the illusion of quick ROI.
Humanoid robots are genuinely working technology developing quickly. But "quickly" in the context of complex mechanical systems still means years before broad industrial application.