I have to disagree with that. Most positions in factory jobs exist because human labour is very cheap, especially in terms of flexibility. I doubt there are many positions where a robot with a less humanoid shape wouldn’t do a better job than a human or a humanoid robot. It’s just often cheaper to employ these workers because you pay them a salary, either on a hourly basis or on a monthly one, yet don’t have to worry about maintenance. With robots you have less hourly costs, but a much bigger overhead, as you now have to hire qualified technicians to perform regular maintenance on those machines, and also semi-regularly order replacement parts. These costs will rise alongside the complexity of these robots. And humanoid robots are much, much more complex than industrial robots, especially as they need to incorporate a lot more sensors that most industrial robots just won’t need. Sensors that might be very sensitive or require regular calibrations to ensure they work properly. That doesn’t come cheap.

Even when we look over the costs, humans will always be more versatile than robots. Give a person a book on how to do a job and they will perform it with the help of the books, and develop their working style to even work more efficiently. In contrast, robots would need a much more thorough training in order to work properly. This could be done traditionally by hardcoding the logic, or by using neural networks, which would be more intuitive, but are prone to create undesired results if one doesn’t have a good eye for the involved factors. And this process would need to be repeated for each job, and again if jobs would be fused together. And of course one would have to adhere for hardware limitations. A processor can only work so fast, and there are limitations on storage space, data transfer speed and reliability that also come to play when it comes to saving the training data.