Using robotics for factory automation is a bad idea. The reason is, that robotics is researched in universities while factory automation is a practical application. Both domains doesn't fit together and the reason why is explained in the following post.
Factory automation has to do with increasing the productivity. The idea is, that the company uses tools which helps to reduce the costs of doing a task. Typical examples of these tools are working gloves for the employees, barcode readers and CNC drilling machines. These tools are utilized in real applications because they are providing an added value. The company buys the new CNC machine for 30000 US$ and after a few month the machine has made a profit.
In contrast, academic robotics works the other way around. The reason is, that within AI research the question is not how to increase the productivity of existing workflow, but the problem is how to realize artificial life. In an artifical life project, the robot and the technology itself is important. Which means, that the AI experts are building a complicated structure which doesn't fulfill external needs but the aim is to explore something not known before.
It doesn't make sense to transfer Robotics projects from the academic domain into real world application. It will simply not work. Let us take a typical example. The Nao robot costs around 15000 US$ and it was used in many successful academic robotics project. Successful means, that the project has resulted into an academic paper which was perceived by other robotics experts as a valuable contribution to robot research. Additionally, the Nao robot was utilized in synthetic challenges for example the robocup competition. So we can say, that the price for the device is fair.
But what will happen if a company who is producing car buys 10 of these robot with the aim to increase the productivity at the assembly line? Sure, technically they can start such a project, but the prediction is, that the money is wasted. They won't increase the productivity by a single percent. The reason is, that Nao is without any doubt a robot but not a CNC machine. It can be used only in an academic context to explore AI problems, but it won't help workers in a factory.
Somebody may argue, that the true reason is the size of the device. And indeed, the Nao robot has the height of a toy robot. So a factory comes to the conclusion, that they should buy something which is capable of doing industrial tasks. A typical robot from that domain is the Rethink Robotics Baxter which is sold for around 30000 US$. Similar to Nao, Baxter is a very successful robot. It was used in many synthetic challenges and was used by universities to research new vision algorithms and graph planning systems. The surprising fact is, that even the Baxter robot won't help a car factory to increase the productivity. The reason is, that Baxter is a real robot who was designed for AI applications in mind. It's not a CNC machine and it's not a tool.
An interesting attempt to automate the service industry was realized by Joseph Engelberger in the 1990s with the helpmate robot. It was a transport vehicle for supporting employees in a hospital. The idea was to take the latest robotics technology and utilize it for increasing the productivity in a hospital. Simliar to other attempt of robotics automation the project failed. The customer didn't like the device and after a short period the company went bankrupt. What was wrong with the helpmate robot? The robot itself was great. In an academic project it was a high end device. The problem was, to utilize the machine for real world applications. Only in synthetic challenges the robot would perform successful. A synthetic challenge is a game not available in reality. For example the game is about pick&place objects in an area. For such a challenge, the helpmate device is a great choice. The problem is, that between the synthetic challenge and a real hospital there is a large gap. And the gap will become larger, if the robot is more advanced. The reason is, that the synthetic challenges are designed for the robots. For example, the latest iteration of the robocup competition was designd in a way, that the existing robots can fulfill the challenge. Even it looks like soccer, it's different.
Perhaps it make sense to investigate the helpmate hospital robot in detail to explain what the problem is. The first thing to do is to imagine a synthetic challenge in which a robot has to deliver objects in a hospital. So we need a stage with a floor, rooms and a concrete task. It's up to the robot to solve the challenge. The interesting point is, that it's possible to program the helpmate robot in such way. At the end, he will avoid the obstacle and is able to deliver all the objects to the goal position.
Great, so the problem is solved right? No it's not, what the robot has demonstrated is that he can solve a synthetic challenge. But a real hospital is working very different. The reason is, that in a hospital the employees have no time to participate at a robot challenge, but they have to solve a different task. The inner working of a robot challenge is defined by the challenge itself and it's described in the rule book. A synthetic delivery challenge contains of rooms, objects and obstacles.
The problem is, that real world applications and synthetic challenges doesn't fit together. The main difference is, that a real hospital is asking for tools which increase the productivity. But a robot is not a tool but it's an Artificial Intelligence. This social role conflict can't be fixed. What happens in reality is, that the gap between practical applications and academic robotics research will become larger. A hospital will not use robot for the delivery task, but they are prefering human workers with a handcart. On the other hand, a university who is researching robotics, will develop new synthetic challenges which fits well to their need of researching pathplanning algorithm in detail.
The simple conclusion from the past is, that it's not possible to convert a robot into a commercial product. It's only a subject of synthetic challenges but robots are not useful for real world applications.