Game building robots

What is the reason why self-driving cars, robot soccer and rescue robots failing always? In the 1960s the answer was easy, because the software of the 1960s was not highly enough developed. Computers in that time had only a RAM of not more than 2 Kilobytes and not operating systems were available. These problems from computer stone age are mostly solved but the problems remains the same. An autonomous robot won’t solve the maze and even if he is doing so he has lost the game. But is the problem?

The problem with robots is, that they are creating a new game on top of the old one. A game is a computer program in which certain actions are valid while others not. Let me give an example. Suppose there is a maze on the floor and the task for the robot is to find a way through that maze. From an technical point of view, there is a game, called maze, a robot who acts in the game and a software installed at the robot who acts hopefully intelligently. Thet is the setup of every micromouse challange and the precondition in which the AI is created by the teams. But is this understanding the reality? Not exactly. What in reality is there is the following. The original game is called “maze game”. If a human player is actor in that game he is trying to do the task right. That means, he plays the game according to the original rules. If a robot plays that game, he can’t play it directly. Instead a second game is created in the memory of the computer and this second game works under different rules than the original one. The first original game is about a maze solving problem. It contains of the layout at the floor, a timelimit and lots of obstacle. This kind of game is different from the software which runs in the robot. The software in the robot is about something else. It is about an operating system called robotics firmware. On top of the firmware, the c compiler executes a program, and in the program some subfunctions and classes are defined. The original game is about a maze solving problem, but the robot has created an operating system plus a C sourcecode. With a bit searching we can identify in the robot’s software also a maze game. But this maze game is different from the first one. Both games are not synchronized.

Let us observe what happens, if we are starting the software in the robot. We press the run button and the software prints something on the screen. From the point of view of the robot, he is the boss. That means, the robot has started a game, and the world around the robot has to play this game. And here we have the conflict. What the robot should do is to play a game, which is already there and which is located outside the robot. That means, the robot should be behave as a subordinate and follow existing rules. What the robot is doing in reality is to ignore all rules and create his own game. The environment has two options. Either it will play the game of the robot. Then everything is great. The robot will say, if the world has won or lost. The other option is, if the world is playing it’s own game and ignores the robot needs. Then the robot will print an error to the screen, and the project has failed.

Perhaps we should describe a situation which is working at best for the robot. If the world is following the rules of the robot software, no problem is there. What robots can do great is playing the judge. The robot is the boss, and everybody else is wrong. That means, the robot is the dominant part. With this background knowledge we can describe in detail what is wrong with robocup rescue, micromouse and all the other robot challenges. The reason why they look staged is because the challenge is build around the robot needs. That means, in the micromouse challenge is not driving in a maze, but in reality the robot decides what is wrong with the maze. The robot can not make a mistake, only the environment is wrong. The maze has not understood what the robot wants.

Creating an environment in which player should act according to rules is called “game design”. Computergames are a well known discpline and they are successful application of computer technology. The game can never be wrong. Only the player in the game can behave the wrong way. What computers are great is to realize such games. It is possible to program at a computer any kind of challenge. A maze problem, a jump’n’run game or a driving challenge. The more interesting question is, if it is possible that software can solve these games? Some reearchers are saying yes, and as a proof they presenting chess engines, Starcraft playing software or AI software which can play pacman. But did the computer really play the game? No he don’t. AI is only possible in the fairy tale not in reality

Perhaps a small example. Suppose we are starting the Mario game, then we are activating the AI controller which can play Mario alone. Now we have to things started in parallel. The well known jump’n’run game and ontop an AI system which plays Mario on a near human level. But why is this not an Artificial Intelligence? Because the human player remains in the loop. In reality the situation is more complicated. There is the Mario game, the Mario AI Controller and in front of the screen is the human who observes the scene. What the human sees on the screen is a computer game. And the game is about teaching Mario. The task for the human is to observe the AI controller and detect what he is doing right / wrong. That means, the game can’t be played without the human, he is needed in the loop, even if Mario makes everything alone.

What does it mean? It means, that the AI Controller doesn’t solve the game, but the AI controller is creating the problem. The new problem is called “judge the AI”. And if this problem is solved the next problem is called “judge the judge of the AI” and so forth. If the last instance in this chain is not there or has leaved the room, the game is over. That means, the game is lost.

To understand the situation better we have to introduce the terminology of automation level. The automation level describes the amount of human in a task. A low automation level means, that the human is doing everything while a high-automation level means, that the AI is working great. how exactly is the automation level of AI software who can play games? Perhaps a look back into the year 1996 will answer the question. In that time, DeepBlue has won in computerchess against Kasparov, https://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov Please estimate the automation level? Was it 100% because the AI software played the game against Kasparov? According to some comments that was the case. The newspaper have written that it was a match man versus machine and man hos lost against the machine. But in reality, it was not a battle, it was a different kind of interaction. Kasparov ha played himself, that means he was the grandmaster and the world best chess player. And Deepblue has played the role of a beginner how needs advice. Deepblue was unsure in every move he made, he had no idea how to play chess right. Kasparov told him how the game works. It was an asymmetric realtionship. Kasparov was the grandmaster, the teacher and the father, while Deepblue was the beginner, the student and the son. The automation level during the game was 0% because Kasprov has played fair. He thought about every move and was concerned if Deepblue would understand it right. Kasparov didn’t lost the game, because he never played against the machine. He played with the machine on the same side. Kasparov and the software have worked together to show the world what chess is.

Advertisements

The joy of teleoperated robots with semantic annotation

Artificial Intelligence is one of the hardest problems in engineering and most projects with that background failed in the past. Usually the scientists are discussing the technical reasons, but perhaps the project itself was started with the wrong idea. The assumption is, that with neural networks, expert systems and path planning algorithm is everything perfect instead the projects failed because of reasons outside of Artificial Intelligence. What does that mean? A robotics project can be realized in two ways: autonomous robots vs. teleoperated robots.

Most AI projects from the past have a fully autonomous robot in mind. This is the case for Robocup soccer, Micromouse challenge, Mario AI challenge and the Amazon picking challenge as well. The teams are trying out different approaches and they are usually not able to fulfill the task. But what is, if the competition is held with a different goal in mind? From a technical point of view, it is possible to construct a manual control robot. The best example is a robotarm, controlled with a mouse or a micromouse robot controlled with a keyboard. The advantage is, that such systems are more easier to realize.

A teleoperated robot is a stable base. It is technology already available. That means, it can be teached in school how to program such systems and it is possible to improve the remote control on minor details. Sure, there is a difference between teleoperation and a real robot. In most cases, a robot is developed to replace the human. And if a human is needed to control the micromouse in the maze there is no real benefit. But in cases of high complexity and hard to solve challenge this kind of approach is the right one. The question which can be answered next is how to make teleoperation more difficult. And here comes semantic annotation into the game.

A annotated remote control is at foremost a teleoperated system with a human-in-the-loop. That means, without the operator, the robot will do nothing. The advantage over a normal RC car is, that the software is recognizing the events and puts them to the screen. That means, the operator sees the video signal and as overlay a textual description of what is going on at the screen. And this annotation can be made simple or more advanced. It make sense, that different teams are competing in realizing the best annotation for a teleoperation robot. In a robot challenge, all the teams are using remote control, but some of them have better technologies in the semantic parser. Such a challenge provides a more realistic entry barrier. The lowest possible submission would only a teleoperated robot without semantic annotation. That means, it is a remote controlled car. Such a lowlevel entry barrier make sense because this is the common standard available today. That means, with normal technology it is possible to control any device from remote. Such technology is cheap and researched very well.

Everything what is going beyond remote control can be called advanced. That means a semantic parser who is using convolutional neural networks is advanced stuff, while a fully autonomous system based on LISP too. What a team is building ontop of a remote control is up to them. It is not researched very well and it is unclear what the best technology is. It make sense to compare different approaches against each other.

Unfortunately, most robotics competitions are working quite different. The common standard for all is, that the robot will work autonomously. That means 100% of the teams are taking away the hands from the keyboard otherwise they have failed. Does it make sense to judge in that way? No, the barrier is too high. Perhaps one of team doesn’t have a fully autonomous robot but only provides a semantic annotation. This team would get 0 points in a robotics challenge, but their technology would be interesting. I think it it important to lower the baseline a bit and perceive remote controlled robots are a valid candidate in Artificial Intelligence research.

On the first look, augmented reality systems which are supporting a human operator are easy to realize. But a closer look shows, that it’s a complicate task to build the domain model. Even if the software shows only some annotations and prints a trajectory into the screen, the amount of software behind the GUI is immense. The good news is, that from day 1 such systems are production ready. That means, a teleoperated UAV is like a normal UAV able to fly in the air. It can also play games.

But perhaps we can reduce the difficulty a bit more? Normal robotics competitions are working with real robots. That means a RC-car or a UAV is needed first. The more cheaper way would be a simulated environment. The task is now to control a car in a game and write the semantic annotation parser. That means, it is a normal racing game which is played by human and the only exception is, that all the actions are recognized by the parser.

Can robots replace human worker?

One myth in robotics research is that today’s or future robotics are needed in the industry to increase the automation level. Companies like Kuka, Fanuc and Rethink robotics are selling the hardware and AI researchers are programming the software. This combination allows to provide pick&place robots, self-driving cars, and industry wide automation of processes. But is this vision realistic?

Let us take a look into some automation projects from the past. It seems, that automation and robotics are different issues. It is correct, that industry has a high demand for automation. That are machines for mass production of goods. For example a machine how is making pizza. But these machines are operated without robots. Instead the assembly line plus additional machines are the backbone in automation. And the remaining tasks are done by humans. But, it is a very few part of the total work what is done by humans. That means, without robotics the automation level is nearly 100%.

Let us focus on a different kind of machine. If the product is ready it gets packaged into boxes. At first let us take a look what robotics companies have to over. Modern pick&place robots for example the latest model of softrobotics is able to take the food and put it into the box. That means, the robot is doing all the work, and he is doing it very fast. On the first look, a softrobotics hand is the perfect example for great usage of robots in industry. But let us take a deeper look. Are these pick&place robots in daily use somewhere in the world? No they don’t. The reason is, that non-robotics automation has solved the issues decades ago. There are endless machines available which are able to packaging food into boxes. The surprising feature is, that these machines doesn’t need to be programmed and they are not working with a robot hand. Instead, the overall machine is the packaging robot.

I’m explaining these details to make clear, that in 10 years, in 20 years and even in 50 years robots won’t entering manufacturing. The problem has nothing to do with writing better software or making robotarms more flexible. The problem is, that non-robotics automation is working great, and the remaining manual work is done by humans cheaper than any robot can do.

This insight is perhaps surprising. Because pick&place robots at the assembly line are imagined as the perfect usecase for today’s and future robots. But exactly this domain won’t entered by robots. That means, that robots won’t replace a single worker, even if their software gets improved. The same is true for so called autonomous cars. Companies like Waymo are pushing the idea forward. The problem is the same like in manufactoring. Non-robotics based automation works too good, and for Artificial Intelligence there is no need.

What does that mean? On the first look, robots on the street would increase the productivity. That means, the AI software is controlling the car but not the driver. The problem is, that even without Artificial Intelligence, transportation was automated in the past. The best example is a bus. A single driver is able to transport hundreds of people. That means, the automation level of a bus is with a human driver very high. There is no need to increase the automation further.

Let us talk about the difference between mechanical automation and robotics automation. In both cases the idea is to reduce the costs. Most part of industry are highly automated. That means all the pizzas in the supermarket are produced with mechanical machines. Also the transportation is highly automatized. Automation means, that not horses are in charge to transport people or goods but diesel engines. This kind of mechanical automation is called industrial revolution. The progress in cost reduction was huge.

The promise of robotics is, to increase the automation level further. And this promise can’t be fulfilled. At the beginning the engineers believed it has to do with their robots. The computers in the 1970s were slow and the first robots for example Unimate wasn’t a success. Since the year 2000, the computers are faster and software engineering is better than ever. But even today, noone is aksing for robots. Especially not to automate an assembly line. This paradoxon is not fully understood yet. Instead only anecdotes are told. For example a company has installed a pick&place robot at their assembly line in the hope to increase the automation level. But a detail calculation came to the conclusion that no advantage is there. If the company replaces the robot by a pure mechanical machine and put a human at the assembly line they will save a lot of money. And this is what most companies like at most, reducing their cost. The surprising insight is, that robots are not efficient, they are useless piece of technology.

This problem can’t be overcome with better software. Even if Waymo improves his software and even Rethink Robotics will develop software which is 100 times better than today, they won’t sell a single of the robotics system. The problem is, that robots are not competing with man they are under pressure because of mechanical automation. That are classical machines which are invented for mass automation. These machines doesn’t need software, they are based on metal-technology invented 100 years ago. In case of industry it is often an assembly line with some extra machines. This kind of system is working great, and no robot in the world can improve the costs.

Let us take a look into packing workflow. Suppose, a company wants to package their product and then put the product into boxes for shipment. On the first look this kind of task is great for robots. Because it’s a hard task, it has to do with mass manufactoring and no humans want’s to make the job. So the company will buy the latest robot from Fanuc and is happy, right? No, they won’t. What the company is really doing is to research the market for mechanical packaging machines, buys some of them, put them into a serial line and at the end a human worker is doing the remaining manual work around the assembly line. That is modern old-school packaging and it is recommended for any company. It was the best solution in the 1970s, in the 2000 and also in the year 2030s. That means, no robot is in the workflow and this helps to save a lot of money.

So where exactly are the Kuka and Fanuc robots in usage? What is the usecase in which a pick&place robot make sense? Nobody knows. Perhaps some researchers have these robots installed in the basement for testing out fancy deeplearning software, but a normal industrial company has no demand for robots.

The question is not if robots or man are the better idea if a company has a packaging problem. The question is, if the mechanical machine A or B is the better choice. Both machines have in common, that no software is required, and that these machines are classical one, that means with an average automation level and without a robotic arm.

What does it mean, that robotics are not needed in industrial automation? It means, that an easy to automate task can’t be done by robots. If a software company has developed on top of ROS and Gazebo a pick&place controller which can control a robotic gripper, it is 100% sure, that this company won’t sell a single copy of this software. There is simply no demand for such a software, because real industry in the world, doesn’t have robot arms at the assembly line and they are not planning so. In contrast, if a company has developed a new all-mechanical machine for packaging boxes and the machine is not very expensive such a machine can be sold. There is a high demand for automation because the demand for food and especially mass produced food is high.

If robots are not useful in repetitive task in industrial context and if nobody knows how to create software for automate more complex tasks, where exactly can Robots be used in a meaningful way? Right, nowhere. Robots are useless in a factory, they are useless on the streets, in the household, they are useless in the classroom and especially in medical environments. That is the reason, why all robotics companies in the past for example Rethink, Unimate, Willow Garage were run into bankrupt or will do in the near future. Robots are not able to replace humans and they can’t replace mechanical machines.

This is called a paradox. From a theoretical point of view a possible explanation is, that the current automation level is at 80% and even a perfect programmed robot won’t increase the level anymore. This problem is called limit of automation and means, that even if a company likes robots very much they won’t use the technology because the costs are too high.

Packaging task

Let us take a look into packaging on industrial level. There are two possible machines available. The first one is a classical mechanical assembly line. That means, the products running through big machines with 300 items per minute and gets packaged into boxes. The other type of machine contains some robot arms, which are doing something with the products. Which kind of system is more advanced? The sad news is, that the pure-mechanical installation has the lower costs. It works more robust. In contrast the robotarm in the second installation slows down the process and is a sign of bad engineering work. That means, of the company has installed a robot and replaces classical assembly line they didn’t have understood automation and their strategy is a dead end.

Let us describe a packing task from a robotics only perspective. An AI researcher would argue, that this kind of task contains of subtasks. In the first step the robot takes the object, then he opens the box, then the object is put into the box, then the box is closed. And now the process is repeating over and over. Each step can be handled with a convolutional neural network for image detection and the actions are planned with an expert system. In the backend a semantic database is used and the programming is C++ plus Python. This is some kind of best practice method in AI software development in the year 2019.

And now let us compare this complicated pipeline with packaging in reality. What a real company is using is no robot at all. Instead the packing is done in a vertical packing machine which was designed 20 years ago. This system contains of some motors and the products are feed from the top. The machine doesn’t even have a microprocessor and no Artificial Intelligence is needed. That means, the packaging process can be done in a pure mechanical way. Or to explain it more obvious. The AI researcher didn’t have understand what packing is or what what the customers wants.

Let us make the counter thought experiment for testing the hypothesis. Suppose, a mechanical packing machine is outdated and modern software driven robots are the better idea. So the company doesn’t need any kind of mechanical automation and buys a modern robot arm with 6 degree of freedoms plus a sophisticated software. What will happen? The result is, that the production will stop. No products are packaged and after two days the company is bankrupt because the customer gets no products at all. The robotarm isn’t able to handle the task and even the best researcher from M.I:T. can’t solve the issue. Robotics is some kind of hoax who pretends to automate the manufacturing process but in reality they have nothing to offer. Robotics promises a lot and has a great media coverage but their benefit is low.

The upraising of spiritual robots

Many people are interested in the future. They want to know how exactly the world will look like in 10 years or in 20 years. Will robots take over? Will intelligent machines have replaced human workers?

The first important hint is, that robots and Artificial Intelligence isn’t able to increase the productivity. In some studies of the 1980s around the Halle 54 at the car manufactorer Volkswagen this was described as a productivity paradoxon. It means, that even after advanced robotics technology was introduced into the company the productivity isn’t higher. From an economical point of view, Robots are a disaster. They are costing a lot and their output is zero. Classical non-robotics automation techniques invented in the 19th centure before the first microchip was produced are the better technology in factory automation. That means, classical assembly lines and packaging machines are superior to robotics arm and software controlled systems.

If robots can’t be used productive in the industry, where exactly are they are used in future? The walking robots of Boston Dynamics and the self-driving cars of Google are reality. That means, the technology is there and they is evolving quickly. But not into useful machines which are helping the humans but future robots are some kind of fifth wheel on the carriage. Robots are there, they are present at the street and in the industry but they have no useful task. That means, the robotarm at the assembly line doesn’t packaging products, instead it is some kind of research project. And the same is true for walking robots which can be seen in the park. The robot walks around, the algorithm works better than ever but he has no real task. Instead, the robots will costs a lot and provide nothing in exchange. They are from an economical point of view a costfactor or a luxaruy. They are financed by humans and by pure mechanical machines. Future robots will have a social role comparable to a priest. That means, they have a fixed place in the society, they have something to do, but they are not used for producing goods and services. Instead robots are the problems.

Let us make an example. The vision of Waymo is, to revolutionize the transport business and replace normal taxis with self-driving cars. What will really happen in 10 years is, that the waymo cars are driving without a driver through the city, but with no guests on board. They are driving alone. Rich companies and private citizen have them bought as some kind of toy or for promotional usage. But these self-driving car are not able to replace real drivers. They won’t reduce traffic accidents and they won’t reduce the transportation costs. They will become some kind of useless invention without any purpose.

To understand the situation let us focus on the opposite: technology which is useful. A normal car or a non-robotics assembly line provides something. That means, a mechanical assembly line is bought by a company to increase their productivity. The idea is to reduce the costs and increase the output per day. The same is true for a normal car. The idea is to travel a distance for lower costs than without the car. Both are an investment. They were bought because of economical reasons. If a persons owns this technology he has a benefit. And now let us compare this with spiritual robots. These machines will have a negative balance sheet. That means, a company buys a dual arm robot for their assembly line and from a financial perspective the investment was a bad choice. The robot will need a lot of electricity, asks for a professional programmer and his output is zero. That means, the company is not able to benefit from the technolgy. The same is true for self-driving cars. These cars will costs a half million US$, they will drive completly autonomous, but their economic impact is negative. They are not replacing existing car or existing human drivers, they are operating on top of existing systems. Robotics and ARtificial Intelligence is generating additional costs and will charge the enviornment.

The idea of useless things isn’t uncommon in modern society. Many people in the world are owner of dogs and cats. These animals have no purpose in the family except from making life more interesting. The human has bought the dog to get a lot of trouble with him. And exactly this is the role of robots too. They will cost a lot but have no benefit. In the future, robots will be everywhere. They are present in schools, in the industry and at home. That means, but not as a helping hand. Their social role will become the opposite. That means, the robots who are making the most trouble are the favorite model. They are acting as some kind of professional trouble make similar to ALF (tv-series in the 1980s).

What is productivity?

An easy way to explain productivity is if we are comparing a pre-industrial society with an industrial society. The difference is, that in the first example all the goods are shipped with horses, while in the second the car is used. In the first use-case the humans have no access to electrical current and diesel-engines in the second case they have. That means, an industrial society has a greater productivity and the costs of the products are lower.

Now let us compare the industrial world with a robotics society. What is the difference? Some science fiction authors have imagined, that robots will increase the productivity further. According to the dream a robot is a very powerful steam engine which can replace human work completely. This kind of outlook is naive and wrong. It misinterprets the nature of human work. In an industrial society, most of the manual work is already automatized. That means, the automation level is 80% and higher. What all the humans are doing is not to produce goods but they are communicating about economy. Only very few jobs have to do with production itself, and even these jobs are mostly organized in teams, so that in reality work has always to do with talking to each other. A robot, which is by definition a non-human worker can’t be part of this discussion. He will reduce the productivity of the humans.

This hypothesis to understand is difficult. And no large scale studies are available which have researched the topic in detail. In some of the literature it is called the productivity paradoxon and means, that it is not possible to increase the automation level over a certain degree. If a company has already an assembly line in charge there is no way in optimize the production further. That means, an industrialized society provides already the maximum in productivity.

At the same time, robots are evolving into a working technology. That means, we see on the first hand each year new and better working robots and at the same time there is no need for such technology. That means, robots are invented without a need but as an extra society on top of the existing. Similar to the existing population of all household dogs and cats. They are there, they are doing something, but they have no real task in the society. This makes robots unique from inventions in the past, like the car or the diesel engine. The car and other machines have a purpose. They will make people more efficient and they are reducing the costs. In contrast, a robot is a negative machine. It takes energy and skills from the environments and doesn’t provide a useful feedback back.

A prominent example it the Unimation company. It was the first robotics company, founded by Joseph Engelberger in 1962. From an economical point of view, the company has produced dogs. That means, the Unimate robots have produced a lot of costs, but doesn’t automate anything. The vision of Engelberger was to produce robots for hospitals, and some older pictures shows these robots in reality. But they were not able to replace a single worker. Most of the time, the robots were defect or they were travelling without a purpose through the house.

The unimata robots are a typical example of spiritual robots. That means, these machines were from productivity point of view a nightmare, and their only purpose was to make a lot of trouble. Hundred of engineers and programmers have tried to bring the robots onlline and they are wasted their life with the robots. Exactly this social pattern will repeat in every robotics company. A modern example is Waymo, which is a spinoff of the Google self-driving car. We will see the same social pattern like in the Unimation example. The engineers will put all their energy into the self-driving cars and at the end they get nothing back.

Are robots useless?

At first we have to define productivity from an economical point of view. A machine make sense for a company, if the machines helps to increase the profit. This is true for mechanical automation. Without the assembly line the company will produce 100 pizzas a day, with the assembly line they can produce 100000. They reason, why machines are bought in industry is to earn more money and to reduce the costs. From that point of view, robots are useless, because they doesn’t fulfill these requirements. Robots are expensive machines, they will need a lot of programming hours and the end the company is not able to increase the number of produced pizza. Also the robots won’t help to reduce the costs.

If a company is buying a robot, this is equal to that the company is buying goldfish. That means, the animal has a price, it will look great in the office but it is doing nothing. Somebody may argue, that a goldfish and robots too have a meaning. But not from an economical point of view. To see the meaning the definition has to be changed. That means, if somebody likes robots he won’t argue with costs but he would say, that a robot is equal to a vision of the future. And perhaps the robot will help him to become more productive. But this explanation is not quite accurate. Because the term useful has a clear economy driven meaning. The correct definition of a hamster, a household dog or a painting of van gogh is that it is all useless from a productivity point of view.

Let us take a look back to Joseph Engelberger and ask him something about robotics. What can we learn from his life or his company? The perception under AI researcher is, that Engelberger was with his robots ahead of his time. He invented the first robots even before the first microcomputer was available. So he failed of technical reasons, right? That is not accurate. It is true, that in the 1970s and 80s computers were too slow to control a robot, bu if Engelberger would program today’s more modern robots he would fail again. A modern version of Engelberger is Rodney Brooks. He has founded Rethink robotics in the year 2008. In that time, the microcomputer was already invented and high performance Raspberry PI computer was sold at the market. Brooks started his company in a time in which computer hardware and software was highly developed. But his company failed of the same reason like Engelberger failed. The problem is, that more advanced robots are not able to provide more sense or a higher productivity. What i want to tell is, that the Unimate robot is useless and the Baxter robot as well. They are not able to replace human workers or increase the productivity in a company. Even worse, they are increasing the costs and lowering the productivity.

To understand Engelberger and Brooks we must see robots as spiritual machines. That are inventions which have never a positive effect. They are equal to a black hole which absorbs money, human intellect and manhours and doesn’t provide something back. And we can go a step further and predict that future robotics companies will run into the same problems. Even the latest generation of Boston Dynamics walking robots are useless. These advanced machines are more powerful than the Baxter robot and the Unimate robot. The Boston Dynamics robots can walk, jump and even make a backflip. But they won’t provide anything back to society. They are absorbing lots of energy and can’t be used in a productive meaning.

Increasing the productivity?

Is it possible to increase the productivity of today’s robot? Is it possible to use them in industry, for rescue missions or in the household? Unfortunately the answer is no. If the aim is to increase the productivity in an economical meaning a look back into the technology of the 19th century make sense. Most of the machines invented before the advent of computers can be used to increase the productivity. Positive examples are the refrigerator, the car, the telephone, the electric light, the tin can and the vacuum cleaner. The common feature of these machines is the absence of computerized control. They are working without microelectronics and they are driven by manual buttons for turning them on or off.

It is not possible to increase the productivity further above the automation level of the 19th century. If a robot is used with that aim, the productivity curve will be lower. This effect can’t be overcome by better programming or newly developed algorithm, it is normal for all robots.

Understanding robots

If we want to understand what robots are, we must built real robots. That are robots designed as a useless machine. The reason is, that all robots are useless so it make sense to invent from the beginning a machine which has no meaning. On the internet some examples of useless machines and especially of useless robots are given. These machines are the real robots. That means, they were constructed without wrong expectations. The robot is providing exactly this behavior what was expected from the begining. That means, the system is called useless and it will behave so.

Limits of automation

Instead of focus on robotics technology itself, it make sense to observe everything apart from robots which is used for automation. Let us take a look back into the time before the first computers were used widespread. The period of the short timetravel is the 1950’s and the 1960’s. That means, much of modern technology was already invented except of the microcomputer revolution of the 1980s. What can be automated with technology from the 1960’s? Surprisingly very much. Harvesters are used in the agriculture, the industry is working with mechanical machines, the products were transported into the city with trucks and the homes are equipped with refrigerators and toasters. That means, 80% of the manual work was automated in the 1960s. Only minor part like driving with the car into the supermarket were not automated.

The question is, what exactly was the job of the humans in that time if everything was automated? Their task was similar to what humans are doing today. They are going to school, They have learn a lot about business, science and politics and they are meeting each other in the company to present new strategies. The term automation describes a situation in which machines are doing the manual work. For example to harvest grain or to package food. The humans are doing the remaining works which can’t be automated. For example driving the truck, teaching the students and discussing problems with employees.

And here we find the problem in automation. If a society has already reached an automation level of 80% and most of the manual work was automated, the potential for future automation is low. That means, even if one engineer makes the invention of the century, he won’t revolutionize production again. Let me give an example. A modern truck has a capacity of 40000 pounds, A state-of-the art harvester has a rotor length of 3 meters, and a coal driven powerstation can produce 100 megawatt per average. Until which level this amount can be increased? Right, the limit is reached. And even advanced robots can’t improve the situation.

Most people have the hope, that robots will help them personal and all the world in becoming more productive. They dream of a society in which robots and not humans are going to work. The truth is, that the reality is the opposite. That means, it is possible to program robots in a way, that they are completely useless, while the humans become more productive. Let us make it more realistic.

The society of the future contains of robots who own humans as their workers. That means, the advanced robot likes to go shopping and he is asking a human if he can drive him to the city. The human controls the car and gets paid for the job, while the robot is sitting in the back and reads the newspaper. This sounds funny, but it will become the future. The reason for this kind of roleplay is that humans have a demand for helping other. That means, they are inventing robots only for the purpose, to become the servant of the new species. And if the humans are not looking advanced enough and have no demand for going shopping they will get programmed in such a direction. The robots from Joseph Engelberger were build for low costs, there demand for energy and software was low. The new cars of Waymo will produce more costs and need more programmers. Robots in 10 years will become so advanced, that they will have demand for thousands of AI experts and they will have wishes like going shopping and going to the doctor.

Replacing humans

Are robots able to replace factory workers, painters, musicians, or supermarket employees? The answer is always no. Robots can’t do it. The funny information is, that even the robot is more intelligent than a human he won’t replace him. Robots are additional entities in the society. They are similar to superintelligent house dogs. They have no purpose, instead they are producing costs. What robots are doing is not provide work, they have a demand for work. That means, it is possible that a human works for a robot. For example, a human can compose a song and play it loud to entertain the robot. A human can deliver a pizza to the robot, because he is hungry. That is the social role robots can play.

I know it sounds a bit wired, but the natural role of robots is to be a useless machine. And per definition such a technology is not able to increase the productivity but it will do the opposite. Future society will organized in a way, that robots can benefit from it. The question is how humans are able to help robots. Humans will work for robots, they will pay taxes for robots, they will repair robots, they will entertain them and humans will obey to the robots. Basically, robots will become the social role of gods on earth. The funny fact is, that this role isn’t intended by anybody but it is the natural choice.

Let us make a thought experiment. A single person and not the overall society builds a new robot from scratch. For example, a simple Lego Mindstorms line follower with a simple software. So in theory, this robot can be programmed in any direction and because it is a single engineer he can do with the machine what he wants. Which social role will the robot become in the one-to-one relationship between the robot and the human? It’s natural role is become a king or a god. That means, the EV3 brick is superior to the human, and the human will do everything to make the life of the robot better.

Perhaps the human will try to adjust the relationship into the other direction. He would invent an application for the Lego brick in which the robot is working for the human and not other way around. Is this rolemodel possible? Unfortunately not. The Mindstorms robot can’t be used for productive tasks, and even if the robot is bigger for example a full blown kuka robot nothing will change. That means, any attempt to utilize the machine for a useful purpose will fail. The human is not able to profit from the robot.

This kind of interaction will take place for the whole society. The single engineer will fail to change the relationship to the robots, and the overall society too. At the end all the humans are working for the robots and even if they are trying they are not able to make it differently. Let us give another example. Since many years the AI community is trying to build medical robots. The idea is, that a robot will replace a human doctor and this will reduce the costs. This effort wasn’t successful until now. And it won’t be able to build robot doctors in the next 100 years. Instead something remarkable is taken place. During the last years, so called robot psychiatrists are available. That are humans who are playing doctor for a robot. If a robot owner thinks that his robot needs a psychiatrist, he can visit such a person. This funny fact is, that such a relation ship works very well. That means, there is a demand for psychiatrists for robots and the prediction is, that in the future more robots needs mental help.

That means, from an engineering perspective it is possible to build robots who asks for a service, but it is impossible to build robots who are providing a service.

Underactuated passive robotics is a dead end

On the first look, so called passive robotics is a smart technology which is future compatible. The idea is to save energy and use external system properties. Typical examples for passive systems are flywheels builtin a bike, air driven robot hands, passive walking machines without any motor and tendon-driven robot arms.

The sad news is, that the absolute amount of saved energy is low. Flywheels were tested in car manufactoring a long time ago. They have only a little advantage but the make the car heavy weight. The same is true for cable-driven robot. They are funny to see because the concept is different but they are not able to save large amount of energy. I would guess in reality the amount of saved energy is around 5% or smaller. That means, non-passive active servo-motors can’t be replaced by passive systems. The same problem is there for sailboats used on the occean. In theory it sounds great to move ahead only with the wind, but in reality all cargo ships are driven with classical fuel driven motors and nothing will change in the next 40 years.

The comparison between passive and active motors were tested out in the past and the active motors have won. That means, they are superior to pure passive concepts. Passive dynamics is a concept which doesn’t work. It is some kind of fantasy concept which is only working right in a fairy tale. Here the machines are not driven by electricty but with thinking. That means, the actor has a wish and this will move the ship. Such kind of dream-oriented science doesn’t work. What real machines are using is pure energy, measured in Watt. That means if somebody want’s to build a robotarm or an electric car he will need a servo motor which has 2000 Watt and more. It is not possible to engineer a car the other way around, which means to ignore the motor and built first the chassis. The result will be a car without a motor, which is equal to a standstil car. That is not the car that is a failed project.

Let us make a simple calculation. The aim is to transport a cargo which has weight of 1 ton over the distance of 100 km. What will we need? We don’t need a car, a ship or a railway what we really need is energy. Without a certain amount of kilowatt the cargo won’t move a single millimetre. And a motor is the only technology available which can provide energy.

Sure, the wind in the air can provide energy too. But there are two problems. At first the amount of wind is small and secondly the direction of the wind is wrong. That means if we need 100 kilowatt energy but the wind only provides 1 watt the cargo can not be transported. I think, that dreams are important but in the domain of physical calculations. Physics has to do with numbers, and it’s important to do the math right.

Let us take at look at the facts about ships. The Emma Mærsk containership https://en.wikipedia.org/wiki/Emma_M%C3%A6rsk has a diesel engine which provides 81000 kw. Or to be more specific, the engine needs 81000 kw which is a lot of fuel. Is it possible to reduce the amount of energy? No it is not. The ship is using one of the most energy efficient motors ever built. No other transport vehicle including a sailboat can do the job better. The reason why such a large scale engine make sense is because only a small amount of employees are needed on the ship. That means a crew of only 13 people are enough to transport large amounts of containers over the ocean. If somebody want’s to cut the costs or make ships more efficient he has to focus on the manpower. That means, if it’s possible to reduce the crew to the half, the Emma Mærsk will cruise more efficient. In contrast, it is not possible to cut down the diesel engine and use a smaller one which provides only the half of power. Because then the ship would drive slower or the amount of cargo has to be reduced.

The weakest point in modern machinery is not the physics, which is equal to the motor, but it is the human power, which means the number of people needed to control the engines.

7.3 How much energy is needed by a robot?

The robotics community is discussing about different kind of realizing robots but what they are ignoring is the energy consumption. For example there are many robot competitions available like Robocup, micromouse or FLL, but what is not discussed is how much watt a certain robot needs or provides. This aspect is important if the aim is not to build demonstration robots but real robots used for practical purpose. The question is not, if a walking robot or a wheeled robot is the better idea, the question is if a robot needs 1 Watt, 100 Watt or 10000 Watt.

Let us take a look at industrial robots used in reality. Are the manufacturing companies wrong if they are using servo motors with 2800 Watt and more? No they don’t. It is normal for industrial applications like a construction crane or an electric car to use large scale motors. Large scale means, that the minimum requirement is 1000 Watt and will grow to 10000 Watt or more. The surprising information is, that this kind of attitude doesn’t depend on a certain manufactorer, but is common over the industry in general. That means, all construction cranes in the world need a motor with 50 kw, and all pick&place robots in the food industry need 15000 Watt for each robot.

But why is the amount so high? Wouldn’t it be great to build a robot for the assembly line which needs only 10 Watt comparable to an iphone 5? The problem is called physics. If the robot should uplift a donut and put it into the box, a certain amount of newton is needed. This amount can’t be ignored or calculated away. No matter if it’s an delta robot, a robotarm or a linear robot they will need all the same amount of energy. This results into a servo motor which has a dimension of 4000 Watt each, and if the robot contains of 3 motors the maximum possible energy is 12000 Watt.

For the newbies this sounds crazy. The good news is, that in reality, not the full power is used. Instead a normal pick&place action will need only a fraction of the overall power. It’s similar to a desktop PC which has a power supply for 300 Watt but will consume only 100 Watt. Additionally, the robot needs only electricity if he is moving. A standstill servo needs less energy.

It was a bit complicated to get exact information, but a typical industrial robot at the assembly needs around 800 Watt in 24/7 per average. Such a robot is able to pick&place small objects, like froozen fish or Cheesecake items. The built in servo motors have a higher rated capacity, which is 8000 Watt.

And now comes the important step. It is important to accept this kind of constraints as fixed. That means, it is not possible to build robot which needs less energy. What robot companies can do is to reduce the consumed electricity by a small fraction, perhaps by 5%, but not more. Even the most advanced robot available today will have three servo motor with 3000 Watt each. The second insight is, that a servo motor in such a dimension can’t be build small. If somebody is trying to build such a motor from scratch, the dimension will always be large.

What i want to explain is, that serious robotics has to do with large scale servo motors with 3×3000 Watt. The system will need a lot of space and is only able to pick&place small objects with a diameter of 10 cm. That is constraints which can’t be discussed away. Any robotics projects which is using smaller motors with less electricity is not a real robot but something else. That means a robot which takes only 100 Watt can be from a physical standpoint not doing any useful things.

What the robotics and AI Community is doing is to make classical large scale machinery more efficient. That means they can write software which has to be run on the robots. The funny thing is, that a robot which takes 800 Watt in 24/7 mode can’t be called costly or overpriced. Compared to other costs in a factory for example for the building or for the light, the robot is a cheap device. What makes modern robotics so expensive is not the large scale servo motor, but the labor costs for programming the machine. This is really expensive.

Let us make a short calculation. Suppose there is a standard industrial robot available at the assembly line. The delta-robot has 3 motors with 4000 Watt each and needs 2000 Watt in normal operation over a longer timespan. The costs for electricity will become 288 US$ per month (0.29 US$/kwh*2 kw*24 hourse*30 days). For a large scale factory which is producing lots of foods this amount of costs is very low. Because the factor is generating profit with the robot.

Let us describe what will happen if we are shrinking the servo motor to a different size. Somebody may argue, that a small robot can do the same. This model will consume only 50 Watt in total and helps to save a lot of money. Right, the energy costs will become slower. But which kind of work can such a machine do? That is exactly the problem. There is no application for a robot which needs 50 Watt. Every washing machine, every construction crane and every servo motor in an assembly line needs more than 50 Watt. In an industrial context the aim is to increase the power consumption. That means, if the robot needs more energy he can produce more.

At the end i want to discuss in which situation the energy costs can be saved. It make sense to separate between software development which can be done in simulation and productive robot. Testing out a robot prototype works fine on a standard notebook which needs 10 Watt of energy. It makes no sense, to use 2000 Watt robot as a programming device for playing around with algorithms. If the software works great in the simulator it can be used on the real machine for grasping real objects. This will produce a lot of energy costs but the factory has no problem with it.

Let us take a look at typical classical machines, like a washing machine, a toaster or a crane. A washing machine for home users needs around 2000 Watt. Industrial washing machines needing more. A coffee machine needs around 1500 Watt. And the models in a restaurant will need 4000 Watt. It seems, that in classical machinery it is not possible to reduce the energy very much. New models have the same specs like the older systems, so it seems there is a general rule in the form “a machine used for practical applications will need 2000 Watt or more”.

Robots are not available in the mainstream, but the prediction is that they will have a similar characteristics. How can a robot be designed if the servo motors will take 2000 Watt each? Right, this is a problem. Large scale energy consumption is equal to large motors and they can’t be built into small robot hands. The main challenge which has to be answered by robotics company is how they can use 2000 Watt servo motor for precise micromanipulation of objects.

So far, they have invented two design ideas. The first are delta robots. Delta robots will take the same amount of energy than classical robotarm, which is 4000 Watt. But they can be built more compact. The other invention is called tendon driven robot. Here has the servo motor also the same dimension, but the force is transferred with cables. Let us describe a tendon robot for industrial application in detail. The first constraint is, that a huge motor is needed which produces 4000 Watt and weights 10 kg. A cable is mounted at the motor, and the end of the cable can provide the force to a point far away from the motor. This is the basic idea. The motor is hidden in the basement and only the tip of the cable is used for manipulation.

On the first look this design looks unusual. Somebody may ask why it’s not possible to use tiny servo motors which are mounted in the robotarm and in the robotfinger directly. The problem is, that these tiny motors will have a much lower energy consumption. A typical arduino servo has around 5 Watt. Using such motors in a robot-hand design prototype is possible The hand is able to open and close and perhaps it can grasp an object. But, such a machine is not able to do usefull work. Because we have shown before, that a typical industrial machine will need 2000 Watt and more. That means, a 5 Watt servo motor can’t compete with a 2000 Watt tendon-driven robot.

And now comes question: If the servo motor needs 2000 Watt minimum, what is the appropriate robot-design? The idea is, that form follows function. The technician determines that 2000 Watt are needed and then the design department can make suggestion how the robot will look like. The problem with most robotics project is, that they are going in the other direction. They are producing first the design concept and then they are surprised, because their robot is useless. The ability to use a robot for practical purpose is the result of large scale motors. That means, a servo motor which consumes 100 Watt is useless, while a motor with 4000 Watt make sense.

Tiny robots

Let us describe a scenario in which a 5 Watt arduino robot motor make sense. Suppose, we are taking 3 motors and are building a delta robot with it. What is important here is, the dimension of the robot. That means, such a system is not able to handle objects with 10 cm lenght, but it’s possible to handle very tiny objects. For example a rice grain. A 5 watt motor has enough power to pick&place such miniature objects.

Perhaps some numbers are usefull. An industrial robot needs 5000 Watt. If the smaller version has only 5 Watt this is equal to thousands times smaller. An industrial robot can handle objects with a sitze of 5 cm maximum. So the smaller robot can handle objects with 0.05 millimeter.

Delta robots plus soft-grippers can automate supermarkets

Artificial Intelligence is usually treated as a software problem, which has to be solved by programmers. But hardware manufacturing companies have done also a good job in tackling large scale issues. The newest invention which is sold for practical application is a delta robot. Delta robots are the perfect choice for pick&place operation. Because their large scale servo motor which consumes 8000 Watt is hidden in the frame, far away from the object itself. A metal linkage connects both entities.

A delta robot itself is not able to grasp an object. A delta robot provides only the platform to reach a point in the 3d space. What is needed else is a gripper. The latest and highly innovative idea so far is a soft gripper. It is a plastic hand, filled with air which can grasp and release an object. Such a soft-gripper works well together with food. It can handle a single apple, a joghurt, bread or a fish dose. The technical working of the air gripper is realized by a remote compressor. A flexible tube is used to transfer the air from the motor to the gripper. Such a compressor motor will take similar to the delta robot itself a lot of energy. Around 1000 Watt is the average, similar to what a vacuum cleaner is needed.

Let us observe the overall system in action. The robot contains of two parts. Large scale motors hidden in the backend which are consuming 8000 Watt for moving the metal linkage plus 1000 Watt for the air compressor. If the robot is activated, it will shown at the power consumption display. And after running it for a month in 24/7 the bill will become much higher than usual. On the other side, the robot is surprisingly small. The object is taken by a small device which is a flexible gripper. This end effector can move with high speed and great accuracy.

So what can the customer do with such a robot? He can automate a supermarket. The objects at the assembly line are taken by the robot and put into the box. This can be done faster than a human can do. The robot will automate repetitive tasks and make things easier.

Why are delta-robots and continuum robots important?

The first impression might be, that both robotics type like cool and it would be nice to build such a machine. But it is not directly a design question it has to do with the combination of limited space plus large servo motors. Let us go into the details.

The constraint under which industrial robots are working is, that the servo motor has 3000 Watt each. Three servo motors have together 9000 Watt. Not the full power is used in 24/7 because sometimes the robot is in standby mode, and if the object is small he won’t need the full power. Per average a standard industrial robot needs around 1000 Watt, that is similar to what a washing machine or a microwave is using. For robotics beginners this seems to be a lots of electrical power and they maybe ask who to reduce the power consumption. That is not possible. Developing a motor which is able to drive a washing machine but will need only half of the watt is task for mechanical engineers not for robotics builder. A robotics builder has to take standard components which are available today and think about how to use this for a pick&place device.

The main question is what a robot project has to answer is how to put 9000 Watt into a small robot hand which can make small grasping movements and move from place A to B. The answer to the problem was given in the headline of this blogpost. Delta robots and tendon driven robots are using a linkage mechanism which is either a metal rod or a flexible snake like system. The main feature is, that at the end of the robot hand, the full 9000 Watt of the motor power is available. This power is used to pick and object, move it through the air and release it at another point. The action can be repeated with highspeed and for 24/7.