When robots "soar in the clouds": The five components of cloud robotics

The potential of using robots has not yet been revealed, but cloud technologies will soon provide us with incredible support. In this article, I will talk about how a new generation of robots can use cloud technologies , Big Data, machine learning, open source systems and the “Internet of things” to help us more in various tasks: from driving a car and doing housework before surgery.

This trend dates back to the early 1990s , when the World Wide Web first appeared. I was a young professor at the University of Southern California and worked with graduate students in the research laboratory of robotics. Once they came to my office and showed something unimaginable: they launched the first Mosaic web browser on my desktop. We visited the first few websites, including a website where anyone could watch what was happening in the student café through a webcam. It was established by graduate students at the University of Cambridge to keep track of when you can get fresh coffee, in connection with the so-called "destructive caffeine addiction."
')
That evening, my students and I stayed up late in the laboratory, trying to develop our idea. Instead of passively watching what is happening with the help of webcams, can we allow remote visitors to actively influence our workspace using our robot?

What if the robot, instead of doing something boring like building a tower of cubes, could take care of a garden with live plants? To do this, we took the industrial robot arm and equipped it with a digital camera, an irrigation system and an air nozzle so that it could collect seeds. Then it was installed in the center of an ordinary round flower bed with aluminum fencing with a radius of three meters and filled it to a depth of half a meter with soil and several seedlings. We developed a graphical web interface that everyone could access to monitor the plants, water them and plant the seeds in this garden, controlling the robot.



"Telesad" (materials: robot, garden, Internet). 1995–2004, Ars Electronica Museum, Linz, Austria. Project leaders: Ken Goldberg and Joseph Santarromana. Project team: George Bekey, Steven Gentner, Rosemary Morris, Carl Sutter, Jeff Wiegley. Ars Electronica team: Erich Berger (Erich Berger), Gerold Hofstadler (Gerold Hofstadler).

"Telesad" [ eng. The Telegarden ] became available on the Net in the summer of 1995. Rumors about him quickly spread, and after a few weeks, thousands of people began to visit this “communal” garden. Many regularly visited the site to water their plants. In the general chat, people asked someone to water their plants while they were on vacation. Thousands of seedlings began to sprout, and Telesad quickly grew. The project eventually turned into a study of the "tragedy of the community."

We were shocked; we feared that gardening might be the least interesting thing people wanted to do on the Internet (and this is ten years before the appearance of the game FarmVille). A year later, we were offered to install “Telesad” in the Museum of Austria, where, as a result, he worked online 24 hours a day for nine years.

As far as we know, more people have used this robot than ever before.

"Telesad" became the first active device operating on the network. MIT Press published two books about it, and soon many other devices and systems were connected to the Web.



Since then, robotics has progressed far enough. Currently there are hundreds of research laboratories and more than ten journals are published; developed more than 5 million service robots like Roomba, retractable in homes and offices, and more than 3,000 robots helping surgeons in operating rooms around the world. Many major achievements have been made in the world of digital cameras, inertia and many other sensors. The invention of Microsoft's Kinect 3D camera for games was a huge breakthrough in robotics and provided a cheap way to get a three-dimensional cloud of points that can help robots navigate space and execute commands. In 2012, US President Barack Obama announced the provision of new funding in the amount of $ 70 million for the US national research program in the field of robotics.

However, robots are not yet able to fold things or load dishes in the dishwasher. This routine work routine is extremely difficult for robots. The main problem is uncertainty. Put yourself in the robot's place: everything around you is unclear and unstable, everything seems to be in low resolution and is disturbing; you cannot determine what you are facing, where it is and how it moves. You cannot fully control the movement of your own hands: as if you are wearing huge gloves and glasses smeared with petroleum jelly. I hope you now feel sympathy for the robots.

I think cloud technology is the key to a new generation of robots. Take, for example, Google's robot car. The car uses the network to access a huge Google database with maps and images from space and StreetView service, and then compares them with a stream of GPS data, cameras and 3D sensors to determine its position to the nearest centimeter and avoid collisions taking into account past and current situations on the road. This gives Google an incredible advantage over automotive companies such as Toyota and General Motors.



Prototype unmanned car google

Why is Google interested in robots? Because Google understands the essence of the Internet.

By the way, I like one of the observations of Brad Templeton:

"The robot will be truly autonomous only when you give him the command to start work, and he instead decides to go to the beach."

Despite the fact that robots have used the Internet for 20 years, in 2010, James Kaffner, a terrific researcher from Google, introduced the concept of "cloud robotics". The cloud is not a new name for the Internet. This is a new paradigm, suggesting new ways of using the Internet. Take, for example, Google Docs. Anyone could send Microsoft Word documents over the Internet, another thing is Google Docs: the document and software are not on your computer. Everything is stored in the cloud using remote server farms with combined memory and processors. This is quite convenient, since you do not have to worry about the drive failing or its operation and updating the software or hardware. In addition, the cloud allows you to save on scaling and share data between applications and users faster than ever (of course, there are also important issues of security and privacy).

Below are the five components of cloud robotics:

The first component is directly related to memory. The 2012 film “ Robot and Frank ” is a masterpiece offering a unique and, I think, quite realistic picture of the future (like Spike Jonze’s “ She ”, released in 2013). The hero of the film is aging and becoming forgetful. His children send him a robot to help him around the house. He cleans up after him, reminds him of healthy food and watering the garden. When I watched the movie, I wanted the same robot that would remind me to eat cabbage, take my medicines and shook the press, and also keep me company, recalling important events in my life, perhaps even telling jokes about recent events or about what is happening. But such household robots are not yet available. The difficulty is that in an ordinary house you can count thousands of items.



Think about creating a robot that puts things in your house into place. After all, it is very important for everyone who has children, and especially important for older people. If an elderly citizen drops something on the floor, he may not pay attention to it due to poor eyesight. But even if they see, it will not be easy to bend down and pick up the object. And a fallen object can lead to tragic consequences: a hip fracture, for example, can subsequently chain you to a bed, which in turn leads to impaired blood supply and depression. After a certain age, things left on the floor can play a cruel joke with you.

What if the robot could work silently while you were sleeping or at work, collecting things from the floor and putting them where they lay? The problem is that no matter how well the robot is programmed or how many items are stored in its internal memory: there will always be something with which he is not familiar. For example, my new remote control for switching slides: if it fell to the floor, my robot might not know what to do with it. Maybe it's a chocolate bar? Did he lie on my desk? In a refrigerator? Or in the trash can?

Fortunately, any robot that works in your home will be connected to a Wi-Fi network, so that it will have access to an extensive repository of information on the Internet, where there is data on almost any subject you can imagine. The amount of data here is huge, and it is constantly growing. The problem is that you cannot save all this memory inside your robot. But the cloud makes this information available when it is requested. Therefore, the first element of cloud robotics is big data [ English. Big Data ].

The second element of cloud robotics focuses on the limitations in processing information on embedded computers. Robots can at best contain several computers, and many problems arise that require much more computation than they can provide.

The robots are beginning to use a statistical approach known as the “space of belief”.

One striking approach is to model the environment, sensors, and actions using a probability distribution. The mathematical concept describing this process is the “space of belief” [ eng. Belief Space ]. I know what sounds like something sublime, or as if it was taken from science fiction. However, the term is a symbol of Markov decision-making processes in partially observable variants of the environment [ Eng. Partially Observed Markov Decision Processes, POMDPs ]. To solve this kind of problem and find the optimal action in a given environment, it is necessary to convolve several distributions. Soon this becomes impracticable, as the probability distributions become more complex, multimodal and non-parametric.

Finding a solution requires extremely high computing power. "The space of belief" did not respond to processing until recently, when access to clusters of computers on demand through the cloud was obtained. Among other things, such high performance facilitates statistical optimization, machine learning and programmed movement in the multidimensional space of several simultaneously moving robots. The second element of cloud robotics is cloud computing.

The third element of cloud robotics takes into account the fact that people are becoming more connected across the Web, sharing information with each other. I will give an example. I was born in Nigeria and returned there several years ago. I was amazed when I learned how much students are interested in robotics. In Africa, as in other places around the world, children see the robot as something tangible, something that can be recognized, and it catches their attention. Robots are like a drug that makes students interested in the natural sciences, technology, engineering, and mathematics [ Eng. STEM - Science, Technology, Engineering, and Math ]. I met students from different parts of Ghana who are interested in robots, but in most cases they did not know anything about each other.

Robots are like drugs that make students interested in science, technology, engineering, and math.

I met an amazing professor, Ayorkor Korseh from Ashesi University. We decided to form an African network of robotics. The idea was to link all these groups not only in Ghana, but also all over Africa - all those who are interested in robots, and give them the opportunity to learn from each other, develop new devices and develop new ideas.

Existing at the moment, training robots are quite expensive. We found the need to create an inexpensive robot for educational purposes. To do this, we created a website and announced a worldwide project competition. We set a target price of $ 10. We thought it was just ridiculous, but our project could make people think seriously. Four months later, 28 applications came to us, a lot of excellent projects, and the owner of the main prize just hit us.

The basis of this robot was Sony's gamepad. The author of the work Tom Tilly modified the game controller, which cost about 3-4 dollars. It has two built-in motors that create vibration during the game. He took them out and screwed two wheels. The game controller case has become a mobile robot case. In addition, he attached the lights, but still wanted to be able to determine the obstacle encountered.

He realized that he could use two switches at the top. The problem was that when testing the switches did not react; they needed an additional lever, that is, Tom needed a counterweight for them. He thought about what would be suitable, and what would be affordable, and a brilliant idea came to his mind: candy.



The prototype of the “candy robot” designed by Tom Tilly

What child will stand in front of a robot with two candies at the top?

Tom calls him "Candy Robot" [ eng. Lollibot ], and you can find on the Internet detailed instructions for building your own robot. The total cost of all the parts is $ 8.96 (and that includes candy)!

Another incredible achievement is the operating system for robots ROS [ eng. Robot Operating System ]. This is an open source software library that has changed the scope of robotics; so to say, Linux for robots. When someone comes up with a new algorithm, he or she can immediately download it to ROS and open it up to other scientists around the world. So, the third element of cloud robotics is an open source system that provides general access to people's ingenuity: code, data, and projects.

The fourth component of cloud robotics is based on direct communication of robots with each other. Take a look at Amazon, which needs to quickly process thousands of orders for books and other goods and pack them in boxes with items that may be in different locations in huge warehouses. A company called Kiva Systems has designed a new type of robot to solve this problem. Robots move shelves with goods in boxes, and since there may be several hundreds or thousands in stock, robots increase productivity by several times. The reason for this success is that all robots interact with each other. They maintain constant communication: all robots agree on the routes of their movement, and when conditions change, for example, if the robot detects a lubricant spilled on the floor, it instantly notifies the others to avoid it. The fourth element of cloud robotics is the mutual learning of robots - sharing data and code to increase overall performance.

The fifth element of cloud robotics concerns the error handling modes in the case when the robot is not able to determine what to do. If you work in the field of robotics, you understand that there will always be situations when the robot does not know what to do.

When a robot gets into a dead end, he can ask for help from a person

The cloud can provide access to a referral service where people can help identify a problem based on the data and video of the robot and suggest steps to solve the problem if all else fails. It seems to me that this is the opposite case when you are now calling technical support, and the voice of the robot answers you. One way to implement this idea is to have the robot track its level of confidence so that it can recognize when the situation is unclear and needs to stop to ask for help. Therefore, the fifth element of cloud robotics is crowdsourcing.

Until recently, robots were considered autonomous systems with limited computing power and memory. Cloud robotics offers a terrific alternative, thanks to which robots access and exchange data and code over a wireless network.



So, the five components of cloud robotics include:

1. Big data : streamlining the global repository of images, maps and data about objects,
2. Cloud computing : distributed computing on demand using statistical learning and route planning,
3. Open source : people sharing code, data, algorithms and hardware for robots,
4. Collective learning of robots : sharing by robots of data on the trajectory of movement, principles of control and possible outcomes of events that can be analyzed using statistical approaches to machine learning, and
5. Crowdsourcing and referral services : autonomous instructions requested by the robot, allowing people to assess the situation, train the robot and correct its mistakes.

The development of cloud-based robotics will be based on common initiatives, including the Internet of Things, IBM's Smart Planet, the idea of ​​Industrial Internet from General Electric, and the concept of Industry 4.0 from Siemens. These approaches have enormous potential, but at the same time open Pandora’s security and privacy issues.

---

“ What we love robots for ” is a five-minute documentary (the director is my wife and laureate of several Tiffany Shlein awards), which combines cultural references and the genre of “found film” [ Eng. found footage ], and also demonstrates a study of our dedication to robots and recent discoveries in the field of cloud robotics. In 2014, he was nominated for an Emmy Award and was awarded the Botskar Award (Oscar for the film about robots) at the robots movie festival.