Knowledge Stream: Where Will Robots Learn?

May 31, 2013, 20:00 lecture


On May 31 the Digital October Centers’ educational project Knowledge Stream held a lecture via web link by Ken Goldberg, American engineer, inventor, and professor at the UC Berkeley.

His lecture, entitled “Where will Robots Learn?”, opened the new “Forecast for Tomorrow” lecture cycle focused on the technology of the near future and developed jointly with IBS.

“The idea to create robots itself has been around for a very long time, just about since ancient Greece. And think about the fantastic mechanisms of the 19th century,” began Ken. “But the big break-through in the field is happening before our eyes, all thanks to the internet as we know it today.

“Just a decade and a half ago, computing power skyrocketed, with new sensors appearing and the cost of components dropping. Robots quickly became a part of our life — remember the millions of robot vacuums, the bomb-sniffing robots we saw in the news… Of course, more than anything I like that there are already almost three thousand robotic surgeons in the world.

“Sure, they can do some things, but, to quote a friend of mine,


a robot will only ever be considered autonomous if you send it to work and it goes to the beach.


“And if we want to see smart machines in robots it’s worth giving some serious thought to how to teach them. For example, today’s everyday robots have a small memory and insignificant on-board computing capability, so teaching them adaptive behavior is impossible.

“But think about one scene from the Matrix, when Trinity needed to learn how to fly a helicopter. She just requested the program and—voila! Robots can do that too. In San Francisco, for example, there is already a company called RobotAppStore that is actively working in that direction.

“The internet is an enormous receptacle for data with huge computing capabilities. If robots are able to connect with some kind of cloud via wi-fi, then the problem will be solved remotely without burdening their frame with additional processors.


“For instance, ROS emerged—an open-source OS


that is already being implemented in thousands of robots around the world. In allows them to quickly exchange information, and since the code is open-source anyone can make their own contribution or addition, making it available for general use.

“Of course, the thought has already occurred to you that soon robots will be walking around your house tidying up. What’s so difficult about that, you ask? From a programming point of view that is an extremely difficult task. The number of objects the robot might come across in just one apartment is practically endless. But there is already a solution for this.

“Google has their object recognition system called Google Goggles — you just photograph an unknown object (a painting, bottle of wine, anything) and the system, on the shoulders of the long history of search terms, gives an answer and tells you what the object is. Now imagine that the robot is connected to Google Goggles. Let’s say it finds a bottle of something carbonated on your table, but doesn’t know what to do with it.


“‘Doofus,’ Google tells it, ‘that’s just a regular coke.’


“And simultaneously it sends the data about the bottle’s measurements, how to grasp it — calculated remotely — and the combination of actions and the trajectory the robot needs to take and move that object. The robot in turn sends feedback about the success of the action and, if there were multiple solutions, which one it chose.

“We have basically built our own such system for robots and even announced it two weeks ago. Of course, we’re working on the lack of data for the geometry of many objects, but we’ve already tested our solution on a small group of objects: 20% of them fell, unfortunately, but we know why that happened. Generally speaking, the problems were related to dim lighting.

“On the other hand, the statistics of successful grasping are impressive, and we think that sometime in the near future we will be able to raise the effectiveness of our system. Robots will then become really useful as helpers around the house.

“One more issue, of course, is the price. Last year I was in South Africa and found out that cloud robotics are very popular there. But their robots cost a fortune.

“After traveling to South Africa we even thought up an African Robotics Association. By the way, in spite of the name anyone can join from anywhere on the planet. Some have even joined: we set a goal for ourselves of making a robot that would cost $10 already assembled, and when we announced a competition for the best project we received 28 applications from around the world.


“An enthusiast sent us the best idea —


not a professional engineer, mind you, but a normal guy from Thailand.

“He took a gaming controller, attached wheels to it, and installed a balancer made from two lollipops. But his creation can be programmed, directed remotely, and costs all of $8.96. Right now we’re preparing that model for mass production — finalizing the design and the software, and even signed an agreement with a candy manufacturer about a sponsorship.

“Cloud technology is equally suited to large, cumbersome objects. Recently General Electric introduced a very interesting term: ‘industrial internet.’ They are referring to the fact that all their giant electromechanical equipment will be able to talk with each other, so that, let’s say, the engines on one plane could give other flights information about the weather somewhere.

“I’m really intrigued by the idea of the Internet of Things. Cars will send information themselves about failures, accidents, or the quality of the air in a given part of the city. I’m sure you’re aware of Google’s experience with robotic cars and have asked yourselves why that company is working on that type of project? Because they have access to a huge amount of information and their own maps, and they can use the capabilities afforded them by their widespread network and data processing centers.


“Another advantage of cloud robotics that I haven’t mentioned yet


is the possibility for a robot to interact with humans. At the end of the day, no robot will be a completely isolated system. I really like the idea that a robot can put a call through to the most ordinary call center to ask people to fix its ‘robot problem.’

“There is no more effective tool than a teacher. I think you’ll agree with that. And so, returning to the theme of the lecture, I’ll say this: I think that robots will learn from each other and from us. Just think, even children will be able to teach a robot something!”

Free event participation was provided by general partner of the Knowledge Stream project – Rostelecom, and RVC, intellectual partner of the lecture.


472 show Danila
Member of the coordination council of the Russian Transhumanist Movement, futurologist, host of TV program For the Future on Nauka 2.0 TV channel
587 show Evgeny
Deputy General Director, Board Member, IBS Group
588 show Anatol
Professor, Head of the Department of Control Systems and Industrial Engineering, Mines Nantes, France
619 show Arseny
Head, FGM Digital
620 show Elad
Founder and CEO, RobotAppStore, RobotsLAB
621 show Vasiliy
Vice President of Technological Strategy, Luxoft (IBS Group)
622 show Vyacheslav
Project Manager, rbot

photo gallery

all photos of event

general partner


intellectual partner of this lecture

Russian Venture Company

series partner


internet partner

Mail.Ru Group

general media partner

Ashmanov & partners


119072, Moscow, Bersenevskaya Naberezhnaya, 6, str.3

+7 (499) 963–31–10
+7 (985) 766–19–25