PR2 - learning companion robot

Explore the world and combine - those two pillars on which human civilization rests. Well, also greed, of course. Designer cups alone in the 21st century came up with more than great inventions over the past century. Although it would seem that it could be easier than a faceted glass of a fly and a glass holder.

However, if for us the process of mastering the objective world (and of our own body) was relatively easy in the distant, remote childhood, then for robots to learn how to cope with numerous beer mugs is an impossible task that is usually passed on to adult bearded programmers.

Therefore, the creation of deep learning algorithms remains the main direction in the development of robotics for the seventeenth consecutive year. And where else to test such algorithms, if not in the field with a variety of different rattles: screws, bolts, keyholes and buttons.
')
The most capable child prodigy today is the brainchild of Willow Garage, under the name PR2 (Personal Robot 2) . The guys didn’t grow together with the first model, but the heirs are already actively working in hotels .



There are several advantages to PR2 . This, of course, quality manipulators is akin to YUMI , good sensors and, most importantly, open source code. Seven nannies usually have a child without an eye, but this time in the role of nurses they are specialists from the best universities in the world, so the training moves by leaps and bounds.

At the head of PR2, there are a lot of fresh libraries, hardware drivers, and blanks that help scientists do something new, and not re-teach the robot to walk. PR2 already knows how to open a beer, go for coffee (doors, elevator, bring it up), play billiards and — most importantly — hold objects. Such successes, by the way, became possible largely due to tactile feedback, so the robot does not flatten a jar of hot tea in front of you, but will carefully place it in front of you on the table.



However, all this was back in 2013. Literally in a year and a half, the algorithms have grown to the point that now PR2 can make you coffee. Attention: make coffee on an unknown coffee machine, guided by the instructions printed on the piece of paper. For example: “Hold an espresso cup under a hot water tap.”



The idea is that the robot independently navigates in a room full of objects, recognizes the necessary buttons and knobs in a form without prompts and a pre-compiled knowledge base. All that he has in his hands is a piece of paper with instructions and objects.



Applause to developers from Cornwall!

One more thing just a few days ago the published achievement in the education of robots by the deep learning algorithm “by trial and error”. This time, the program was written by scientists from the Berkeley Research Center.



In this case, the developers managed to reproduce the learning process, which is as close as possible to the human one. Naturally, little PR2 still needs to be learned, but the scientists have outlined the right way.

Now, to set the task to the developer, it takes about 10 minutes of time. To complete the task, the car takes about 3 hours. This, of course, is a bit too much (if we consider that it is painful to attack the rake in the real world), however, this algorithm is only the beginning and scientists hope that the robot will start to develop exponentially.