{"id":19635,"date":"2017-03-10T12:26:11","date_gmt":"2017-03-10T11:26:11","guid":{"rendered":"https:\/\/www.ferrovial.com\/blog\/en\/?p=19635"},"modified":"2025-12-12T11:27:03","modified_gmt":"2025-12-12T10:27:03","slug":"teaching-robots-to-improve-the-world","status":"publish","type":"post","link":"https:\/\/www.ferrovial.com\/blog\/en\/2017\/03\/teaching-robots-to-improve-the-world\/","title":{"rendered":"The robots that stopped asking questions and turned to improving the world"},"content":{"rendered":"

Man designs and programmes, and robots perform on the basis of such programming. Up until now, robotics, in all its complexity, has been a discipline based on the performance of a series of orders<\/strong>. This keeps the ball<\/em> in the court of developers, with robots lacking initiative and independence, though perhaps such qualities would make them more useful.<\/p>\n

But we humans intend to change this by programming robots in a way that makes them capable of self-programming<\/em> themselves<\/strong>. We\u2019ve been pursuing this idea for many decades, and it is likely that we now have the technological tools required for designing truly intelligent robots<\/strong>.<\/p>\n

Teaching robots to learn<\/h2>\n

[inlinetweet prefix=”” tweeter=”” suffix=””]Teaching robots to learn is the great challenge of our century. [\/inlinetweet]We realised already a long time ago that it is not possible to teach a robot everything<\/strong> in order to let it loose<\/em> on the world, in the same way that we do not teach our children everything at once. There are simply no beings<\/em> that have all the information required to survive or to work right from the start.<\/p>\n

We don\u2019t have it either. Or does a person who starts working in a new office know all there is to know from day one? Why have we been insisting on the opposite for centuries? Probably because we thought that robots would be different from us. Robotics<\/strong>, however, is now<\/strong> progressing towards learning<\/strong>.<\/p>\n

 <\/p>\n

\"Retro<\/a><\/p>\n

For example, think of the latest trend, the driverless cars which are already being used on streets around the world<\/a>. It would be impossible to teach these types of robots to distinguish each and every one of the points of the earth and drive from memory<\/em>.<\/p>\n

Instead, they are taught to drive on the basis of traffic regulations and learning from error. Exactly in the same way as we would teach a human being. With the advantage, of course, that they are not human.<\/p>\n

Genetic algorithms and group learning<\/h3>\n

Autonomous vehicles are not the only application of robots that learn on their own. One of the key barriers to the integration of robotics in such diverse fields as industrial production, construction, or the care of the elderly, is how complicated it is for robots to recognise their environment and move about without destroying everything in their path.<\/p>\n

And here\u2019s where the generational learning<\/strong>, \u2013 or\u00a0 \u00a0robotic genes<\/a> \u2013 comes into play. The basic idea can be seen in this video: a simple virtual robot learning to walk. After setting it the objective of \u2018go far using your \u201cmuscles\u201d\u2019, it is left to progress and learn on its own. In the beginning, it trips and falls, as would a baby.<\/p>\n