Service robots that learn from you and like you

Although the industrial sector has been the main user of robots for many years, nowadays there is a clear shift towards the service sector. This expansion of service robots responds to a strong demand of an ageing society (in 40 years from now, nearly 35 per cent of the population of Europe is projected to be 60 years or over), in which the urban centres continue to grow in size and density. The growth of service robots has not progressed due to technological limitations that make their progress difficult. The lack of growth in no way reflects a lack of demand. This view of a growing market is shared by all national and international robotic organizations (such as the Spanish Committee of Automation, the European Robotics Technology Platform, the Spanish Robotics Technology Platform and the International Federation of Robotics, etc).

However, if we consider the main scenarios where future robots are expected to move, or the tasks they are expected to carry out (assisting with the housework, security and vigilance, rehabilitation, collaborating in the care-entertainment, etc), we immediately realize that this new generation of robots must be able to learn on their own. They can not rely on an expert programmer, on the contrary, once they are bought they should be “trainable”. Nevertheless, this learning or robot-adaptation can not only consist on a demonstration process, in which the user shows the robot what to do. The limited nature of the human patience, the ambiguous nature of the information provided by the robot owners, the advanced years or impared mobility of the robot owners, deem it neccesary that not only should robots be able to learn from what the user does (“robots that learn from you”), but also from their interaction with the physical and social environment. Like humans (“robots that learn like you”), the mistakes and successes the robot makes should influence its future behaviour. Furthermore, this adaptation should not be constrained to a time interval, but on the contrary it should be continuous, i.e., during the life of the robot.


To enable robots to “learn from their owners and like their owners”, in this project we will have to deal with the following challenges:

    1. Development of algorithms that allow robot learning from experience and human observation. We need to develop algorithms that allow not only fast and continuous robot learning processes, but also knowledge acquision at anytime, or the ability of using environment identification to improve the performance of the robot.


    1. Knowledge retrieval from scene recognition and robot localization. Like people, the behaviour of the robots must change not only as a consequence of the time goes by, but also according to where the robots are. Today’s robots are unable to understand their environments, they are not aware if they are moving in a room that is similar to another one where they have been moving previously. In this project we will try to make progress in the development of artificial vision based techniques to carry out scene understanding. All the information about the environment will be used to retrieve robot controllers that have been previously learnt.


  1. Development of multimodal robot-human interfaces to support robot lerning processes. These interfaces must allow an easy and friendly communication with the robot, either using gestual guiding or the identification of probable objects in different parts of the environment, etc.