iCub Philosophy, Some History & Recent Results – video lecture by Prof. Giorgio Metta
The iCub project was started 10 years ago within the field of Human Robotics, focused mostly on building models of cognitive behaviours. The goals of the project include understanding how the human brain resolves certain problems just to use this knowledge within robotics and build models that can allow robots to solve autonomously the same problems, hence become highly intelligent systems.
20 years ago neurophysiologists discovered special types of neurons called visuomotor neurons. This discovery triggered a shift in the way robots were programmed to learn and perceive as now, mirroring the function of these neurons, the goal was not only for the robot to perceive the surroundings but also to act upon the new information acquired, to carry out a task, solve a problem, imitate what a person would do when he or she perceives and processes external information.
To translate these goals into practice, first experiments where focused on registering human grasping actions via data gloves and trackers and then these registered actions, translated into images, were used to build a classifier which would aid the robot to predict better what action to carry out when perceiving certain information. Furthermore, neuro studies showing that it is not that important to recognise the object to be grasped it but it’s enough to recognise it’s shape, helped Prof. Metta and his team build models that prepared the robot to recognise the shape of an object just before executing an action like grasping it. This made possible for the robot to execute more complex behaviours such as recognise not only the objects to be grasped or pushed, but also to recognise tools which may help the robot to do something with the object in question. In one of the experiments lead by Prof. Metta, this was translated into an iCub grasping a tool to bring closer an object placed too far away from the robot, in order for the iCub to grasp it.
Read More: RoboHub.org
Source: International Journal of Advanced Robotic Systems (IJARS)
– Future Robots
– Artificial Intelligence
– Robotics Books
– A.I. Books
– Robots for the Home
– Neuroscience Books
Donate ETH: 0x981FcEAAa895C6cee76D2876e9AfC649Dc0C4c75