Developed by Lars Kopp 1998.

The aim of this research is to build a robot that is able to grasp objects on a table.

To grasp an object the robot needs to know:

  • What the object is
  • Where it is located
  • Its orientation

Together, these properties tells the robot how to grasp the object. It was suggesten by Goodale and Milner that such a HOW-system can be found in the parietal lobe in the human brain. According to Zeki, this system is separate from the what and where streams described by Mishkin and Ungerleiter. It is the goal of this research to use ideas from biology and apply them to a new concept of robot construction.

The main features of this work is:

  1. Learning new object by an exploratory behaviour. The robot will learn how to grasp new objects with very little help from a supervisor. The robot will play with novel objects and learn how an object looks from different views and how to grasp it correctly.
  2. Object tracking and grasping objects in flight.
  3. Attending to appropriate objects in a complex scene.

The HOW-system is developed from the XT-1 architecture (Balkenius and Kopp, 1996). This architecture was developed for a moving stereo-robot-head that could track a moving object. It was also able to automatically select an appropriate goal-object.

In the current project, a more complex form-analyzing system has been added. The form-system learns objects from different views and associate the learned views with motor-action commands for the robot. Thus, the robot learns to associate a certain stimulus with an action.