Direkt zum Inhalt springen
Computer Vision & Artificial Intelligence
TUM School of Computation, Information and Technology
Technical University of Munich

Technical University of Munich

Menu

Links

Informatik IX
Chair of Computer Vision & Artificial Intelligence

Boltzmannstrasse 3
85748 Garching info@vision.in.tum.de

Follow us on:
CVG Group DVL Group SRL Group


Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
members:sturmju:research:imitation [2011/07/12 11:18]
sturmju
members:sturmju:research:imitation [2012/01/20 12:01] (current)
Line 1: Line 1:
 +====== Imitation Learning ======
 +
 +To accomplish a particular manipulation task, a robot needs a detailed description 
 +of how to execute it. 
 +However, it is not possible to specify all potential tasks of a
 +manipulation robot beforehand.  For example, robotic assistants
 +operating in industrial contexts are frequently faced with changes in
 +the production process. As a consequence, novel manipulation skills become relevant
 +on a regular basis.  For this reason, there is a need for
 +solutions that enable normal users to quickly and intuitively teach
 +new manipulation skills to a robot. 
 +
 +
 +More concretely, we consider the problem of learning generalized descriptions of
 +object manipulation tasks from human demonstrations. 
 +We employ dynamic Bayesian
 +networks (DBN) as a compact representation where special nodes 
 +encode geometrical constraints between the relevant objects in the scene and the
 +hand of the demonstrator. This formulation allows the robot to 
 +learn generalized task descriptions from multiple demonstrations so that 
 +it can reproduce them also under changed conditions. 
 +Furthermore, novel constraints 
 +can easily be added during the reproduction of a task. This is, for example, relevant to allow a robot 
 +to deal with obstacles.  To reproduce a task,
 +the robot searches for the action sequence that maximizes
 +the data likelihood in the DBN. In experiments carried out in simulation as well as on 
 +a real manipulation robot, we show that
 +our approach enables robots to efficiently learn novel manipulation skills from human 
 +demonstrations and to robustly reproduce them in different situations. 
 +
 +<html>
 +<iframe width="425" height="349" src="http://www.youtube.com/embed/LBLvpYOnGU0?hl=en&fs=1" frameborder="0" allowfullscreen></iframe>
 +</html>
 +
 +====== Related Publications ======
 +
 +<bibtex>
 +<author>J. Sturm</author>
 +<topic>imitation-learning</topic>
 +</bibtex>
  

Rechte Seite

Informatik IX
Chair of Computer Vision & Artificial Intelligence

Boltzmannstrasse 3
85748 Garching info@vision.in.tum.de

Follow us on:
CVG Group DVL Group SRL Group