Change search
ReferencesLink to record
Permanent link

Direct link
Robot Learning from Demonstration: A Task-level Planning Approach
KTH, School of Computer Science and Communication (CSC), Centres, Centre for Autonomous Systems, CAS.ORCID iD: 0000-0003-2965-2953
2008 (English)Article in journal (Refereed) Published
Abstract [en]

In this paper, we deal with the problem of learning by demonstration, task level learning and planning for robotic applications that involve object manipulation. Preprogramming robots for execution of complex domestic tasks such as setting a dinner table is of little use, since the same order of subtasks may not be conceivable in the run time due to the changed state of the world. In our approach, we aim to learn the goal of the task and use a task planner to reach the goal given different initial states of the world. For some tasks, there are underlying constraints that must be fulfille, and knowing just the final goal is not sufficient. We propose two techniques for constraint identification. In the first case, the teacher can directly instruct the system about the underlying constraints. In the second case, the constraints are identified by the robot itself based on multiple observations. The constraints are then considered in the planning phase, allowing the task to be executed without violating any of them. We evaluate our work on a real robot performing pick-and-place tasks.

Place, publisher, year, edition, pages
2008. Vol. 5, no 3, 223-234 p.
Keyword [en]
Programming by demosntration, task learning, task planning, object, manipulation, motion
URN: urn:nbn:se:kth:diva-18234ISI: 000264063000001ScopusID: 2-s2.0-51349119092OAI: diva2:336280
QC 20100525Available from: 2010-08-05 Created: 2010-08-05Bibliographically approved

Open Access in DiVA

No full text


Search in DiVA

By author/editor
Kragic, Danica
By organisation
Centre for Autonomous Systems, CAS

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 44 hits
ReferencesLink to record
Permanent link

Direct link