Robot task error recovery using Petri nets learned from demonstration

Guoting Chang, Dana Kulić

Research output: Contribution to conferencePaper

5 Citations (Scopus)

Abstract

The ability to recover from errors is necessary for robots to cope with unexpected situations in a dynamic environment. Efficient error recovery should allow the robot to utilise existing knowledge of the task and learn new error recovery strategies from observation. This paper proposes an automatic error recovery procedure that allows the robot to handle both known and unknown error states using a Petri net representation of the task. For known error states, the robot can directly adjust the sequencing of actions using the Petri net representation to complete the task, while for unknown error states, the robot can learn how to perform error recovery from a human demonstrator by extending the existing Petri net. The proposed method is verified on a real robot performing a block stacking task.

Original languageEnglish
DOIs
Publication statusPublished - 1 Jan 2013
Externally publishedYes
Event2013 16th International Conference on Advanced Robotics, ICAR 2013 - Montevideo, Uruguay
Duration: 25 Nov 201329 Nov 2013

Conference

Conference2013 16th International Conference on Advanced Robotics, ICAR 2013
CountryUruguay
CityMontevideo
Period25/11/1329/11/13

Cite this