Gal A. Kaminka: Publications

Sorted by DateClassified by Publication TypeClassified by TopicGrouped by Student (current)Grouped by Former Students

Robots Predictive Execution Monitoring in BDI Recipes

Mika Barkan and Gal A. Kaminka. Robots Predictive Execution Monitoring in BDI Recipes. In Proceedings of the 2019 AAMAS Workshop on Autonomous Robots and Multirobot Systems (ARMS), 2019.

Download

[PDF]578.0kB  

Abstract

Execution monitoring allows robots to assess the execution of plans, determine the need for re-planning, identify opportunities, and re-evaluate their commitments. There exists extensive literature on monitoring the execution of classical and HTN plans. However, execution monitoring of BDI plans is often left implicit in the BDI control loop. In practice, many BDI plan execution systems monitor the current plan steps only. They do not project ahead the current knowledge of the robot to determine implications on future steps. Thus a failure of a future plan-step, which may already be predictable given the current knowledge of the robot, is not detected until the last possible moment. This paper examines the task of predictive execution monitoring in BDI plans. It provides a base algorithm, and shows that its complexity is super-exponential in the general case, even under mild assumptions. It then discusses several methods for pruning the search space, and formally shows their completeness. It evaluates these methods in hundreds of experiments, utilizing approximately 4000 hours of modern CPU time.

BibTeX

@InProceedings{arms19ws,
	author = {Mika Barkan and Gal A. Kaminka},
	title =  {Robots Predictive Execution Monitoring in BDI Recipes},
	year = {2019},
	booktitle = {Proceedings of the 2019 {AAMAS} Workshop on Autonomous Robots and Multirobot Systems ({ARMS})},
	abstract = {Execution monitoring allows robots to assess the execution 
	of plans, determine the need for re-planning, identify opportunities, and 
	re-evaluate their commitments. There exists extensive literature on monitoring 
	the execution of classical and HTN plans. However, execution 
	monitoring of BDI plans is often left implicit in the BDI control loop. In 
	practice, many BDI plan execution systems monitor the current plan 
	steps only. They do not project ahead the current knowledge of the 
	robot to determine implications on future steps. Thus a failure of a 
	future plan-step, which may already be predictable given the current 
	knowledge of the robot, is not detected until the last possible moment. 
	This paper examines the task of predictive execution monitoring in BDI 
	plans. It provides a base algorithm, and shows that its complexity is 
	super-exponential in the general case, even under mild assumptions. It 
	then discusses several methods for pruning the search space, and formally 
	shows their completeness. It evaluates these methods in hundreds 
	of experiments, utilizing approximately 4000 hours of modern CPU time. 
	},
}

Generated by bib2html.pl (written by Patrick Riley ) on Mon Nov 16, 2020 22:25:46