Reasoning About When to Start Acting

Facedwith a complicatedtask, someinitial planningcan significantlyincreasethe likelihoodof successandincrease efficiency, butplanningfor toolongbeforestarting to act canreduceefficiency. Thispaperexploresthe questionof whento beginacting for a resourceboundedagent. Limitations of an idealizedalgorithmsuggestedin the literature are presentedandillustrated in the contextof a robotcourier.A revised, idealizedalgorithmis givenandjustified. Therevisedidealizedalgorithmis usedas a basisfor developinga new"step choice"algorithmfor makingon-the-flydecisions for a simplifiedversionof the robotcouriertask. Aset of experimentsare usedto illustrate the relative advantageof the newstrategy overalwaysact, alwayscomputeandanytimealgorithmbasedstrategies for decidingwhento begin execution.

• Type: Conference
• Year: 1994
• Where: AIPS
• Authors: Richard Goodwin
• Keywords: AIPS 1994, Artificial Intelligence, Decidingwhento Begin Execution, Significantlyincreasethe Likelihoodof Successandincrease