Heuristic search is a technique used pervasively in artificial intelligence and automated planning. Often an agent is given a task that it would like to solve as quickly as possible. It must allocate its time between planning the actions to achieve the task and actually executing them. We call this problem planning under time pressure. Most popular heuristic search algorithms are ill-suited for this setting, as they either search a lot to find short plans or search a little and find long plans. The thesis of this dissertation is: when under time pressure, an automated agent should explicitly attempt to minimize the sum of planning and execution times, not just one or just the other.
This dissertation makes four contributions. First we present new algorithms that use modern multi-core CPUs to decrease planning time without increasing execution. Second, we introduce a new model for predicting the performance of iterative-deepening search. The model is as accurate as previous offline techniques when using less training data, but can also be used online to reduce the overhead of iterative-deepening search, resulting in faster planning. Third we show offline planning algorithms that directly attempt to minimize the sum of planning and execution times. And, fourth we consider algorithms that plan online in parallel with execution. Both offline and online algorithms account for a user-specified preference between search and execution, and can greatly outperform the standard utility-oblivious techniques. By addressing the problem of planning under time pressure, these contributions demonstrate that heuristic search is no longer restricted to optimizing solution cost, obviating the need to choose between slow search times and expensive solutions.
","abstract_html":"<p>Heuristic search is a technique used pervasively in artificial intelligence and automated planning. Often an agent is given a task that it would like to solve as quickly as possible. It must allocate its time between planning the actions to achieve the task and actually executing them. We call this problem planning under time pressure. Most popular heuristic search algorithms are ill-suited for this setting, as they either search a lot to find short plans or search a little and find long plans. The thesis of this dissertation is: when under time pressure, an automated agent should explicitly attempt to minimize the sum of planning and execution times, not just one or just the other.</p><p>This dissertation makes four contributions. First we present new algorithms that use modern multi-core CPUs to decrease planning time without increasing execution. Second, we introduce a new model for predicting the performance of iterative-deepening search. The model is as accurate as previous offline techniques when using less training data, but can also be used online to reduce the overhead of iterative-deepening search, resulting in faster planning. Third we show offline planning algorithms that directly attempt to minimize the sum of planning and execution times. And, fourth we consider algorithms that plan online in parallel with execution. Both offline and online algorithms account for a user-specified preference between search and execution, and can greatly outperform the standard utility-oblivious techniques. By addressing the problem of planning under time pressure, these contributions demonstrate that heuristic search is no longer restricted to optimizing solution cost, obviating the need to choose between slow search times and expensive solutions.</p>","abstract_has_math":false,"creators":["Burns, Ethan"],"institution":null,"degree_name":"Doctor of Philosophy","degree_level":"Dissertation","degree_discipline":null,"degree_department":null,"school":null,"contributors":["Wheeler Ruml"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2013,"date_issued":"2013-01-01T08:00:00Z","date_published":"2013-01-01T08:00:00Z","updated_at":"2026-07-24T05:22:42Z","subjects":["Computer Science","Artificial Intelligence"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholars.unh.edu/dissertation/725","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Wheeler Ruml"]},{"key":"dc:creator","label":"Author","values":["Burns, Ethan"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer Science","Artificial Intelligence"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://scholars.unh.edu/dissertation/725"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Heuristic search is a technique used pervasively in artificial intelligence and automated planning. Often an agent is given a task that it would like to solve as quickly as possible. It must allocate its time between planning the actions to achieve the task and actually executing them. We call this problem planning under time pressure. Most popular heuristic search algorithms are ill-suited for this setting, as they either search a lot to find short plans or search a little and find long plans. The thesis of this dissertation is: when under time pressure, an automated agent should explicitly attempt to minimize the sum of planning and execution times, not just one or just the other.
This dissertation makes four contributions. First we present new algorithms that use modern multi-core CPUs to decrease planning time without increasing execution. Second, we introduce a new model for predicting the performance of iterative-deepening search. The model is as accurate as previous offline techniques when using less training data, but can also be used online to reduce the overhead of iterative-deepening search, resulting in faster planning. Third we show offline planning algorithms that directly attempt to minimize the sum of planning and execution times. And, fourth we consider algorithms that plan online in parallel with execution. Both offline and online algorithms account for a user-specified preference between search and execution, and can greatly outperform the standard utility-oblivious techniques. By addressing the problem of planning under time pressure, these contributions demonstrate that heuristic search is no longer restricted to optimizing solution cost, obviating the need to choose between slow search times and expensive solutions.
"]},{"key":"dc:title","label":"Title","values":["Planning under time pressure"]}]}],"canonical_facts":{"dc:contributor":["Wheeler Ruml"],"dc:creator":["Burns, Ethan"],"dc:description.abstract":["Heuristic search is a technique used pervasively in artificial intelligence and automated planning. Often an agent is given a task that it would like to solve as quickly as possible. It must allocate its time between planning the actions to achieve the task and actually executing them. We call this problem planning under time pressure. Most popular heuristic search algorithms are ill-suited for this setting, as they either search a lot to find short plans or search a little and find long plans. The thesis of this dissertation is: when under time pressure, an automated agent should explicitly attempt to minimize the sum of planning and execution times, not just one or just the other.
This dissertation makes four contributions. First we present new algorithms that use modern multi-core CPUs to decrease planning time without increasing execution. Second, we introduce a new model for predicting the performance of iterative-deepening search. The model is as accurate as previous offline techniques when using less training data, but can also be used online to reduce the overhead of iterative-deepening search, resulting in faster planning. Third we show offline planning algorithms that directly attempt to minimize the sum of planning and execution times. And, fourth we consider algorithms that plan online in parallel with execution. Both offline and online algorithms account for a user-specified preference between search and execution, and can greatly outperform the standard utility-oblivious techniques. By addressing the problem of planning under time pressure, these contributions demonstrate that heuristic search is no longer restricted to optimizing solution cost, obviating the need to choose between slow search times and expensive solutions.
"],"dc:identifier":["https://scholars.unh.edu/dissertation/725"],"dc:subject":["Computer Science","Artificial Intelligence"],"dc:title":["Planning under time pressure"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Doctor of Philosophy"]},"updated_at":"2026-07-24T05:22:42Z"}