When to Commit to an Action in Online Planning and Search

Abstract

In online planning, search is concurrent with execution. Under the formulation of planning as heuristic search, when a planner commits to an action, it re-roots its search tree at the node representing the outcome of that action. For the system to remain controlled, the planner must commit to a new action (perhaps a no-op) before the previously chosen action completes. This time pressure results in a real-time search. In this time-bounded setting, it can be beneficial to commit early, in order to perform more lookahead search focused below an upcoming state. In this paper, we propose a principled method for making this commitment decision. Our experimental evaluation shows that our scheme can outperform previously-proposed fixed strategies.

Planning and Execution in Multi-Agent Path Finding: Models and Algorithms

“…Then the algorithm commits the next k actions and re-roots the tree from the last location of the committed actions. Since RTHS generates only partial solutions during each commit, studies in this direction focus on improving the look-ahead overheads and developing dynamic commitment strategies to optimise the feasibility of the final plan (Gu et al 2022;Elboher et al 2023;Koenig and Sun 2009).…”

“…In concurrent planning and execution, both planning time and execution time are measured together. In single agent planning, this is called Goal Achievement Time (GAT) (Gu et al 2022), which is the time for a single agent from the planning start to reach the goal location in execution. For MAPF, the GAT is defined for each agent, as the sum of planning time and path length.…”

“…During the execution of these actions, PIE optimises the remaining paths of agents and then commits to a new set of actions. Concurrent algorithms have been previously studied in single-agent search (Korf 1990) where they are known to reduce waiting time for agents and improve Goal Achievement Time (GAT) (Gu et al 2022).…”

Planning and Acting While the Clock Ticks

“…Previous work on combining planning and execution, such as I X T E T-E XEC (Lemai and Ingrand 2004) and ROS-Plan (Cashmore et al 2015) (among others), focused on how to integrate an offline planner with a reactive executive to create an online planning and execution system. This does not address problems with tight deadlines, where plans can become infeasible during the planning process due to a deadline expiring during search.…”

“…This does not address problems with tight deadlines, where plans can become infeasible during the planning process due to a deadline expiring during search. Other work addressed the question of when to commit to dispatching an action during search (Gu et al 2022), but this work also does not reason about deadlines and is thus inapplicable to such settings.…”