Causal Search Tree Pruning for Faster Automated Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated planning systems struggle to rapidly filter out intermediate state nodes that are unrelated to the target node during search tree expansion, leading to inefficiencies in action plan generation for autonomous objects.
Innovation Solution
A search tree pruning method that utilizes a causal involvement between actions to construct and prune the search tree in an action space, reducing the search space by adding only those nodes with a causal relation to the current action as sub-nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional search tree expansion is used in automated planning systems, then the search tree can be constructed to represent possible states and transitions, but the system cannot rapidly filter out intermediate state nodes that are unrelated to the target node, leading to inefficient action plan generation
Solution Approach 1:
The patent pre-computes and stores causal relationships between actions in a causal graph before the search tree expansion begins. This preliminary organization of causal information allows the system to rapidly filter intermediate nodes during search by checking pre-established causal links rather than evaluating all possible transitions, thereby significantly reducing search time and improving action plan generation speed.
Solution Approach 2:
The patent introduces a causal graph as an intermediary structure that mediates between the action space and the search tree expansion process. This causal graph serves as a mediator that provides causal involvement information, enabling the system to efficiently identify and prune irrelevant intermediate nodes during search tree construction, thus reducing the time lost in filtering unnecessary nodes.
2Reliability
If the search tree is expanded to include all possible intermediate nodes, then the completeness of the action plan can be ensured, but the search space becomes excessively large, increasing computational resource requirements
Solution Approach 1:
The patent applies local quality by differentiating the treatment of nodes based on their causal relevance to the target state. Instead of uniformly expanding all intermediate nodes, the system selectively expands only those nodes that have causal involvement with the target node, as determined by the pre-computed causal graph. This localized expansion strategy maintains the completeness of the action plan for relevant paths while significantly reducing the overall search space size and computational complexity.
3Productivity
If conventional search tree methods are used, then the basic planning function can be implemented, but the system lacks efficient mechanisms for pruning irrelevant branches, resulting in suboptimal computational efficiency
Solution Approach 1:
The patent performs the complex causal relationship analysis as a preliminary action before the search tree expansion begins. By pre-computing the causal graph and storing causal involvement information, the system transforms the complex pruning decision-making process into simple lookups during search tree expansion. This preliminary computation shifts the computational complexity to an offline phase, enabling efficient online pruning without adding complexity to the main search algorithm.
Data Source
AI summary
Disclosed are a search tree pruning device and method for an automated planning system. The search tree pruning device for an automated planning system of the present invention includes a processor, and a memory configured to store instructions executed by the processor, wherein the processor selects a current action node from among front nodes of a search tree in an action space to execute an action of the current action node, and adds a new action node as a sub-node of the current action node according to whether the new action node has a preset causal involvement with the current action node.