Causal Action Network Heuristics for Faster Task Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous task planning systems for devices like robots and self-driving cars are inefficient in generating action plans due to the slow creation of search trees, as they often select unrelated nodes during expansion, leading to increased time and resource consumption.
Innovation Solution
An automated task planning system using heuristic evaluation values calculated on the basis of a causal action network, where a processor estimates distances to a target node and selects nodes with short heuristic evaluation values for expansion, creating an action space search tree by connecting action nodes through edges corresponding to states, and utilizing a knowledge base to process precondition and postcondition data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional search tree expansion methods are used to generate action plans, then completeness of search is maintained, but search time and computational resources increase significantly
Solution Approach 1:
The patent pre-calculates and stores heuristic evaluation values for all possible action nodes before actual search execution. This preliminary preparation allows the search algorithm to quickly retrieve and compare pre-computed values rather than calculating them on-demand during search, significantly reducing search time while maintaining complete exploration of relevant nodes through informed expansion selection
Solution Approach 2:
The system uses previously computed heuristic evaluation values to guide its own search process. By storing evaluation results from prior computations and using them to select which nodes to expand next, the search algorithm serves itself with pre-prepared information, avoiding redundant calculations and focusing computational resources on promising search paths
2Reliability
If all nodes are expanded uniformly during search tree creation, then no node is missed, but the search tree size increases and creation time extends
Solution Approach 1:
The patent changes the parameter used for node selection from uniform or random selection to selection based on heuristic evaluation values. By using these pre-calculated values that estimate the promise of each node, the system dynamically adjusts which nodes are expanded at each step, ensuring that high-potential nodes are explored while avoiding unnecessary expansion of low-potential nodes, thus maintaining completeness for relevant nodes while improving creation speed
3Device complexity
If heuristic evaluation values are calculated without using causal action networks, then calculation simplicity is maintained, but search efficiency and performance deteriorate
Solution Approach 1:
The patent introduces causal action networks as an intermediary structure that pre-organizes action nodes and their relationships based on causal dependencies. This intermediary representation allows the heuristic evaluation to leverage pre-computed causal information, reducing the complexity of on-the-fly evaluation while dramatically improving search efficiency by guiding expansion toward causally relevant nodes
Data Source
AI summary
Provided are an automated task planning system and method using a heuristic evaluation value calculated on the basis of a causal action network. The automated task planning system is a task planning system for calculating heuristic evaluation values of search tree nodes for an action space on the basis of a causal action network, creating a search tree related to completion or achievement of a task, and generating an action plan of an autonomous thing on the basis of the search tree, and includes a processor configured to create an action space search tree. During the creation of the action space search tree, the processor calculates heuristic evaluation values of front nodes and determines an expansion node. The action space search tree is created by connecting action nodes which are nodes corresponding to actions through edges corresponding to states.