Technical System Planning with Dynamic Heuristic Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing heuristic search methods in planning systems are inefficient due to varying heuristic performance across domains and failure to account for internal search dynamics, leading to suboptimal solutions and increased search effort.
Innovation Solution
Dynamic Heuristic Selection using a policy trained via Dynamic Algorithm Configuration (DAC) that learns to adjust algorithm parameters based on contextual Markov decision processes, selecting the most informative heuristic at each step to minimize state expansions and optimize goal state reachability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single heuristic is used for planning, then the planning process is simple, but the performance varies significantly across different domains and heuristics may be inaccurate
Solution Approach 1:
The patent applies dynamics by making the heuristic selection adaptive rather than static. The system dynamically selects heuristics based on the current search state, domain characteristics, and problem features. This allows the planning system to adapt to different domains and search situations, resolving the contradiction between simplicity and accuracy by introducing controlled dynamic complexity only when needed.
Solution Approach 2:
The patent changes the parameter of heuristic selection from a fixed single heuristic to a dynamic choice among multiple heuristics. By varying which heuristic is applied based on search progress and domain characteristics, the system achieves high accuracy across diverse domains while maintaining manageable complexity through systematic selection criteria.
2Loss of information
If multiple heuristics are combined in satisficing planning, then more information can be utilized, but combining inadmissible heuristic estimates is difficult and may have negative effects
Solution Approach 1:
The patent segments the heuristic evaluation process by maintaining separate heuristic estimates for different aspects of the search space. Instead of combining all heuristics into a single aggregated value, the system selectively applies individual heuristics or groups of heuristics based on their suitability for the current search state, domain type, and problem characteristics. This segmentation avoids the complexity of combining inadmissible heuristics while still utilizing multiple information sources.
Solution Approach 2:
The patent introduces an intermediary selection mechanism that mediates between multiple heuristics and the search process. This intermediary layer evaluates which heuristics are appropriate to apply at each search step, filtering out potentially harmful combinations while amplifying useful heuristic information. The selection mechanism acts as a bridge that manages the complexity of multiple heuristics without requiring direct combination of all estimates.
3Reliability
If heuristic search expands many states to find a solution, then a solution can be found, but the search effort and time increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-evaluating heuristic performance on training data and pre-determining which heuristics are most effective for different domain types and search situations. This preliminary analysis allows the system to make informed heuristic selections during actual search, avoiding the need to explore excessive states by choosing the right heuristic from the start or early in the search process.
Solution Approach 2:
The patent implements feedback mechanisms where the system monitors search progress, heuristic performance, and search state characteristics in real-time. Based on this feedback, the system dynamically adjusts heuristic selection to maintain effective search direction. This feedback loop prevents wasteful expansion of states by continuously adapting heuristic choices to the actual search dynamics, thereby reducing search time while maintaining solution reliability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method for planning an operation of a technical system within its environment. The method comprises the following steps: i) Obtaining state information comprising: a current domain, a time step and a current state; ii) Determining by heuristics costs for reachable states from the current state; iii) Selecting a heuristics by a policy out of a set of predefined heuristics depending on the state information and costs; iv) Choosing the state with the lowest cost returned by the selected heuristic from the reachable states, and v) Determining an operation of the technical system out of the set of possible operation that has to be carried out by the technical system to reach said state with the lowest costreturned by the selected heuristic.