Quantum path planning method based on reference point constraint and hierarchical strategy
By introducing reference point constraints and a hierarchical strategy into quantum path planning, key topological information is filtered and large-scale maps are decomposed into small-scale sub-regions. This solves the problems of high resource consumption and limited processing power in complex environments, and achieves efficient path planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV OF INFORMATION SCI & TECH
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-23
Smart Images
Figure CN122062695B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of path planning technology, and more specifically to a quantum path planning method based on reference point constraints and a hierarchical strategy. Background Technology
[0002] Quantum computing, as an emerging computing paradigm, leverages quantum superposition and quantum parallelism to demonstrate computational power far exceeding that of classical algorithms for specific problems. For example, quantum random walks (QRW) enable parallel exploration of space, while quantum amplitude amplification (QAA) accelerates the selection of target paths through interference phenomena. Although existing research has introduced quantum algorithms into motion planning, these studies are mostly limited to theoretical analysis and small-scale discrete environments (such as 4×4 or 8×8 grids), and primarily serve as heuristic optimization tools embedded in classical frameworks, lacking independent, efficient planning schemes for real-world complex environments. This leads to an exponential increase in the size and depth of quantum circuits constructed by existing methods when facing large-scale maps or complex scenarios with narrow passages.
[0003] Currently, quantum path planning methods still face significant challenges in practical applications. First, current quantum computing is in the mid-range noisy quantum (NISQ) era, and physical limitations hinder the realization of large-scale stable quantum computers. As map dimensions increase, the number of qubits and the depth of quantum circuits increase exponentially or significantly, exceeding the processing capabilities of existing quantum hardware and making quantum simulations on classical computers extremely difficult. Second, existing quantum path planning methods struggle with complex real-world terrains containing critical information such as narrow passages due to a lack of environmental feature extraction capabilities, resulting in an excessively large search space and limited pathfinding success rate and scalability. The low search efficiency and high resource consumption of traditional path planning methods in complex environments, coupled with the increased size and depth of quantum circuits due to increased map dimensions, pose challenges to quantum devices and simulation environments in the NISQ era. Summary of the Invention
[0004] Purpose of the invention: The main purpose of this invention is to provide a quantum path planning method based on reference point constraints and hierarchical strategies, which solves the problems of low search efficiency and high resource consumption of traditional path planning methods in complex environments, as well as the challenges that quantum path planning methods pose to quantum devices and simulation environments in the NISQ era due to the increased size and depth of quantum circuits caused by the increase in map size.
[0005] Technical solution: The quantum path planning method based on reference point constraints and hierarchical strategy described in this invention includes the following steps:
[0006] Step 1: Obtain map information of the robot's environment and preprocess the map based on the reference point constraint strategy. By analyzing the distribution characteristics of feasible regions in the neighborhood of the sampling points in the map, reference points containing key topological information are selected to construct a sparse map for narrowing the search space.
[0007] Step 2: Based on the hierarchical strategy, the original map containing reference points is recursively divided into a macro map layer and a micro map layer, decomposing the original large-scale path planning problem into planning problems for multiple small-scale sub-regions;
[0008] Step 3: Map the starting point and ending point of the path planning to a hierarchical coordinate system containing macroscopic and microscopic coordinates, and use quantum ground state encoding technology to encode the map information, reference point information and the hierarchical coordinate information into a quantum register to complete the initialization of the quantum state;
[0009] Step 4: On the macro map layer, use the quantum motion planner to search and generate a macro path consisting of a sequence of macro cells to determine the global direction of motion;
[0010] Step 5: Within each macroscopic cell determined by the macroscopic path sequence, using the geometric center of the next macroscopic cell as a guide, call the quantum motion planner within the current cell to generate an accurate local microscopic path;
[0011] Step 6: Based on the global direction of the macroscopic path, the generated local microscopic paths are sequentially spliced together, and a collision-free path from the starting point to the ending point is output through quantum measurement.
[0012] Furthermore, in step 1, the selection of reference points is as follows: uniformly sample the map and mark the points within the feasible area as valid points; calculate the spatial distribution density of other valid points within the neighborhood of each valid point; based on the spatial distribution density, select valid points representing key areas such as intersections or narrow passages as map reference points.
[0013] Further, step 2 is as follows: Construct a spatial segmentation tree with the entire original map as the root node; recursively divide the spatial region represented by each node into multiple sub-regions of the same size until the sub-regions meet the preset stopping conditions, thereby forming a macro map composed of macro cells and a micro map composed of micro cells.
[0014] Furthermore, step 4 is as follows: the state space is explored by combining quantum random walk and the quantum amplitude amplification algorithm is used.
[0015] Further, step 5 is as follows: calculate the centroid position of the next macro cell in the macro path sequence; within the current macro cell, find the feasible position closest to the centroid as a local temporary target point; on the micro map layer corresponding to the macro cell, call the quantum motion planner to generate a collision-free micro path from the current micro position to the local temporary target point.
[0016] The quantum path planning system based on reference point constraints and a hierarchical strategy described in this invention includes:
[0017] Acquisition module: used to acquire map information of the robot's environment and preprocess the map based on the reference point constraint strategy. By analyzing the distribution characteristics of feasible regions in the neighborhood of the sampling points in the map, reference points containing key topological information are selected to construct a sparse map to reduce the search space.
[0018] Layered module: Used to recursively divide the original map containing reference points into macro map layer and micro map layer based on a layered strategy, decomposing the original large-scale path planning problem into planning problems of multiple small-scale sub-regions;
[0019] Initial module: used to map the starting point and ending point of the path planning to a hierarchical coordinate system containing macroscopic and microscopic coordinates, and to use quantum ground state encoding technology to encode map information, reference point information and the hierarchical coordinate information into a quantum register to complete the initialization of the quantum state;
[0020] Macro Path Module: Used on the macro map layer to search using the quantum motion planner and generate a macro path consisting of a sequence of macro cells to determine the global direction of motion;
[0021] Microscopic Path Module: Within each macroscopic cell determined by the macroscopic path sequence, the geometric center of the next macroscopic cell is used as a guide to call the quantum motion planner to generate a precise local microscopic path within the current cell;
[0022] Output module: Based on the global direction of the macroscopic path, it sequentially splices together the generated local microscopic paths and outputs a collision-free path from the starting point to the ending point through quantum measurement.
[0023] Furthermore, in the acquisition module, the specific steps for filtering reference points are as follows: uniformly sample the map and mark the points within the feasible area as valid points; calculate the spatial distribution density of other valid points within the neighborhood of each valid point; and based on the spatial distribution density, filter out valid points representing key areas such as intersections or narrow passages as map reference points.
[0024] Furthermore, in the layered module, the specific steps are as follows: a spatial segmentation tree is constructed, with the entire original map as the root node; the spatial region represented by each node is recursively divided into multiple sub-regions of the same size until the sub-regions meet the preset stopping conditions, thereby forming a macro map composed of macro cells and a micro map composed of micro cells.
[0025] Furthermore, in the macroscopic path module, the specific steps are as follows: state space exploration is performed by combining quantum random walks, and quantum amplitude amplification algorithm is used.
[0026] Furthermore, in the micro-path module, the specific steps are as follows: calculate the centroid position of the next macro-cell in the macro-path sequence; within the current macro-cell, find the feasible position closest to the centroid as a local temporary target point; on the micro-map layer corresponding to the macro-cell, call the quantum motion planner to generate a collision-free micro-path from the current micro-position to the local temporary target point.
[0027] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: By introducing a reference point constraint strategy, this invention preprocesses complex environments and filters out map reference points with key topological information, thereby significantly reducing the spatial range of quantum path search and improving the algorithm's adaptability to complex terrains such as narrow passages; by utilizing a hierarchical strategy and a spatial segmentation tree structure, the large-scale original map is recursively divided into macro-maps and micro-maps. While maintaining a high pathfinding success rate, this invention significantly shortens pathfinding time and enhances the practicality and scalability of quantum path planning methods in real-world complex environments. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the map reference point selection process, used to select reference points from the original map.
[0029] Figure 2 This is a schematic diagram of map spatial segmentation and coordinate mapping based on a hierarchical strategy. It is used to recursively divide a two-dimensional grid map into two levels: a macroscopic map and a microscopic map, using a spatial segmentation tree. Figure 2 (a) in the figure shows the process of recursively dividing an initial two-dimensional space into macro-cells and finer-grained micro-cells using a spatial partitioning tree; Figure 2 (b) in the diagram illustrates the coordinate representation method after the corresponding layering;
[0030] Figure 3 This is an overall flowchart of the quantum path planning method based on reference point constraints and hierarchical strategies proposed in this invention.
[0031] Figure 4 It is a change diagram of the actual map processing process, used to illustrate the processing flow. Detailed Implementation
[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings.
[0033] like Figure 3 The embodiment of the present invention shown provides a quantum path planning method based on reference point constraints and a hierarchical strategy, including the following steps: Figure 3 The core processing flow of the algorithm is intuitively demonstrated: First, the original map is preprocessed using a reference point selection algorithm to extract key information; then, a layering strategy is used to divide the map into two levels: macro and micro; next, a quantum motion planner is used to plan the global direction on the macro map, and centroid guidance is used to perform local precise planning on the micro map; finally, the macro path and micro path are organically combined to output the final path from the starting point to the destination.
[0034] Step 1: Map Preprocessing and Reference Point Extraction: The original map is preprocessed based on the Reference Point Constraint (RPC) strategy. Reference point constraints involve preprocessing the map before path planning to select as few map reference points as possible that possess key information and satisfy the connectivity requirements for path construction. The specific operations are as follows:
[0035] (11) Perform uniform sampling on the map at fixed intervals, mark obstacles as invalid points and the remaining pixels as valid points, and construct a set of sampling points. .
[0036] (12) For each sampling point With the center as the center, Define a circular neighborhood for the radius, calculate the Euclidean distance between every pair of valid points within this neighborhood, and then perform a weighted average. For any two valid points within the neighborhood... and The Euclidean distance is calculated using the following formula:
[0037]
[0038] in, and Indicates the valid point Position in a two-dimensional coordinate system and Indicates the valid point Position in a two-dimensional coordinate system.
[0039] And calculate the weighted average of the distances between all valid point pairs within the neighborhood. This is used to characterize the degree of aggregation of valid points in the neighborhood of the sampling point.
[0040] (13) Weighted average The smaller the value, the higher the concentration of valid points, which usually corresponds to a large unobstructed area on the map; weighted average. The larger the value, the more dispersed the effective points are, typically corresponding to key areas on the map such as intersections or narrow passages. Finally, by setting an appropriate threshold range Th, when... When that happens, select that point as a map reference point.
[0041] like Figure 1 As shown, Figure 1 The left side shows the original map, with black areas representing obstacles and white areas representing feasible areas; Figure 1 The right side shows the specific operation of reference point filtering. The blue dots represent valid points obtained by uniform sampling, and the yellow dashed circles represent the neighborhood calculation range with the current sampling point as the center and R as the radius. This illustrates the process of extracting key map reference points by calculating the degree of aggregation of valid points within this range.
[0042] Step 2: Hierarchical Map Partitioning: A hierarchical strategy (HS) is adopted, using a spatial partitioning tree (such as a quadtree structure) to recursively divide the 2D grid map into two levels: a macro-map composed of macro cells and a micro-map composed of micro cells. The macro-map is used to determine the general direction of global planning, while the micro-map is used to achieve precise obstacle avoidance locally. This partitioning method decomposes the originally single large-scale quantum circuit into multiple smaller quantum circuit modules, effectively reducing the algorithm's requirement for the number of qubits. The specific partitioning process is as follows:
[0043] (21) Initial space: The root node of the spatial partitioning tree represents the entire two-dimensional space and serves as the starting point for partitioning.
[0044] (22) Recursive partitioning: Each node divides the current space into m sub-regions of the same size. This process is performed recursively until the termination condition is met. The value of m can be updated according to the actual situation of the map during each recursion.
[0045] (23) Termination condition: The segmentation process stops when the number of objects in a certain region is lower than the preset threshold or the region size reaches the minimum granularity. Figure 2 (a) in the figure shows the process of recursively dividing an initial two-dimensional space into macro-cells and finer-grained micro-cells using a spatial partitioning tree; Figure 2 (b) shows the coordinate representation method after the corresponding layering, that is, each smallest grid unit is determined by both macroscopic and microscopic coordinates, which can be represented as:
[0046]
[0047] in, Indicates the grid positions in the original map. Indicates the corresponding macroscopic coordinates, This represents the corresponding micro coordinates, thus decomposing a large-scale map into multiple small-scale submaps that can be independently planned.
[0048] Step 3: State Initialization and Coordinate Mapping: The global start and global goal of the path planning are divided into coordinates according to the hierarchical logic described above, and mapped to corresponding macroscopic and microscopic coordinates respectively. The initial quantum state of the system is prepared using the ground state encoding method. For the coordinates that need to be encoded Its ground-state encoding mapping is:
[0049]
[0050] in, and The coordinates to be encoded are represented by "Encode", and "Encode" represents the ground-state encoding mapping process. This represents the initial quantum position state.
[0051] The system's total state space after initialization From position state Coin state Map information and endpoint status tensor product Composition, namely:
[0052]
[0053] Step 4: Macro-level Global Path Planning: Apply a quantum motion planner (optional: Quantum Global Path Planner QFMP or Goal-Guided Quantum Motion Planner QGMP) at the macro-level map. Utilize quantum random walk (QRW) for state evolution, and combine it with the quantum amplitude amplification (QAA) algorithm to amplify the amplitude of the target solution. Search for and generate a macro-path (macro_path) composed of a sequence of macro-level cells to determine the global motion trend. The specific operations are as follows:
[0054] (41) State evolution: Applying a step size of Quantum random walk operator Generate path superposition state:
[0055]
[0056] in, Represents the initial superposition state of the system. This represents the superposition of paths generated after applying the walk operator. This represents a quantum random walk operator with a step size of M. , Encode the starting operator. Coin flipper (in two-dimensional space) ), It is a conditional shift operator.
[0057] (42) Amplification and Measurement: Constructing Amplification Operators Collision detection and target recognition are performed.
[0058]
[0059] in, For the constructed amplitude amplification operator, For quantum random walk operators, Its inverse operator, A phase of -1 is added to the ground state that simultaneously satisfies "no collision occurred" and "the endpoint is the target macroscopic unit" using a Boolean function. Iterative application. Second-rate The operator, whose estimated number of iterations is:
[0060]
[0061] Where r represents the map obstacle concentration. A macroscopic path (macro_path) composed of a sequence of macroscopic cells is output through quantum measurement. The step size.
[0062] Step 5: Centroid-guided micro-local planning: Within the cell sequence defined by the macro-path, centroid-guided precise local planning is performed sequentially. Specifically, the geometric centroid position of the next macro-cell in the path sequence is first calculated. Secondly, within the current macro cell, the nearest feasible location to the centroid is found as a local temporary target point. The distance is calculated using the Manhattan distance metric to optimize computational overhead.
[0063]
[0064] in, Represents the calculated Manhattan distance, point Represents the currently feasible location (local temporary target point) and , Its coordinates, point Represents the geometric centroid position of the next macroscopic cell and , Let its coordinates be given.
[0065] Finally, the quantum motion planner is invoked to generate a micropath within the current micro-region, enabling precise local obstacle avoidance and exploration.
[0066] Step 6: Path Integration and Output: Based on the global direction determined by the macroscopic path, the microscopic sub-paths generated in each stage are connected and integrated sequentially. Finally, by measuring the quantum state, a collision-free, asymptotically optimal complete motion sequence (move_seq) from the global starting point to the global ending point is output.
[0067] like Figure 4 As shown, Figure 4 This demonstrates the changes during the actual map processing. First, redundant information in the original map is processed by removing color information and shop names. Next, to facilitate pixel information processing, the image is binarized. Finally, interfering paths are eliminated and outlines are added. The final map information is as follows: Figure 4 As shown in the bottom right corner: the size is 900×900 pixels, the area outside the black border is an obstacle, and the area inside the black border is a feasible area.
[0068] experiment:
[0069] Dataset: The experiment uses a real map environment for verification. The map dataset was collected from the floor plan overview of the second floor of Longhu Tianjie in Jiangbei District, Nanjing (located at No. 99 Wenjing Road, Pukou District, Jiangbei District, Nanjing).
[0070] Data preprocessing: To ensure the real map information is simple and clear and meets the needs of subsequent algorithm construction, preprocessing is required. First, redundant information in the original map is removed by eliminating color information and shop names. Next, to facilitate pixel information processing, the image is binarized. Finally, interfering paths are eliminated and outlines are added.
[0071] Experimental Platform: This experiment uses Python and Qiskit as the quantum computing simulation platform.
[0072] Experimental setup:
[0073] Regarding parameter configuration, after extensive digital simulation experiments, the reference point selection parameters were determined to be interval=10, radius R=30, and threshold range Th=(7,+∞). Based on these parameter settings, 75 effective map reference points were selected from the map and mapped onto a 32×32 two-dimensional grid map. Using a layered strategy, this 32×32 map was further divided into 64 4×4 dimension sub-maps to adapt to the processing capabilities of existing sub-simulation environments.
[0074] The experiment selected four mainstream RRT variants (RRT, Goal-bias RRT, RRT*, and RRT-Connect), two quantum path planning methods (QPRM and q-RRT), and a quantum heuristic algorithm (RL-QPSO) as control groups. Ten different start and end points were selected for the test task, and 100 simulation experiments were conducted for each group to record the average running time and success rate.
[0075] Experimental results:
[0076] Experiments compared the average pathfinding time and success rate of different algorithms in real-world complex environments, and the results are shown in Table 1. Among them, the quantum path planning methods (QFMP+HS+RPC and QGMP+HS+RPC) proposed in this invention based on reference point constraints and hierarchical strategies showed significantly improved performance. However, the QFMP, QGMP, and QPRM methods could not operate normally due to the quantum circuit size exceeding the platform's processing capacity or algorithmic limitations, and no simulation results were obtained.
[0077] QFMP+HS+RPC saves 67.5% of pathfinding time compared to RRT, 33% compared to RRT-Connect, 53% compared to q-RRT, and 32% compared to RL-QPSO. QGMP+HS+RPC saves 71% of pathfinding time compared to RRT, 42% compared to RRT-Connect, 59% compared to q-RRT, and 41% compared to RL-QPSO. Comparing the pathfinding success rates in Table 1 reveals that QFMP+HS+RPC and QGMP+HS+RPC both have a success rate of 1, higher than other algorithms. This further verifies the effectiveness of the proposed method.
[0078] Table 1. Comparison of pathfinding time and success rate of different algorithms in real-world maps.
[0079] .
Claims
1. A quantum path planning method based on reference point constraints and a hierarchical strategy, characterized in that, Includes the following steps: Step 1: Obtain map information of the robot's environment and preprocess the map based on the reference point constraint strategy. By analyzing the distribution characteristics of feasible regions in the neighborhood of the sampling points in the map, reference points containing key topological information are selected to construct a sparse map for narrowing the search space. Step 2: Based on the hierarchical strategy, the original map containing reference points is recursively divided into a macro map layer and a micro map layer, decomposing the original large-scale path planning problem into planning problems for multiple small-scale sub-regions; Step 3: Map the starting point and ending point of the path planning to a hierarchical coordinate system containing macroscopic and microscopic coordinates, and use quantum ground state encoding technology to encode the map information, reference point information and the hierarchical coordinate information into a quantum register to complete the initialization of the quantum state; Step 4: On the macro map layer, use the quantum motion planner to search and generate a macro path consisting of a sequence of macro cells to determine the global direction of motion; Step 5: Within each macroscopic cell determined by the macroscopic path sequence, using the geometric center of the next macroscopic cell as a guide, call the quantum motion planner within the current cell to generate an accurate local microscopic path; Step 6: Based on the global direction of the macroscopic path, the generated local microscopic paths are sequentially spliced together, and a collision-free path from the starting point to the ending point is output through quantum measurement.
2. The quantum path planning method based on reference point constraints and hierarchical strategy according to claim 1, characterized in that, In step 1, the selection of reference points is as follows: uniformly sample the map and mark the points within the feasible area as valid points; calculate the spatial distribution density of other valid points within the neighborhood of each valid point; based on the spatial distribution density, select valid points representing key areas of intersections or narrow passages as map reference points.
3. The quantum path planning method based on reference point constraints and hierarchical strategy according to claim 1, characterized in that, Step 2 is as follows: Construct a spatial segmentation tree with the entire original map as the root node; recursively divide the spatial region represented by each node into multiple sub-regions of the same size until the sub-regions meet the preset stopping conditions, thereby forming a macro map composed of macro cells and a micro map composed of micro cells.
4. The quantum path planning method based on reference point constraints and hierarchical strategy according to claim 1, characterized in that, Step 4 is as follows: Combine quantum random walk to explore the state space and use the quantum amplitude amplification algorithm.
5. The quantum path planning method based on reference point constraints and hierarchical strategy according to claim 1, characterized in that, Step 5 is as follows: Calculate the centroid position of the next macro cell in the macro path sequence; within the current macro cell, find the feasible position closest to the centroid as the local temporary target point; on the micro map layer corresponding to the macro cell, call the quantum motion planner to generate a collision-free micro path from the current micro position to the local temporary target point.
6. A quantum path planning system based on reference point constraints and a hierarchical strategy, characterized in that, include: Acquisition module: used to acquire map information of the robot's environment and preprocess the map based on the reference point constraint strategy. By analyzing the distribution characteristics of feasible regions in the neighborhood of the sampling points in the map, reference points containing key topological information are selected to construct a sparse map to reduce the search space. Layered module: Used to recursively divide the original map containing reference points into macro map layer and micro map layer based on a layered strategy, decomposing the original large-scale path planning problem into planning problems of multiple small-scale sub-regions; Initial module: used to map the starting point and ending point of the path planning to a hierarchical coordinate system containing macroscopic and microscopic coordinates, and to use quantum ground state encoding technology to encode map information, reference point information and the hierarchical coordinate information into a quantum register to complete the initialization of the quantum state; Macro Path Module: Used on the macro map layer to search using the quantum motion planner and generate a macro path consisting of a sequence of macro cells to determine the global direction of motion; Microscopic Path Module: Within each macroscopic cell determined by the macroscopic path sequence, the geometric center of the next macroscopic cell is used as a guide to call the quantum motion planner to generate a precise local microscopic path within the current cell; Output module: Based on the global direction of the macroscopic path, it sequentially splices together the generated local microscopic paths and outputs a collision-free path from the starting point to the ending point through quantum measurement.
7. The quantum path planning system based on reference point constraints and hierarchical strategy according to claim 6, characterized in that, In the acquisition module, the specific steps for filtering reference points are as follows: uniformly sample the map and mark the points within the feasible area as valid points; calculate the spatial distribution density of other valid points within the neighborhood of each valid point; and based on the spatial distribution density, filter out valid points representing key areas of intersections or narrow passages as map reference points.
8. The quantum path planning system based on reference point constraints and hierarchical strategy according to claim 6, characterized in that, In the layered module, the specific steps are as follows: Construct a spatial segmentation tree with the entire original map as the root node; recursively divide the spatial region represented by each node into multiple sub-regions of the same size until the sub-regions meet the preset stopping conditions, thereby forming a macro map composed of macro cells and a micro map composed of micro cells.
9. The quantum path planning system based on reference point constraints and hierarchical strategy according to claim 6, characterized in that, In the macroscopic path module, the specific steps are as follows: state space exploration is performed by combining quantum random walks, and quantum amplitude amplification algorithm is used.
10. The quantum path planning system based on reference point constraints and hierarchical strategy according to claim 6, characterized in that, In the micro-path module, the specific steps are as follows: calculate the centroid position of the next macro cell in the macro path sequence; within the current macro cell, find the feasible position closest to the centroid as a local temporary target point; on the micro map layer corresponding to the macro cell, call the quantum motion planner to generate a collision-free micro path from the current micro position to the local temporary target point.