A hybrid map construction method, device and medium for mobile robot navigation

Abstract

The application discloses a kind of hybrid map construction method, equipment and medium of mobile robot navigation, it is related to navigation hybrid map construction technical field, including acquisition multi-modal sensor observation stream, construct metric representation, topological representation and semantic representation, generate time version chain starting point time version;Metric representation, topological representation and semantic representation are updated in control period, generate difference representation, write difference representation into time version chain, generate continuous expansion time version chain;Generate structure stability, semantic drift degree, topological evolution consistency and update cost density quality curve, obtain quality curve set;Select current running hybrid map time version, carry out the local update of metric representation, semantic representation and topological representation, adjust time window length and time version switching strategy.Improve the objectivity of map quality evaluation and time version selection decision accuracy, improve map updating efficiency and overall operation performance.

Description

Technical Field

[0001] This invention relates to the field of navigation hybrid map construction technology, and in particular to a hybrid map construction method, device and medium for mobile robot navigation. Background Technology

[0002] Mobile robots rely on unified modeling of the spatial structure, connectivity, and semantic attributes of the surrounding environment to achieve autonomous navigation. They employ multi-sensor fusion technology to integrate observation data from multiple sources, such as laser, vision, and odometry, to construct a hybrid map that includes occupancy grids, topological nodes, and semantic tags. This hybrid map provides mobile robots with key information such as traversable areas, path connectivity, and environmental objects, supporting mobile robots in localization, path planning, and decision-making in structured or semi-structured scenarios.

[0003] In multi-source observation scenarios, conventional methods for constructing hybrid maps still have room for improvement in the temporal evolution dimension. On the one hand, conventional methods for constructing hybrid maps focus on local incremental updates and consistency correction, and do not adequately quantify the structural stability, semantic consistency, and topological evolution consistency across time windows, making it difficult to systematically evaluate map quality in the long-term evolution. On the other hand, the update strategy adopts a fixed periodic mechanism, and the time version switching and evolution chain management are relatively static, lacking a dynamic decision-making mechanism, which limits the adaptability of the navigation system under complex observation sequences. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a hybrid map construction method for mobile robot navigation to solve the problems of insufficient time version chain evolution expression system and lack of cross-time window quality quantification and control capabilities in existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a hybrid map construction method for mobile robot navigation, comprising,

[0008] Collect observation streams from multimodal sensors, perform time alignment and coordinate unification, generate fused observation frames, construct metric representations, topological representations and semantic representations, combine them to generate an initial version of the hybrid map, and generate the starting time version of the time version chain.

[0009] Based on the starting time version of the time version chain, the metric representation, topological representation, and semantic representation are updated within the control period, update events are recorded, differential representations are generated, and the differential representations are written into the time version chain to generate a continuously expanding time version chain.

[0010] Based on the continuously expanding time version chain, generate quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density, and obtain a set of quality curves;

[0011] Based on the set of quality curves, select the current running hybrid map time version, perform local updates to metric representation, semantic representation, and topological representation, and adjust the time window length and time version switching strategy.

[0012] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the steps of acquiring multimodal sensor observation streams, performing time alignment and coordinate unification, and generating fused observation frames are as follows:

[0013] Collect spatial geometric observation data, environmental semantic observation data, and odometry observation data to generate a multimodal sensor observation stream;

[0014] A reference timeline is constructed, and the multimodal sensor observation stream is sequentially subjected to monotonicity verification, frame loss detection, frequency alignment, and clock drift correction to generate a time-aligned multimodal sensor observation stream.

[0015] The time-aligned multimodal sensor observation stream is transformed to obtain a time-aligned and coordinate-unified multimodal sensor observation stream, which is then used to generate a fused observation frame.

[0016] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the construction of metric representation, topological representation and semantic representation, and the combination to generate an initial version of the hybrid map, and the generation of the starting time version of the time version chain, refers to constructing metric representation, topological representation and semantic representation by occupying grids according to the fused observation frames, combining them to generate an initial version of the hybrid map, establishing a cross-representation alignment index, and writing it into the starting time version of the time version chain.

[0017] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the steps of updating the metric representation, topological representation, and semantic representation within a control period based on the starting time version of the time version chain and recording update events are as follows:

[0018] Obtain the current control cycle fused observation frame, update the metric representation in the occupied raster index space in the order of addition, deletion and correction, obtain the sequence of metric representation change events, and generate the set of occupied raster indexes for metric representation changes.

[0019] Infer environmental connectivity observations within the raster index set occupied by metric representation changes, update the topology representation in the order of node update, edge update, and consistency check, and obtain the sequence of topology representation change events.

[0020] Within the raster index set occupied by metric representation changes, the semantic representation is updated in the order of region update, entity update, and category stability review to obtain the sequence of semantic representation change events.

[0021] In a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the step of generating a differential representation, writing the differential representation into a time version chain, and generating a continuously expanding time version chain includes the following specific steps:

[0022] Based on the sequence of metric representation change events, the sequence of topological representation change events, and the sequence of semantic representation change events, combined with the sequence of cross-representation alignment index change events, a differential representation is generated.

[0023] Based on the differential representation, the time version chain is extended, the time window boundaries are bound, archive segments are generated, and the continuously extended time version chain is obtained.

[0024] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the specific steps for generating structural stability, semantic drift, topological evolution consistency, and update cost density quality curves based on a continuously expanding time version chain, and obtaining a set of quality curves, are as follows:

[0025] Construct a time window event sequence, calculate and normalize the basic values ​​of structural stability and semantic drift respectively, and generate structural stability quality curves and semantic drift quality curves.

[0026] Based on the time window event sequence, the basic values ​​of topology evolution consistency and update cost density are calculated and normalized to generate the quality curves of topology evolution consistency and update cost density, and a set of quality curves is obtained.

[0027] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the specific steps of selecting the currently running hybrid map version based on the quality curve set are as follows:

[0028] Obtain the current control cycle timestamp, determine the target time window number and time version number range, read the quality curve set, and generate historical sequences of structural stability, semantic drift, topological evolution consistency, and update cost density;

[0029] Based on historical sequences of structural stability, semantic drift, and topological evolution consistency, historical baseline values ​​are calculated, and the current running hybrid map time version is selected.

[0030] As a preferred embodiment of the hybrid map construction method for mobile robot navigation described in this invention, the steps of performing local updates of metric representation, semantic representation, and topological representation, and adjusting the time window length and time version switching strategy, are as follows:

[0031] Trigger local updates to metric representation, semantic representation, or topological representation based on the current running hybrid map time version;

[0032] Adjust the time window length configuration parameters and time version switching strategy based on the historical sequence of update cost density.

[0033] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the hybrid map construction method for mobile robot navigation as described in the first aspect of the present invention.

[0034] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the hybrid map construction method for mobile robot navigation as described in the first aspect of the present invention.

[0035] The beneficial effects of this invention are as follows: by generating quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density, quantitative assessment and monitoring of the temporal evolution quality of hybrid maps are achieved, improving the objectivity of map quality assessment and the accuracy of time version selection decisions; by dynamically selecting the current running hybrid map time version based on the quality curve set and triggering local updates, precise control over the timing and scope of map maintenance is achieved, reducing computational resource consumption and improving map update efficiency and overall operational performance. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart of a hybrid map construction method for mobile robot navigation.

[0038] Figure 2 A flowchart for generating the starting time version of the time version chain.

[0039] Figure 3 A flowchart for generating a continuously expanding time version chain.

[0040] Figure 4 A flowchart for selecting a time version and adjusting the strategy. Detailed Implementation

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0044] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a hybrid map construction method for mobile robot navigation, including the following steps:

[0045] S1. Collect multimodal sensor observation streams, perform time alignment and coordinate unification, generate fused observation frames, construct metric representations, topological representations and semantic representations, combine them to generate an initial version of the hybrid map, and generate the starting time version of the time version chain.

[0046] Spatial geometric observation data, environmental semantic observation data, and odometry observation data are collected to generate a multimodal sensor observation stream. A reference time axis is constructed, and the multimodal sensor observation stream is sequentially subjected to monotonicity verification, frame loss detection, frequency alignment, and clock drift correction to generate a time-aligned multimodal sensor observation stream.

[0047] Furthermore, the mobile robot is equipped with a distance sensor, a vision sensor, and an odometry sensor. The distance sensor continuously collects spatial geometric observation data, represents the spatial geometric observation data using distance point sets or depth frames, and includes distance sensor timestamps to generate a distance sensor timestamp sequence. The vision sensor continuously collects environmental semantic observation data, represents the environmental semantic observation data using image frames, and includes vision sensor timestamps to generate a vision sensor timestamp sequence. The odometry sensor continuously collects odometry observation data, represents the odometry observation data using pose increment sequences, and includes odometry sensor timestamps to generate an odometry sensor timestamp sequence.

[0048] Spatial geometric observation data, environmental semantic observation data, and odometry observation data are arranged in chronological order to form a multimodal sensor observation stream.

[0049] Furthermore, time alignment is performed. Specifically, the timestamp of the odometer sensor is used as a reference time axis, and monotonicity checks are performed on the distance sensor sequence and the vision sensor sequence. If the timestamps are in reverse order, the observation data enters the out-of-order buffer, is reordered according to the timestamps, and then time alignment is performed.

[0050] Frame loss detection is performed. Specifically, if the timestamp interval exceeds the set upper limit of the time interval, the frame loss interval is recorded and frequency alignment is performed. Specifically, the control period is divided by the reference time axis, and within the control period, the observation frame with the timestamp closest to the center time of the control period is selected by the nearest neighbor selection rule.

[0051] Furthermore, clock drift correction is performed. Specifically, the cumulative offset of the distance sensor timestamp and the vision sensor timestamp relative to the reference time axis is statistically analyzed, and drift trend fitting mapping is performed to generate a time-aligned multimodal sensor observation stream.

[0052] The time-aligned multimodal sensor observation stream is subjected to coordinate transformation to obtain a time-aligned and coordinate-unified multimodal sensor observation stream, generating a fused observation frame. Based on the fused observation frame, by occupying the grid, metric representation, topological representation, and semantic representation are constructed and combined to generate an initial version of the hybrid map. A cross-representation alignment index is established and written to the starting time version of the time version chain.

[0053] Furthermore, coordinate unification is performed. Specifically, by using external parameter calibration parameters, a rigid body mapping is established from the distance sensor coordinate system and the vision sensor coordinate system to the robot coordinate system. The time-aligned multimodal sensor observation stream is transformed to the robot coordinate system through rigid body mapping, thereby obtaining a time-aligned and coordinate-unified multimodal sensor observation stream.

[0054] It should be noted that extrinsic parameter consistency maintenance is performed within the control cycle. Specifically, spatial geometric observation boundary segments and environmental semantic observation projection boundary segments are extracted within the overlapping view domain, and the average spatial offset is calculated. The average spatial offset is used as the alignment error. The alignment error is compared with the tolerance range. If the alignment error is within the tolerance range, the extrinsic parameter calibration parameters remain unchanged. If the alignment error exceeds the tolerance range, the extrinsic parameter calibration parameters are determined to be invalid. When the extrinsic parameter calibration parameters are invalid, boundary segment pairs from the most recent control cycles are selected, rigid body estimation is performed again, and the rigid body mapping is updated.

[0055] To obtain the tolerance interval, specifically, during the calibration phase and the initial stage of normal operation, a control cycle is selected, and the average spatial offset of each control cycle is used as the alignment error sample. The alignment error is measured by spatial distance in the robot coordinate system. Multiple alignment error samples are combined into an alignment error history sequence, and statistical analysis is performed to calculate the center position statistic and dispersion statistic of the alignment error history sequence. The upper and lower bounds of the alignment error are determined and used as the tolerance interval. During online operation, new alignment error samples are accumulated within the time window, the alignment error history sequence is updated, and the tolerance interval is updated.

[0056] Furthermore, the single-frame spatial geometric observation data, single-frame environmental semantic observation data, and corresponding odometry observation data within each control cycle are combined to generate a fused observation frame. The fused observation frame includes a set of distance points or depth frames in the robot coordinate system, an image frame, and pose increments.

[0057] To construct a metric representation, specifically, based on the spatial geometric observation data of the fused observation frames, the reachable space covering the fused observation frames' field of view is projected onto the occupied grid according to the projection model. The grid occupancy status is updated along the projection ray path, the projection path grid is marked as a passable grid, and the projection termination grid is marked as a non-passable grid, thus obtaining the spatial occupied element set. The spatial geometric observation data of the fused observation frames is then subjected to plane fitting, edge extraction, and clustering segmentation to obtain wall fragments, ground fragments, and obstacle fragments, which are written into the spatial geometric element set. Projection mapping from point sets to occupied grids is performed on the wall fragments, ground fragments, and obstacle fragments respectively, generating a geometric fragment coverage index set. The geometric fragment coverage index set is associated with the geometric fragment identifier and stored in the spatial geometric element set. Within the occupied grid index space, consistency cleanup is performed on the spatial occupied element set and the spatial geometric element set. Specifically, occupied grids covered by obstacle fragments in the spatial geometric element set are updated to non-passable grids, and occupied grids corresponding to the ground fragment support areas in the spatial geometric element set are updated to passable grids, while retaining correction logs.

[0058] Furthermore, based on the set of spatially occupied elements, deterministic derivation of environmental connectivity observations is performed. Specifically, the traversable grids within the set of spatially occupied elements are segmented into connected domains. Adjacent traversable grids sharing edges or corners are used as connectivity criteria for segmentation. The connected set is searched layer by layer through a queue until the queue is empty, obtaining traversable region blocks. Occupied grid indices with the same traversable region block number are aggregated to generate a traversable region block index set. Distance transformation and refinement are performed on the traversable region blocks to obtain the region centerline, which is used as the channel skeleton. Channel skeleton bifurcation points, channel skeleton endpoints, and continuous line segments satisfying length constraints are used as channel candidates, represented by the occupied grid index set. Each grid on the channel skeleton is traversed, and grid indices are matched. Matched grid indices are arranged sequentially according to the spatial extension direction of the channel skeleton to generate a channel skeleton index sequence. Connectivity boundaries are extracted from the set of boundary points of traversable region blocks and represented by the occupied grid index set. Channel candidates and connectivity boundaries are integrated to generate environmental connectivity observations characterizing the environmental connectivity features of the current observation range.

[0059] It should be noted that the length constraint refers to the minimum allowable length of continuous line segments in the occupied grid index space of the channel skeleton. Only continuous line segments with a length greater than or equal to the length constraint are considered as channel candidates to obtain the length constraint. Specifically, during the deployment phase, environmental connectivity observations are performed over multiple control cycles. The length of each continuous line segment is represented by the length of the occupied grid index sequence. A sample sequence of continuous line segment lengths in the channel skeleton is obtained, sorted, and statistically analyzed to determine the median position and the length distribution interval near the median. Combined with the distribution of continuous line segment lengths of stable travel paths in the map, the minimum length covering common stable travel paths and excluding obviously short and noisy line segments is taken as the length constraint. New continuous line segment length samples of the channel skeleton are accumulated within the time window, the sample sequence of continuous line segment lengths of the channel skeleton is updated, and the length constraint is updated.

[0060] Furthermore, based on environmental connectivity observations and the set of spatially occupied elements, a topological representation is constructed, the channel skeleton bifurcation points and endpoints are extracted, and candidate topological nodes are obtained. The candidate topological nodes are aggregated according to the merging distance to generate topological nodes. The consistency of the topological nodes is checked to obtain a set of topological nodes. A continuous path without bifurcation is searched along the channel skeleton between topological nodes to obtain topological edges. The grid index sequence occupied by the topological edges is recorded. If there are non-passable grids in the area covered by the topological edge, the topological edge is removed and the topological node is retained. If the set of spatially occupied elements in the area covered by the topological edge is a passable grid, the topological edge is retained. A topological connectivity consistency check is performed to generate a set of topological edges.

[0061] It should be noted that, specifically, in the offline calibration stage, to obtain the merging distance, a multi-frame occupancy grid map and a channel skeleton set are constructed based on long-term collected environmental connectivity observations and spatial occupancy element sets. Channel skeleton bifurcation points and endpoints are extracted as candidate topology nodes. The Euclidean distance between any two candidate topology nodes is calculated in the mobile robot coordinate system and converted into grid spacing according to the occupancy grid resolution, forming a sample of candidate topology node distances. These are then sorted and histograms are used to identify short-distance clusters close to zero distance using cluster analysis. The upper boundary of these short-distance clusters is used as the upper bound of the positioning error distribution. The left and right obstacle occupancy grids are searched along the channel skeleton normal direction, and the channel lateral width samples are calculated. The median value is statistically analyzed, and half of the median value of the channel lateral width samples is used as the estimated lateral half-width of the stable passage path. Based on the upper bound of the positioning error distribution and the lateral half-width of the stable passage path, a candidate merging distance interval is constructed. Historical topology representations are aggregated and simulated under the candidate merging distance to evaluate the degree of topology connectivity preservation. The smallest candidate merging distance that satisfies the evaluation constraints is selected as the merging distance. For example, in conventional indoor mobile robot navigation applications, the upper bound of the positioning error distribution is usually (0.03, 0.15) meters, and the lateral width of the stable passage path is usually (1.2, 3) meters. In the mobile robot coordinate system, the merging distance is preferably set to 1 to 3 times the upper bound of the positioning error distribution, and the ratio of the merging distance to the lateral width of the stable passage path is (0.2, 0.6), corresponding to a merging distance of (0.1, 1) meters. In the occupied grid index space, according to the occupied grid resolution, the value of the merging distance in the mobile robot coordinate system is converted into the number of grid spacings. The converted merging distance is (2, 12) grid spacings. Topology node aggregation can suppress positioning jitter, avoid different channels or inflection points being mistakenly merged, and ensure the accuracy of the topology connectivity structure expression.

[0062] Based on environmental semantic observation data, a set of spatial occupancy elements, and a set of spatial geometric elements, a semantic representation is constructed. Specifically, passable regions in the set of spatial occupancy elements are used as semantic region candidates, and the spatial range of the semantic region candidates is represented by an occupancy grid index set. Based on the environmental semantic observation data, a region type recognition algorithm is executed to generate region type labels, which are limited to a preset set of collectable categories and written into the semantic region set. Based on the environmental semantic observation data, object detection and multi-frame tracking are performed to obtain a set of semantic entities, where the semantic entity categories are limited to a preset set of collectable categories. The semantic entities correspond to image coordinate domain spatial range markers, and the spatial range markers are aligned and calibrated with depth values ​​or distance point sets within the same fused observation frame. Combined with visual projection technology, the data is back-projected onto the robot coordinate system to obtain the spatial range of the semantic entities. The spatial range of the semantic entities is discretized into the occupancy grid index space to generate a semantic entity index set, which represents the spatial distribution of semantic entities. Within a sliding time window, the semantic region set and the semantic entity set are checked for category consistency. If category conflicts exist, the category with the most occurrences is taken as the stable category, ensuring that the object tracking numbers of semantic entities are unique and continuous.

[0063] Furthermore, within the same control cycle, the metric representation, topological representation, and semantic representation are combined to generate an initial version of the hybrid map. A cross-representation alignment index is established within the occupied grid index space. The channel skeleton index sequence is associated with the set of spatial occupied elements, the set of topological nodes, and the set of topological edges. The set of passable area block indexes is associated with the set of spatial occupied elements and the set of semantic regions. The intersection of the set of semantic entity indexes and the set of geometric fragment coverage indexes is associated with the set of semantic entities and the set of spatial geometric elements.

[0064] Write the initial version of the hybrid map, the cross-representation alignment index, and the control cycle timestamp into the starting time version of the time version chain. Initialize the time window management data structure, create a time window index table, use the control cycle timestamp corresponding to the starting time version of the time version chain as the starting timestamp of the first time window, use the starting time version number of the time version chain as the starting time version number within the first time window, and register the correspondence between the first time window number, the starting timestamp of the first time window, and the starting time version number in the time window index table.

[0065] S2. Based on the starting time version of the time version chain, update the metric representation, topological representation, and semantic representation within the control period, record update events, generate differential representations, write the differential representations into the time version chain, and generate a continuously expanding time version chain.

[0066] Obtain the current control cycle fused observation frame, update the metric representation in the occupied raster index space in the order of addition, deletion, and correction, obtain the metric representation change event sequence, and generate the metric representation change occupied raster index set; derive the environmental connectivity observation within the metric representation change occupied raster index set, update the topology representation in the order of node update, edge update, and consistency check, and obtain the topology representation change event sequence.

[0067] Furthermore, the mobile robot repeatedly performs multimodal sensor observation stream acquisition, time alignment, coordinate unification, and external parameter consistency maintenance in each control cycle to obtain the fused observation frame of the current control cycle. Based on the fused observation frame of the current control cycle, the hybrid map is updated. The hybrid map update order is: incremental update of metric representation, re-derivation of environmental connectivity observation, incremental update of topological representation, incremental update of semantic representation, and category stability verification.

[0068] Incremental updates of metric representations are performed based on the occupied raster index space. Specifically, the update sequence includes addition, deletion, and correction operations. The addition operation updates the spatial occupied element set and spatial geometric element set through spatial geometric observation projection, writing the new raster and geometric fragment into the corresponding set. The deletion operation marks rasters and geometric fragments that have not been updated or verified by fused observation frames for a long time as invalid elements and moves them to the historical archive area, which stores the deleted occupied raster index set and timestamp. Correction operations are performed when loop closure detection or global consistency correction is triggered, correcting the pose of affected geometric fragments, re-overwriting the occupied raster, synchronously correcting the spatial occupied element set according to consistency cleanup rules, and writing the corrected occupied raster index set and correction timestamp into the correction log. The newly added occupied raster index set, the deleted occupied raster index set, and the corrected occupied raster index set generated by the incremental updates of metric representations are merged to obtain the metric representation change occupied raster index set.

[0069] Furthermore, the raster occupancy status data corresponding to the raster index set occupied by the metric representation change is extracted to obtain the local spatial occupancy element set. Within the outer connected domain of the raster index set occupied by the metric representation change, the environmental connectivity observation is re-derived based on the local spatial occupancy element set to obtain local environmental connectivity observation. The re-derived environmental connectivity observation includes extracting passable regions, extracting channel skeletons, generating channel candidates, and generating connectivity boundaries. Based on the local environmental connectivity observation and the local spatial occupancy element set, the topology representation is incrementally updated to obtain the topology representation change event sequence. The order of the topology representation incremental update is node update, edge update, and consistency check.

[0070] Specifically, node updates are performed by extracting topology node candidates from the channel skeleton bifurcation points and endpoints, aggregating these candidates to generate topology nodes, and temporarily storing any topology node that falls into a non-travelable grid as an invalid topology node. Topology edge updates are also performed by searching for non-bifurcation continuous paths between adjacent topology nodes along the channel skeleton to generate topology edges and recording the occupied grid index sequence. If an old topology edge conflicts with a new non-travelable grid, the old topology edge is removed. Consistency checks are performed by checking the traversability of the topology edge covering the index sequence one by one. If a non-travelable index exists, the corresponding topology edge is removed, and the topology node is retained for subsequent reconnection.

[0071] Within the raster index set occupied by metric representation changes, the semantic representation is updated in the order of region update, entity update, and category stability review to obtain the semantic representation change event sequence. Based on the metric representation change event sequence, topological representation change event sequence, and semantic representation change event sequence, combined with the cross-representation aligned index change event sequence, a differential representation is generated.

[0072] Furthermore, based on the environmental semantic observations, current metric representations, and current spatial geometric element sets of the current control cycle fusion observation frames, incremental updates of semantic representations are performed. The update order is region update, entity update, and category stability verification. Specifically, within the connected domain circumscribed by the raster index set occupied by the metric representation changes, traversable region blocks are re-segmented, and these traversable region blocks are used as semantic region candidates. Based on environmental semantic observations, region type labels are identified and written into the semantic region set. Through object detection and tracking, semantic entity set candidates are obtained. Through visual projection and spatial geometric observation matching, the spatial range of semantic entities is determined, and the spatial range of semantic entities is mapped to the occupied raster index set. The semantic entity set is added, deleted, and corrected. Category consistency verification is performed on the semantic region set and the semantic entity set to suppress category jitter caused by short-term misidentification and maintain the continuity of object tracking numbers.

[0073] It should be noted that the differential representation is stored in the form of a sequence of change events, including a sequence of change events for metric representation, a sequence of change events for topological representation, a sequence of change events for semantic representation, and a sequence of change events for cross-representation alignment index.

[0074] The metric represents the sequence of change events, recording events such as adding raster, deleting raster, correcting raster, adding geometric fragment, deleting geometric fragment, and correcting geometric fragment. Each event includes the change type, the set of raster indexes occupied by the change, the change timestamp, the local fragment data before the change, and the local fragment data after the change. The local fragment data only covers the range of the change index set.

[0075] Furthermore, the topology representation change event sequence records events such as adding topology nodes, deleting topology nodes, correcting topology nodes, adding topology edges, deleting topology edges, and correcting topology edges; each event includes change type, topology identifier, occupied raster index sequence, change timestamp, connected data before change, and connected data after change.

[0076] The semantic representation change event sequence records events such as adding semantic regions, deleting semantic regions, correcting semantic regions, adding semantic entities, deleting semantic entities, and correcting semantic entities; each event includes change type, raster index set occupied by the change, category tag, change timestamp, category before change, and category after change.

[0077] It should be noted that the cross-representation alignment index change event sequence records the events of adding, deleting, and correcting cross-representation alignment in the hybrid map, including the change type, the set of raster indexes occupied by the change, the change timestamp, the correspondence data before the change, and the correspondence data after the change. Among them, the correspondence data before the change and the correspondence data after the change include corresponding entries between metric representation elements, topological representation elements, and semantic representation elements within the occupied raster index space.

[0078] The updated cross-representation alignment index of the current control cycle is compared with the cross-representation alignment index stored in the previous time version. Within the occupied raster index space, the set of occupied raster indexes that have added, deleted, or changed corresponding relationships are generated to form cross-representation alignment index change events. The cross-representation alignment index change events are added to the cross-representation alignment index change event sequence according to the change timestamp.

[0079] Furthermore, the sequence of metric representation change events, the sequence of topological representation change events, the sequence of semantic representation change events, and the sequence of cross-representation alignment index change events are merged into a differential representation.

[0080] Based on the differential representation, the time version chain is extended, the time window boundaries are bound, archive segments are generated, and the continuously extended time version chain is obtained.

[0081] Furthermore, based on the differential representation, the time version chain is extended by adding a new time version at the end of the time version chain. The new time version includes a time version number, a change timestamp, a differential representation, and an updated cross-representation alignment index. The updated cross-representation alignment index establishes the correspondence between the three types of change event sequences under the same change timestamp.

[0082] Write the time window boundary in the time version metadata area. The time window for the time version chain archive and the time window for the computation quality curve use the same window boundary.

[0083] The time window start timestamp is determined based on the control cycle timestamp when the time window opens, and the time window end timestamp is determined based on the current control cycle timestamp when the time window length reaches the time window length configuration parameter. The time version numbers when the time window opens and closes are recorded, a time version number range is generated, the time window boundary is written into the time version metadata area, and a new index record is added to the time window index table. The index record includes the time window number, the time window start timestamp, the time window end timestamp, and the correspondence between the time version number range.

[0084] Furthermore, at the end of the time window, the set of consecutive time versions within the time window is archived, and redundant change events are merged. Redundant change events are determined based on the same event type, adjacent timestamps, and overlapping raster index sets. The merged event sequence maintains the original time order, generates archive segments, writes them to the time version chain archive area, and registers the correspondence between the archive segment identifier and the time window number in the time window index table.

[0085] It should be noted that the time version chain is backtracked by accumulating the archive segment differential representation and the differential representation outside the time window in reverse according to the time version number, and the target time version mixed map state is reconstructed.

[0086] S3. Based on the continuously expanding time version chain, generate quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density, obtain the set of quality curves, and write them into the time version metadata area.

[0087] Construct a time window event sequence, calculate and normalize the basic values ​​of structural stability and semantic drift respectively, and generate structural stability quality curves and semantic drift quality curves.

[0088] Furthermore, the target time window number and the corresponding start timestamp, end timestamp, and time version number range are read from the time window index table. Within the time version number range, each time version is traversed in order of time version number, and metric representation change events, topological representation change events, semantic representation change events, and cross-representation alignment index change events are extracted respectively. According to the change timestamp order, they are written into the same time window event sequence. The time window event sequence fully covers the four types of change events, the occupied raster index set, and related attributes within the time window.

[0089] From the event sequence within the time window, select subsequences of events with the characteristic type of metric characteristic change to obtain the metric characteristic change event sequence. Perform a union operation on the occupied grid index set of each change event to obtain the occupied grid index set of metric characteristic change within the time window. From the metric characteristic change event sequence, select change events with the change type of correction. Perform a union operation on the occupied grid index set of the correction change events to obtain the occupied grid index set of metric correction within the time window. Calculate the basic value of structural stability within the time window.

[0090] Specifically, the basic numerical value of structural stability within the time window is expressed as follows:

[0091] ;

[0092] in, Indicates time window The basic values ​​of structural stability, Indicates time window Internal metric corrections occupy the raster index set. Indicates time window Internal metric representations of changes occupy the raster index set, subscript Represents the metric representation. Indicates the time window number.

[0093] It should be noted that when the number of raster index sets occupied by the change in metric representation within the time window is 0, it means that there is no change in metric representation within the time window, and the basic value of structural stability is 1, indicating that the metric structure remains completely stable.

[0094] Determine the minimum and maximum values ​​of the basic structural stability values ​​across all historical time windows, normalize the current basic structural stability values ​​to obtain the normalized structural stability, and store the time window number and the normalized structural stability in the structural stability mass curve storage sequence in chronological order, while simultaneously updating the minimum and maximum values ​​of the basic structural stability values.

[0095] Furthermore, subsequences of events with semantic representation change type are filtered from the event sequence within the time window to obtain the semantic representation change event sequence. The set of occupied grid indices in the semantic representation change event sequence is obtained and a union operation is performed to generate the set of occupied grid indices of semantic representation change within the time window. The category labels before and after the change are compared to obtain the set of occupied grid indices of the change event with the changed category label and the union is calculated to generate the set of occupied grid indices of semantic category change within the time window. The ratio of the number of occupied grid indices of semantic category change within the time window to the number of occupied grid indices of semantic representation change within the time window is calculated to obtain the basic value of semantic drift.

[0096] Furthermore, based on the minimum and maximum values ​​of the semantic drift baseline, the current semantic drift baseline is normalized to obtain the normalized semantic drift. The time window number and the normalized semantic drift are written into the semantic drift quality curve storage sequence, and the minimum and maximum values ​​of the semantic drift baseline are updated.

[0097] Based on the time window event sequence, the basic values ​​of topology evolution consistency and update cost density are calculated and normalized to generate the quality curves of topology evolution consistency and update cost density, and a set of quality curves is obtained.

[0098] Furthermore, subsequences of events with topological representation change type are filtered from the time window event sequence to obtain the topological representation change event sequence. The set of occupied raster indices in the topological representation change event sequence is obtained, and the set of occupied raster indices of topological representation change within the time window is generated by taking the union. The set of adjacent topological nodes and topological edge index sequences in the data before and after the change of the topological representation change event are examined. Change events with changes in adjacency or topological edge sequences are filtered, and the set of occupied raster indices of topological connectivity change within the time window is obtained by taking the union.

[0099] The basic numerical value of topology evolution consistency is calculated by measuring the number of raster index sets occupied by topology connectivity changes within the time window and the number of raster index sets occupied by topology representation changes within the time window.

[0100] Specifically, the basic numerical representation of topological evolution consistency is as follows:

[0101] ;

[0102] in, Indicates time window The basic numerical value of topological evolution consistency. This indicates the amount of raster index set occupied by topological connectivity changes within a time window. Indicates the area occupied by topological representation changes within the time window, subscript Represents topological characterization.

[0103] It should be noted that when the number of raster index sets occupied by topological representation changes within a time window is 0, the basic value of topological evolution consistency is 1. Based on the minimum and maximum values ​​of the basic value of topological evolution consistency in historical time windows, the current basic value is normalized to obtain the normalized topological evolution consistency. The time window number and the normalized topological evolution consistency are stored in the topological evolution consistency quality curve storage sequence, and the minimum and maximum values ​​of the basic value of topological evolution consistency are updated.

[0104] To obtain the update cost density quality curve, specifically, based on the time window event sequence, the total number of changing events in the time window event sequence is counted. The total number of changing events includes the sum of four types of changing events. For each changing event, the set of occupied raster indices is read, and the union of these sets is used to obtain the set of occupied raster indices covered by the time window event. The basic value of the update cost density is calculated by the ratio of the total number of changing events to the number of occupied raster indices covered by the time window event.

[0105] It should be noted that when the number of raster indexes occupied by the time window event coverage is 0, the basic value of the update cost density is 0. Based on the minimum and maximum values ​​of the basic values ​​of the update cost density in the historical time window, the current basic value of the update cost density is normalized to generate a normalized update cost density. The time window number and the normalized update cost density are written into the update cost density quality curve storage sequence, along with the minimum and maximum values ​​of the basic value of the update cost density.

[0106] It should be noted that the set of quality curves includes structural stability quality curves, semantic drift quality curves, topological evolution consistency quality curves, and update cost density quality curves.

[0107] Generate the latest fragment of the quality curve set, select the last time version within the time version number range of the time window, and write the latest fragment of the quality curve set into the time version metadata area of ​​the last time version.

[0108] The time window index table records the correspondence between the time window number and the end time version number, as well as the storage location of the quality curve set in the time version chain.

[0109] S4. Select the current running hybrid map time version based on the quality curve set, perform local updates of metric representation, semantic representation and topological representation, and adjust the time window length and time version switching strategy.

[0110] Obtain the current control cycle timestamp, determine the target time window number and time version number range, read the quality curve set, and generate historical sequences of structural stability, semantic drift, topological evolution consistency, and update cost density.

[0111] Furthermore, at the beginning of each navigation control cycle, the timestamp of the current control cycle is obtained, the time window records in the time window index table are traversed, and the time window record whose endpoint timestamp is less than or equal to the current control cycle timestamp and is closest is used as the target time window number and the corresponding time version number range.

[0112] Select the final time version number within the time version number range, read the quality curve set, extract the historical normalized values ​​of the target time window, and generate the historical sequence of structural stability, semantic drift, topological evolution consistency, and update cost density.

[0113] Based on the historical sequence of structural stability, semantic drift, and topological evolution consistency, calculate the historical baseline value and select the current running hybrid map time version; trigger local updates of metric representation, semantic representation, or topological representation according to the current running hybrid map time version.

[0114] Furthermore, by using median statistics, historical benchmark values ​​for structural stability, semantic drift, and topological consistency are determined in the historical sequences of structural stability, semantic drift, and topological consistency.

[0115] The normalized quality curve values ​​of the target time window are compared with the corresponding historical benchmark values. If the normalized structural stability is greater than or equal to the historical benchmark value of structural stability, the normalized semantic drift is less than or equal to the historical benchmark value of semantic drift, and the normalized topological evolution consistency is greater than or equal to the historical benchmark value of topological evolution consistency, then the time version at the end of the target time window is used as the current running hybrid map time version for positioning fusion, path planning, and motion control.

[0116] If the end time version of the target time window does not meet the comparison condition, the end time versions of earlier time windows and their corresponding quality curve sets are accessed in descending order of time window number. The first end time version that meets the comparison condition is taken as the current running hybrid map time version.

[0117] If none of the historical time windows meet the comparison criteria, then the time version at the end of the target time window will be used as the time version of the currently running hybrid map.

[0118] Furthermore, based on the recent sequences of the structural stability quality curve, semantic drift quality curve, and topological evolution consistency quality curve, the local update requirement is determined. Specifically, if the normalized structural stability of multiple consecutive time windows is consistently lower than the historical benchmark value of structural stability, then based on the metric representation change events in the target time window event sequence, the outer connected component of the metric correction occupying the raster index set within the time window is constructed as the local reconstruction space range of the metric representation. Incremental updates and consistency cleanup of the metric representation are performed, new metric representation change events are generated, new differential representations are obtained, written to the end of the time version chain, the cross-representation alignment index field is updated, and the metric representation is locally reconstructed.

[0119] If the normalized semantic drift of multiple consecutive time windows is greater than the historical baseline value of semantic drift, then based on the semantic representation change events in the event sequence of the target time window, the outer connected component of the set of raster indexes occupied by semantic category changes within the time window is constructed. Incremental updates of semantic representation and category stability verification are performed. The set of semantic regions and the set of semantic entities are identified and their consistency is verified again. A new sequence of semantic representation change events and a new set of raster indexes occupied by semantic representation changes are generated and merged into the differential representation. A time version is added at the end of the time version chain to correct the local semantic layer of the hybrid map.

[0120] If the normalized topological evolution consistency of multiple consecutive time windows is less than the historical benchmark value of topological evolution consistency, based on the topological representation change events in the event sequence of the target time window, the outer connected domain of the topological connectivity change occupying the raster index set within the time window is constructed. Environmental connectivity observation is re-derived and topological representation is incrementally updated again. The topological node set and topological edge set are regenerated, the channel skeleton connectivity structure is repaired, a new topological representation change event sequence and topological representation change occupying the raster index set are generated, written to the new time version, and the cross-representation alignment index is updated synchronously.

[0121] Adjust the time window length configuration parameters and time version switching strategy based on the historical sequence of update cost density.

[0122] Furthermore, based on the statistical relationship between the historical and recent sequences of update cost density, the time window length configuration parameter is increased when the update cost density is consistently low, and decreased when the update cost density is consistently high. The time window number range corresponding to the time version identifier of the currently running hybrid map is maintained in the time version metadata area.

[0123] Specifically, the normalized update cost density of all time windows is read from the update cost density quality curve storage sequence to construct a historical update cost density sequence. The normalized update cost density of the most recent time window is selected to construct a recent update cost density sequence. The historical update cost density baseline value is obtained by median statistics in the historical update cost density sequence. The normalized update cost density value is compared with the historical update cost density baseline value to determine whether the update cost density is in a long-term low or long-term high state. If the update cost density is in a long-term low state, the time window length configuration parameter is increased according to a preset incremental step size. If the update cost density is in a long-term high state, the time window length configuration parameter is decreased according to a preset decrementing step size.

[0124] It should be noted that obtaining the time window length configuration parameter involves, specifically, constructing a time version chain and a time window index table based on the offline navigation task set, statistically analyzing the changes in the number of time windows and the quality curves for structural stability, semantic drift, and topology evolution consistency, and combining storage resource constraints and real-time computing capability constraints to determine the target range for the number of time windows. Within multiple candidate time window lengths, the time windows are re-divided, and the quality curve set is reconstructed. The minimum candidate time window length is selected that satisfies the condition that the number of time windows is within the target range and that the quality curve set can distinguish between the slow and drastic change phases of the mixed map state. The ratio of the minimum candidate time window length to the control cycle duration is calculated, the number of control cycles is obtained, and the result is rounded up to obtain the time window length configuration parameter.

[0125] To obtain the preset incremental step size, specifically, after determining the time window length configuration parameters, construct multiple time window length increments, redivide the time windows of the offline navigation task set, reconstruct the quality curve set, statistically analyze the number of time windows and the historical baseline value of update cost density, select the minimum time window length increment that satisfies the conditions of significantly reducing the number of time windows, maintaining smooth changes in the quality curve, and not increasing the historical baseline value of update cost density, calculate the ratio of the minimum time window length increment to the control cycle duration, obtain the number of control cycles, round up, and obtain the preset incremental step size.

[0126] To obtain the preset decreasing step size, specifically, after determining the time window length configuration parameters, construct multiple time window length reduction amounts, re-divide the time windows for the offline navigation task set, reconstruct the quality curve set, and statistically analyze the number of time windows, the number of quality curve segments, the historical baseline value of update cost density, and the number of change events in a single time window. Select the maximum time window length reduction amount that satisfies the requirements of quality curve resolution improvement, historical baseline value of update cost density, and number of change events in a single time window meeting storage resource constraints and real-time computing capability constraints, and whose time window number is within the target range of time window number. Calculate the ratio of the maximum time window length reduction amount to the control cycle duration, round it up to obtain the number of control cycles, and use the rounded-up control cycle number as the preset decreasing step size.

[0127] It should be noted that the current running hybrid map time version identifier is only updated when the new quality curve set satisfies the following conditions within multiple consecutive time windows: normalized structural stability is greater than or equal to the historical baseline value of structural stability; normalized semantic drift is less than or equal to the historical baseline value of semantic drift; normalized topological evolution consistency is greater than or equal to the historical baseline value of topological evolution consistency; and normalized update cost density has not entered a long-term high or long-term low state. The hybrid map time version is selected and locally updated by driving the selection of the quality curve set, while maintaining the stability of time version switching during navigation control.

[0128] This embodiment also provides a computer device applicable to the hybrid map construction method for mobile robot navigation, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the hybrid map construction method for mobile robot navigation as proposed in the above embodiment.

[0129] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0130] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the hybrid map construction method for mobile robot navigation as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] In summary, this invention achieves quantitative assessment and monitoring of the temporal evolution quality of hybrid maps by generating quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density, thereby improving the objectivity of map quality assessment and the accuracy of time version selection decisions. By dynamically selecting the current running hybrid map time version based on the quality curve set and triggering local updates, it achieves precise control over the timing and scope of map maintenance, reduces computational resource consumption, and improves map update efficiency and overall operational performance.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A hybrid map construction method for mobile robot navigation, characterized in that: include, Collect observation streams from multimodal sensors, perform time alignment and coordinate unification, generate fused observation frames, construct metric representations, topological representations and semantic representations, combine them to generate an initial version of the hybrid map, and generate the starting time version of the time version chain. Based on the starting time version of the time version chain, the metric representation, topological representation, and semantic representation are updated within the control period, update events are recorded, differential representations are generated, and the differential representations are written into the time version chain to generate a continuously expanding time version chain. Based on the continuously expanding time version chain, generate quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density, and obtain a set of quality curves; Based on the set of quality curves, select the current running hybrid map time version, perform local updates to metric representation, semantic representation, and topological representation, and adjust the time window length and time version switching strategy.

2. The hybrid map construction method for mobile robot navigation as described in claim 1, characterized in that: The process of acquiring multimodal sensor observation streams, performing time alignment and coordinate unification, and generating fused observation frames involves the following steps: Collect spatial geometric observation data, environmental semantic observation data, and odometry observation data to generate a multimodal sensor observation stream; A reference timeline is constructed, and the multimodal sensor observation stream is sequentially subjected to monotonicity verification, frame loss detection, frequency alignment, and clock drift correction to generate a time-aligned multimodal sensor observation stream. The time-aligned multimodal sensor observation stream is transformed to obtain a time-aligned and coordinate-unified multimodal sensor observation stream, which is then used to generate a fused observation frame.

3. The hybrid map construction method for mobile robot navigation as described in claim 2, characterized in that: The construction of metric representation, topological representation, and semantic representation, combined to generate an initial version of the hybrid map, and the generation of the starting time version of the time version chain refers to constructing metric representation, topological representation, and semantic representation based on the fused observation frames by occupying grids, combining them to generate an initial version of the hybrid map, establishing a cross-representation alignment index, and writing it into the starting time version of the time version chain.

4. The hybrid map construction method for mobile robot navigation as described in claim 3, characterized in that: The process involves updating the metric representation, topological representation, and semantic representation within the control period based on the time-version chain starting time version, and recording update events. The specific steps are as follows: Obtain the current control cycle fused observation frame, update the metric representation in the occupied raster index space in the order of addition, deletion and correction, obtain the sequence of metric representation change events, and generate the set of occupied raster indexes for metric representation changes. Infer environmental connectivity observations within the raster index set occupied by metric representation changes, update the topology representation in the order of node update, edge update, and consistency check, and obtain the sequence of topology representation change events. Within the raster index set occupied by metric representation changes, the semantic representation is updated in the order of region update, entity update, and category stability review to obtain the sequence of semantic representation change events.

5. The hybrid map construction method for mobile robot navigation as described in claim 4, characterized in that: The process of generating a differential representation and writing it into a time version chain to generate a continuously expanding time version chain involves the following steps: Based on the sequence of metric representation change events, the sequence of topological representation change events, and the sequence of semantic representation change events, combined with the sequence of cross-representation alignment index change events, a differential representation is generated. Based on the differential representation, the time version chain is extended, the time window boundaries are bound, archive segments are generated, and the continuously extended time version chain is obtained.

6. The hybrid map construction method for mobile robot navigation as described in claim 5, characterized in that: The step involves generating quality curves for structural stability, semantic drift, topological evolution consistency, and update cost density based on a continuously expanding time version chain, and obtaining a set of quality curves. The specific steps are as follows: Construct a time window event sequence, calculate and normalize the basic values ​​of structural stability and semantic drift respectively, and generate structural stability quality curves and semantic drift quality curves. Based on the time window event sequence, the basic values ​​of topology evolution consistency and update cost density are calculated and normalized to generate the quality curves of topology evolution consistency and update cost density, and a set of quality curves is obtained.

7. The hybrid map construction method for mobile robot navigation as described in claim 6, characterized in that: The specific steps for selecting the currently running hybrid map time version based on the quality curve set are as follows: Obtain the current control cycle timestamp, determine the target time window number and time version number range, read the quality curve set, and generate historical sequences of structural stability, semantic drift, topological evolution consistency, and update cost density; Based on historical sequences of structural stability, semantic drift, and topological evolution consistency, historical baseline values ​​are calculated, and the current running hybrid map time version is selected.

8. The hybrid map construction method for mobile robot navigation as described in claim 7, characterized in that: The specific steps for performing local updates to metric representation, semantic representation, and topological representation, and adjusting the time window length and time version switching strategy are as follows: Trigger local updates to metric representation, semantic representation, or topological representation based on the current running hybrid map time version; Adjust the time window length configuration parameters and time version switching strategy based on the historical sequence of update cost density.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the hybrid map construction method for mobile robot navigation as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the hybrid map construction method for mobile robot navigation as described in any one of claims 1 to 8.

Images