A local grid map updating method and apparatus

Abstract

The application discloses a local grid map updating method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: determining historical attribute information of each grid in a latest historical local grid map; acquiring multi-source perception data at a current time, and determining grids in the historical local grid map that are matched with the coordinates of each piece of perception data as target grids; determining the trusted features of each piece of perception data, the update intention of the historical attribute information of the target grids, combining the source types and the target grids, and constructing update candidates of each piece of perception data; determining the update range of the historical attribute information of the target grids based on each update candidate acting on the target grids; and updating the historical attribute information of the target grids based on the update range and the update intention, obtaining current attribute information, and generating a local grid map at the current time. The embodiment realizes the differentiated update of the adaptive update range of different source perception data for the same target grid.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and apparatus for updating local raster maps. Background Technology

[0002] In the construction of existing local raster maps, the updating of raster attribute information typically employs a fixed incremental approach. However, this "one-size-fits-all" fixed incremental approach ignores the actual impact of sensing data from different sources on the updating of raster attribute information, leading to the excessive amplification of low-confidence noise and insufficient response from high-confidence observations. Therefore, the existing fixed incremental approach struggles to balance the map's noise resistance and response sensitivity, thus affecting the overall stability and security of the map and failing to meet the real-time and security requirements of map updates in high-frequency navigation scenarios. Summary of the Invention

[0003] In view of this, the present invention provides a method and apparatus for updating local raster maps, which can at least solve the problem that the fixed incremental mode in the prior art cannot reflect the actual impact of sensing data from different sources on the updating of raster attribute information.

[0004] To achieve the above objectives, according to one aspect of the present invention, a local raster map updating method is provided, comprising: Determine the historical attribute information of each grid in the latest historical local grid map; Acquire multi-source sensing data at the current moment, and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid for each sensing data. Determine the credible characteristics of each of the sensing data and the intention to update the historical attribute information of the target raster, and construct update candidates for each of the sensing data by combining the source type of each of the sensing data and the target raster to which it is applied; Based on each of the update candidates acting on the target raster, the update magnitude of the historical attribute information of the target raster is determined; Based on the update magnitude and the update intention, the historical attribute information of the target raster is updated to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

[0005] To achieve the above objectives, according to another aspect of the present invention, a local raster map updating apparatus is provided, comprising: The determination module is used to determine the historical attribute information of each grid in the latest historical local grid map; The matching module is used to acquire the multi-source sensing data at the current moment and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid of each sensing data. A construction module is used to determine the credible characteristics of each of the sensed data and the intention to update the historical attribute information of the target raster, and to construct update candidates for each of the sensed data by combining the source type of each sensed data and the target raster to which it is applied. A calculation module is used to determine the update magnitude of the historical attribute information of the target raster based on each of the update candidates acting on the target raster; The update module is used to update the historical attribute information of the target raster based on the update magnitude and the update intention, to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

[0006] To achieve the above objectives, according to another aspect of the present invention, a local grid map updating electronic device is provided.

[0007] The electronic device of this invention includes: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement any of the local raster map update methods described above.

[0008] To achieve the above objectives, according to another aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, which, when executed by a processor, implements any of the local raster map update methods described above.

[0009] To achieve the above objectives, according to another aspect of the present invention, a computing program product is provided. One computing program product of the present invention includes a computer program, which, when executed by a processor, implements the local raster map update method provided in the present invention.

[0010] According to the solution provided by the present invention, one embodiment of the invention has the following advantages or beneficial effects: After acquiring the multi-source sensing data of the current frame, it is not directly superimposed, but rather the update intention of each sensing data on the matched target raster is analyzed, and update candidates are generated by combining the source type and credibility features of each sensing data. This decouples the multi-source sensing data into independent update intentions, effectively avoiding the coverage conflict and semantic loss problems caused by the direct superposition of multi-source sensing data. At the same time, the update magnitude is determined based on multiple update candidates acting on the same target raster, realizing adaptive differentiated updates for sensing data from different sources. This suppresses the excessive amplification of low-confidence noise and ensures sufficient response of high-confidence observations, thereby significantly improving the update reliability, noise resistance, and real-time performance of local raster maps in high-frequency navigation scenarios.

[0011] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0012] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein: Figure 1 This is a schematic diagram of the main process of a local raster map update method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an optional local raster map update method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the architecture for updating a local raster map according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the main modules of a local raster map updating device according to an embodiment of the present invention; Figure 5 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied; Figure 6 This is a schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present invention, such as a mobile device or server. Detailed Implementation

[0013] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in this disclosed technical solution all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to safeguard user personal information security, network security, and national security.

[0015] The following is an explanation of the terms used in this plan: Local grid map: A two-dimensional gridded map of a limited area around the robot, where each grid is used to represent occupied, idle, unknown, cost, semantic or dynamic auxiliary states.

[0016] Main sensing data (or main sensing sources): Data sources used to obtain information about the main environmental obstacles around the robot, such as LiDAR, depth camera, RGB (Red Green Blue)-D camera, millimeter-wave radar, or fused point cloud.

[0017] Near-field blind spot sensing data (or near-field blind spot source): Data sources used to supplement the insufficient near-range coverage of the main sensing, which can come from blind spot cameras, wide-angle / fisheye cameras, near-range depth, ToF (Time of flight), short-range radar or algorithm modules.

[0018] Semantic-aware data (or semantic sources): Data sources used to output target category, location, size, pose, or coverage area. Dynamic-aware data (or dynamic sources): Data sources used to represent the impact of moving targets, temporary obstacles, or dynamic point observations. Semantic and dynamic data can be processed using detection, segmentation, tracking, or dynamic point cloud classification modules.

[0019] Update Candidate: An update intent proposed by a source for one or more rasters in the current frame. It typically includes the raster or region to be affected, the source type, the base weight of the source type, the confidence feature, and the update intent.

[0020] Trustworthy features: Features used to describe the reliability of observations, such as distance, number of observations, observation density, target category, target velocity, source confidence, or algorithm confidence.

[0021] Historical observation characteristics: persistent or credible evidence formed by the grid in historical updates, such as historical height, historical occupancy level, number of consecutive observations, cumulative confidence, duration of existence, or source reliability.

[0022] Adaptive update magnitude: A mechanism that dynamically determines the occupancy probability, clearance probability, log-odds (also known as log odds), or change in cost based on the source type, the basic weight of the source type, reliable features, and historical observation features.

[0023] Adaptive forgetting: The forgetting period is determined based on the historical sensing data set used in the last raster update, the last update time, and historical observation characteristics, and forgetting processing such as decay, clearing, undoing, or partial reset is performed.

[0024] In scenarios such as warehousing and logistics, last-mile delivery, industrial inspection, and service robots, it is necessary to build a local environment map of the mobile robot in real time and provide this local environment map to the local path planning and obstacle avoidance control module to ensure the safety and efficiency of robot navigation. Currently, the industry generally uses local grid maps (or local cost maps) as a form of environmental indication, which discretizes the space around the robot into multiple grids, and each grid maintains attribute information such as occupancy probability, idle probability, unknown state, or navigation cost.

[0025] The construction of existing local raster maps primarily relies on master sensing data provided by master sensing units (such as LiDAR, depth cameras, millimeter-wave radar, or fused point clouds). A typical process includes transforming the master sensing data to a map coordinate system, using obstacle observation markers to occupy grids, and utilizing observable empty areas or ray paths to reduce grid occupancy probability. To compensate for insufficient near-field coverage by master sensing units, engineering systems typically introduce near-field blind spot sensing data; in dynamic mixed-traffic scenarios involving people and vehicles, semantic sensing data or dynamic sensing data are further introduced. However, existing technologies often use this multi-source sensing data as additional layers, additional obstacle sources, or temporary information during the planning phase, lacking a unified source representation mechanism and map maintenance rules.

[0026] Furthermore, for gratuitous grid cells in local raster maps that have not been updated for a long time, existing methods often employ strategies such as setting a fixed timeout threshold, periodic map decay, or full map scan reset. While these methods are simple to implement, in high-frequency navigation scenarios involving multi-source sensing, near-field blind spots, and dynamic targets, it is difficult to simultaneously ensure the reliability of map updates, the safety of obstacle removal, and the stability of long-term operation. Specifically, existing technologies mainly suffer from the following drawbacks: 1. The presence of near-field blind spots in the main sensing unit leads to unreliable near-field region updates for the robot. Depth cameras, LiDAR, or other main sensing units typically have limitations such as minimum effective ranging, installation height, pitch angle, horizontal field of view, or occlusion. If the local grid map relies primarily on the main sensing unit for marking and clearing, areas invisible to the main sensing unit at close range may remain unobserved for extended periods. This can result in near-field obstacle remnants (e.g., obstacles removed but still showing as occupied on the map), missed near-field obstacle detection (e.g., obstacles present but the map shows as free), or, in the absence of real-world observations, incorrect assumptions about free areas leading to the accidental clearing of actual obstacles, severely threatening the robot's operational safety.

[0027] 2. The lack of a unified hierarchical management mechanism for multi-source sensing data fusion easily leads to overlay conflicts or loss of source semantics. Existing simple overlay methods often treat sensing data from different sources as independent layers or ordinary obstacle sources, resulting in ambiguity regarding the source type, occupancy meaning, clearing permissions, and subsequent maintenance rules for the same raster. For example, short-term point clouds generated by pedestrians or other dynamic obstacles may be incorrectly fixed as static obstacles; sensing data provided by near-field blind spot sensors may cover up real obstacle evidence formed by the main sensing or semantic sources.

[0028] 3. Traditional occupancy status updates use a fixed incremental model, which fails to reflect the differences between multi-source sensing data. The reliability of obstacle observations is not constant but dynamically changes with factors such as the type of observation source, observation distance, target category (e.g., people, vehicles, static objects), and dynamic attributes. If a fixed update increment is uniformly adopted, on the one hand, it is easy to over-amplify low-confidence noisy observations, introducing false obstacles; on the other hand, it may lead to insufficient updates of high-confidence key observations, resulting in a delayed response.

[0029] 4. Map timeliness maintenance typically fails to differentiate between raster sources and stability, making it difficult to balance real-time performance, dynamic obstacle fading, and static obstacle protection. Strategies such as fixed timeout thresholds and periodic map decay cannot distinguish the data source and stability behind each raster. For example, occupancy caused by dynamic targets like pedestrians should be quickly forgotten to free up passage space, while static obstacles confirmed by long-term stable observations (such as walls and shelves) should not be cleared due to short-term observation gaps. A uniform timeout strategy can easily lead to the premature removal of genuine static obstacles (creating collision risks), the creation of dynamic obstacle residues or false occupancy markers (reducing passage efficiency), or the long-term retention of short-term false detections. Furthermore, if the entire local raster map is scanned for timeliness assessment in every frame, it significantly increases the computational burden of map updates, impacting the system's real-time performance.

[0030] See Figure 1 The diagram shows the main flowchart of a local raster map update method provided by an embodiment of the present invention, which includes the following steps: S101: Determine the historical attribute information of each grid in the latest historical local grid map; S102: Obtain the multi-source sensing data at the current moment, and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid of each sensing data. S103: Determine the credible characteristics of each of the sensing data and the intention to update the historical attribute information of the target raster, and construct update candidates for each of the sensing data by combining the source type of each of the sensing data and the target raster to which it is applied; S104: Based on each of the update candidates acting on the target raster, determine the update magnitude of the historical attribute information of the target raster; S105: Update the historical attribute information of the target raster based on the update magnitude and the update intention to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

[0031] For steps S101 and S102, this scheme is applicable to maintaining and updating various forms of local raster maps, such as sliding local maps and local cost maps, without limitation. A local raster map consists of multiple rasters, each raster m_j representing a map unit corresponding to a certain spatial location. Each raster m_j can maintain multi-dimensional attribute information, including but not limited to historical occupancy status p_j, the historical sensing data set (or source set) S_j used in the last update, the last update time t_j, and historical observation features H_j. Historical observation features H_j can be extracted from the historical sensing data set, which may include historical sensing data used by raster m_j from its creation to the most recent update, or may only include historical sensing data used in the last update; there is no limitation in this regard.

[0032] Here, the historical occupancy state p_j represents the probability that grid m_j is occupied by an obstacle, and its value is usually between 0 and 1. The source set S_j includes primary sensing sources (such as LiDAR, depth camera), near-field blind spot filling sources (such as ultrasonic radar), semantic sources (such as visual semantic segmentation results), or dynamic sources (such as multi-target tracking data).

[0033] Historical observation features H_j include, but are not limited to: historical height information (such as obstacle height), historical occupancy level (such as the number of times it has been continuously identified as occupied; for example, if it is marked as occupied in 50 consecutive frames, it indicates that it is a stable static obstacle; if it is marked as occupied in only 3 out of 50 frames, it indicates that it may be a dynamic target or sensor noise), cumulative confidence (such as the cumulative value or weighted average of the confidence scores given by the sensor or perception algorithm when observing grid m_j in each observation), number of consecutive observations (such as the number of consecutive frames effectively observed by the sensor), continuous update time (i.e., the total length of time that it is continuously updated), source reliability score (a dynamic reliability score of the sensor source (such as main perception, near-field blind filling, semantic, dynamic, etc.) that provides the observation data according to the current environmental conditions; for example, the reliability scores of visual and semantic sources are higher in daylight with sufficient light, but the reliability score of visual sources will automatically decrease at night or in strong backlight) or a combination of the above information.

[0034] For example, assuming that grid m_j has been identified as occupied by the primary sensing source in the last 10 frame updates, and the confidence level of each update is high, then the historical observation features of grid m_j can be reflected as a high "historical occupancy level" and a high "cumulative confidence level", indicating that there is a high probability that there is a stable static obstacle at this location. Conversely, if grid m_j is only marked as occupied by the dynamic source in a single frame update, and is subsequently not observed or marked as free multiple times, then the "historical occupancy level" in its historical observation features is low, and the "source reliability" may show a specific decay trend due to the rapid movement of the dynamic target, indicating that there is a high probability that there is a temporary dynamic obstacle at this location.

[0035] This scheme uses the updated local raster map M_t-1 at time t-1 as the latest historical local raster map. When the preset map update cycle is reached, multi-source sensing data (including at least one of main sensing, near-field blind spot filling, semantic, and dynamic data) at the current time t is acquired to update M_t-1, thereby obtaining the updated local raster map M_t at time t. During the update process, the corresponding source type is first configured for each acquired sensing data, and these multi-source sensing data are preprocessed, including but not limited to coordinate alignment, data validity checks, and timestamp checks.

[0036] For example, in terms of source type configuration, point cloud data acquired by LiDAR or the main depth camera can be configured as "Main Perception Source," data acquired by ultrasonic radar can be configured as "Near Field Blind Spot Completion Source," semantic segmentation results output by visual algorithms can be configured as "Semantic Source," and dynamic object position information output by multi-target tracking algorithms can be configured as "Dynamic Source." In terms of preprocessing, coordinate alignment refers to uniformly converting the perception data from each sensor coordinate system to the robot's body coordinate system or the global map coordinate system; data validity checking refers to removing flying point noise from LiDAR, abnormal jump values ​​from ultrasonic sensors, or invalid visual targets with confidence levels below a preset threshold; timestamp checking refers to verifying whether the timestamps of each data point are within a reasonable time window, and performing motion compensation or interpolation alignment on data with time deviations to ensure strict synchronization of multi-source perception data in the time dimension.

[0037] Next, the grid cells in the historical local grid map M_t-1 that match the coordinates of each sensing data point are identified, and these grid cells are designated as the target grid cells for that sensing data point. For example, for point cloud data generated by LiDAR or depth cameras, the ray casting algorithm is used to calculate all grid cells that each ray passes through in the map coordinate system, and these grid cells penetrated by the ray, as well as the grid cells where the ray ends, are identified as target grid cells. For distance information returned by ultrasonic radar, a fan-shaped area within the beam coverage is calculated based on its installation angle and ranging value, and all grid cells contained within this area are identified as target grid cells. For visual semantic segmentation results or dynamic target tracking boxes, their pixel coordinates or bounding boxes are first projected onto the map plane coordinate system, and then all grid cells covered within the projected area are identified as target grid cells.

[0038] For step S103, after determining the target raster affected by each sensing data, the update intention of the sensing data on the corresponding target raster is further determined. The update intention is used to indicate the adjustment trend of the historical attribute information of the target raster, including but not limited to: increasing occupancy status, decreasing occupancy status, writing semantic auxiliary status, writing dynamic auxiliary status, or revoking the historical status of a certain source. Specifically, it can be comprehensively judged based on the source type of the sensing data and its current observation characteristics at the target raster. Taking occupancy status as an example, this process may include at least one of the following situations: For the primary sensing source (such as LiDAR or primary depth camera), if it detects the echo signal reflected by the obstacle at the target grid, the update intention is determined to improve the historical occupancy status; if it does not detect the obstacle echo signal in the ray-penetrating area (i.e., indicating free space), the update intention is determined to reduce the historical occupancy status.

[0039] For near-field blind spot fill sources (such as ultrasonic radar), if it detects the echo signal reflected by a nearby obstacle, it determines that the update intention is to improve the historical occupancy status; otherwise, it reduces the historical occupancy status to avoid false clearing caused by physical blind spots of the sensor.

[0040] For semantic sources (such as visual semantic segmentation results and high-precision map data), if the semantic category corresponding to the target raster identified by it belongs to the preset impassable category (such as curb, building, wall), then the update intention is determined to improve the historical occupancy status; if it belongs to the preset passable category (such as lane, flat ground), then the update intention is determined to reduce the historical occupancy status, and semantic auxiliary status can also be written to assist subsequent planning.

[0041] For dynamic sources (such as multi-target tracking data, such as pedestrians, pets, etc.), if the data indicates that the spatial location of the dynamic obstacle at the current time t is within the target grid, then the update intention is determined to improve the historical occupancy status; if the data indicates that the trajectory of the dynamic obstacle at the previous time t-1 passed through the target grid but has now left, then the update intention is determined to reduce the historical occupancy status, or the historical dynamic status written by the dynamic source on the target grid can be directly revoked.

[0042] The integration of multi-source sensing data, update intentions, and descriptions is shown in Table 1 below. It should be noted that after determining the "update intention of the sensing data for the corresponding target raster," the final update magnitude is not immediately calculated, nor is the map state directly overwritten.

[0043] Table 1

[0044] Furthermore, it is necessary to construct reliable features for each piece of sensing data. These reliable features can be determined based on at least one of the following: observation distance (e.g., distance from the camera), number of observations (e.g., the number of projected points within the target grid, such as obstacles), observation density (e.g., the density of data points distributed within the target grid and its adjacent grid areas; higher observation density typically indicates the presence of a continuous physical surface at that location (e.g., walls, shelves)), target category (e.g., the category of identified obstacles), target velocity (e.g., the movement speed of obstacles), source confidence (e.g., the overall reliability score of the specific sensor or sensing algorithm module providing the observation data under current environmental conditions), and algorithm confidence (e.g., the probability value output by the detection model, which can also be set as an empirical value). In terms of specific calculation methods, numerical accumulation, weighted summation, mean calculation, or nonlinear mapping can be used; this application does not impose any restrictions on these methods. Therefore, the reliable features intuitively reflect the overall reliability of the sensing data; the higher the score, the more reliable the data.

[0045] To avoid coverage conflicts caused by directly writing multi-source sensing data into the historical local raster map M_t-1, this invention does not directly use the multi-source sensing data at the current time t to update the historical local raster map M_t-1. Instead, it constructs corresponding update candidates c_i to generate an update candidate set C_t. Update candidate c_i = (G_i, s_i, a_i, ... _i, u_i), where G_i represents the raster or region being acted upon, s_i represents the source type, a_i represents the base weight of the source type, _i represents a trusted feature, and u_i represents an update intent.

[0046] This approach unifies perceptual data from different sources, such as primary sensing, near-field blind spot filling, semantic, and dynamic data, into update candidates for the next stage of the fusion decision-making process. During this process, even if perceptual data from multiple sources affects the same target raster, the coverage relationship between the data is not directly determined in the update candidate construction stage. Instead, the source type and reliable features are fully preserved, and the subsequent adaptive update model performs comprehensive evaluation and processing. Furthermore, the candidate data sources include, but are not limited to, classification maps, depth maps, point clouds, distance points, polygonal regions, raster regions, object boxes, object velocities, or auxiliary probability maps; there are no restrictions on these sources.

[0047] For steps S104 and S105, the update magnitude characterizes the actual influence strength of the update candidate c_i on the target grid m_j. In an optional implementation, the update magnitude of the update candidate c_i on the target grid m_j can be expressed as:

[0048] Where s_i represents the source type, a_i represents the base weight of the source type, _i represents the trusted feature, u_i represents the update intention, and g( _i,H_j) represents the adaptive factor determined by both reliable features and historical observation features. `clip` indicates that the update magnitude is limited to a preset upper and lower bound. This adaptive update model can be constructed using log-odds, Bayesian probability, confidence scores, cost values, or learning models; the adaptive factor can be determined by distance, quantity, source, category, dynamic attributes, or a noise model, without restriction.

[0049] Then, the actual update magnitudes of the historical attribute information of the target raster m_j by multiple update candidates c_i acting on the same target raster m_j are merged, and the update intentions of each update candidate c_i on the target raster m_j are combined to obtain the total update magnitude.

[0050]

[0051] Where C_j represents the set of update candidates c_i acting on the same target raster m_j, and p_j-1 represents the historical occupancy status in the historical attribute information of the target raster m_j at the previous time step. clip indicates that the updated occupancy status is restricted to a preset upper and lower limit.

[0052] As an optimization method, the update magnitude can be calculated. Further optimization is performed. Specifically, the current observation feature is extracted from the sensing data applied to the target grid m_j, and it is determined whether the historical observation feature H_j of the target grid m_j includes the current observation feature. If it does, the update magnitude of the update candidate is increased according to a preset enhancement strategy (e.g., increase by 0.1); if it does not, the update magnitude of the update candidate is decreased according to a preset decay strategy (e.g., decrease by 0.1).

[0053] For example, if target grid m_j has historically consistently identified a static wall with a height of approximately 1.8 meters, and the current observation feature also identifies a static wall of the same height and category, then the current observation feature indicates a reconfirmation of historical observation features, with high reliability. The update amplitude should be appropriately increased to strengthen the occupancy probability of target grid m_j. Conversely, if target grid m_j has historically been identified as a static wall, but the current observation feature suddenly identifies it as a fast-moving pedestrian, this indicates that the current observation feature may be due to sensor noise, algorithmic false detection, or sudden environmental changes. The update amplitude should be appropriately reduced to suppress potential false updates and prevent stable historical states from being easily overturned by a single abnormal observation.

[0054] Furthermore, for update candidate c_i whose update intention is to clear, if historical observation features H_j indicate that the target grid m_j has recently had highly credible obstacle evidence, the system automatically reduces the clearing intensity of the update candidate c_i, such as downgrading strong clearing to weak clearing; conversely, if the historical observation features H_j of the target grid m_j are outdated or the evidence accumulation is insufficient, strong clearing is allowed. This mechanism is particularly crucial when there is a conflict between near-field blind spot observations and historical obstacle evidence, effectively preventing the erroneous clearing of real static obstacles caused by near-field sensor false alarms or physical blind spots, thereby significantly improving the reliability of local grid maps and the operational safety of the robot.

[0055] When multiple update candidates c_i exist for a target grid m_j, the system does not require data from different sources to overlap. Instead, it can apply data to the primary occupancy state, semantic auxiliary state, dynamic auxiliary state, or source marker separately. After fusion, the system writes the final result into the same local grid map and synchronously updates the historical occupancy state p_j of the target grid m_j, the historical sensing data set S_j used in the last update, the last update time t_j, and the historical observation features H_j. The updated local grid map can be used in path planning, obstacle avoidance control, or navigation cost map generation modules in robot navigation scenarios without restriction.

[0056] It is understood that the specific values ​​listed in the above hypothetical description of local raster map updates are merely illustrative examples and not restrictive. In practice, other values ​​can be set according to requirements. Similarly, the specific values ​​listed in the subsequent embodiments are also just examples and are not the only limitation.

[0057] The method provided in the above embodiments, after acquiring the multi-source sensing data of the current frame, does not directly overlay it. Instead, it analyzes the update intention of each sensing data point for the matched target raster, and generates update candidates by combining the source type and credibility features of each sensing data point. This decouples the multi-source sensing data into independent update intentions, effectively avoiding the coverage conflicts and semantic loss problems caused by directly overlaying multi-source sensing data. At the same time, based on the multiple update candidates acting on the same target raster, the update magnitude is determined separately, realizing adaptive differentiated updates for sensing data from different sources. This suppresses the excessive amplification of low-confidence noise while ensuring sufficient response of high-confidence observations, thereby significantly improving the update reliability, noise resistance, and real-time performance of local raster maps in high-frequency navigation scenarios.

[0058] See Figure 2 The diagram illustrates an optional local raster map update method according to an embodiment of the present invention, including the following steps: S201: Determine the unupdated grids in the local grid map that have not been updated based on the multi-source sensing data, input the historical sensing data set and historical observation features of the unupdated grids into a preset forgetting period model, and output the forgetting period of the unupdated grids; S202: Calculate the time difference between the current time and the last update time of the unupdated raster, and perform forgetting processing on the unupdated raster in response to the time difference being greater than the forgetting period of the unupdated raster.

[0059] After updating the historical local raster map based on the current frame observation, rasters that have not been updated based on the current frame observation (i.e., unupdated rasters) also need to be maintained. Specifically, the forgetting period of this invention is not a uniform fixed timeout threshold, but is jointly determined based on the historical sensing data set S_j and historical observation features H_j used in the last update of the unupdated raster. In an optional implementation, the forgetting period of the unupdated raster m_j can be determined by a forgetting period model:

[0060] If the time difference between the current time t and the last update time t_j of the unupdated raster m_j exceeds the forgetting period T_j, then forgetting processing is performed on the unupdated raster m_j.

[0061]

[0062] The forgetting process Forget(m_j) performed on the unupdated raster m_j can include, but is not limited to: reducing the occupancy probability, canceling semantic auxiliary states, clearing dynamic auxiliary states, resetting historical information from a specific source, or partially resetting the raster's attributes. The system matches a corresponding forgetting strategy based on the historical observation features H_j of the unupdated raster m_j, and dynamically adjusts the raster's historical attribute information (such as occupancy probability, historical feature values, etc.) according to this forgetting strategy. Specifically, for states with weak historical observation features (such as a small number of consecutive observations), low source reliability, or only short-term occurrences, the system matches a forgetting strategy with a shorter forgetting period or a larger decay strength to quickly clear its historical attribute information; for states with strong historical observation features and continuous stability, the system matches a forgetting strategy with a longer forgetting period or a smaller decay strength to maintain the stability of its historical attribute information.

[0063] For example, if historical observation features H_j indicate that the unupdated grid m_j was previously identified as a dynamic obstacle with a large movement speed, the system matches an aggressive forgetting strategy, which quickly eliminates the dynamic obstacle residue or false occupancy mark left after the dynamic target leaves by significantly reducing its occupancy probability or rapidly clearing the relevant historical observation features. Conversely, if historical observation features H_j indicate that the unupdated grid m_j has continuously received stable observations of static obstacles over a period of time (such as many consecutive observations and high cumulative confidence), the system matches a conservative forgetting strategy, which avoids the grid from being mistakenly cleared due to short-term observation loss (such as being temporarily occluded), thereby achieving long-term protection for stable static obstacles.

[0064] As an optimization method to reduce computational burden, the local raster map can be divided into multiple sub-regions, with each maintenance cycle processing only one or more of these sub-regions. For example, these sub-regions can be configured with a sorting order, and the corresponding sub-regions can be polled and updated sequentially in each update cycle according to this order. This regional processing mechanism effectively avoids the computational overhead of performing a full scan of the entire local raster map every frame. Simultaneously, through batch and gradual maintenance, outdated dynamic target artifacts, short-term false detection noise, and low-confidence historical states can smoothly and gradually fade away over multiple cycles, thus maintaining the stability of the local environment while ensuring the real-time nature of map updates. As an optional implementation, in addition to this regional polling mechanism, other mechanisms such as time-slice queues, update-time-priority queues, random sampling, spatial hashing, or maintaining only the most recently modified region can also be used; there are no restrictions on the specific implementation.

[0065] The method provided in the above embodiments introduces a forgetting cycle model, dynamically calculating a personalized forgetting cycle based on the historical sensing data set and historical observation features of the unupdated raster, breaking away from the traditional "one-size-fits-all" timeout decay strategy with fixed timeouts. This enables the system to quickly clear residual false occupancy markers for dynamic targets, while providing long-term protection for stable static obstacles to avoid accidental removal due to short-term occlusion. Thus, while ensuring the timeliness and security of local raster maps, it effectively reduces the computational burden caused by indiscriminate scanning of the entire map.

[0066] In summary, this invention constructs a local raster map update framework based on "source hierarchical design, adaptive update, and adaptive forgetting," as detailed in [link to documentation]. Figure 3 As shown, it can solve existing problems such as "unreliable near-field region updates due to near-field blind spots, source conflicts and semantic loss caused by simple superposition of multiple sources, fixed update amplitude failing to reflect differences in observation reliability, and fixed forgetting strategies failing to distinguish between retention and removal based on source and historical observation characteristics": 1. Provide a source-layered map candidate update mechanism: After acquiring multi-source sensing data at the current moment, each sensing data is represented as a unified update candidate, rather than being directly superimposed into independent layers. Each candidate carries the source type, update intent, and reliability characteristics of the sensing data, enabling sensing data acting on the same target raster to be collaboratively updated within the same local raster map, and providing a basis for subsequent adaptive updates and adaptive forgetting.

[0067] 2. Candidate-based adaptive update amplitude strategy: Different update candidates have different source types, update intentions, and reliable features. Combining the historical observation features of the target raster, the update amplitude of the update candidate on the historical attribute information of the target raster is determined, and then the total update amplitude is obtained by fusion. This achieves raster adaptive update and avoids the problem that the existing fixed update amplitude ignores the "actual impact of different source sensing data on the update of raster attribute information".

[0068] 3. Adaptive Forgetting Strategy Based on Update Candidates. Based on the historical sensing data set used in the last raster update, the last update time, and historical observation characteristics, an adaptive forgetting period is determined to replace the uniform timeout decay strategy, and forgetting strategies such as retention, decay, or clearing are matched. Furthermore, partitioned maintenance reduces the computational burden. This strategy allows short-term false detections and dynamic impacts to dissipate promptly, while retaining effective map information with strong historical observation characteristics.

[0069] See Figure 4 The diagram shows the main modules of a local raster map updating device 400 provided in an embodiment of the present invention, including: Module 401 is used to determine the historical attribute information of each grid in the latest historical local grid map; Matching module 402 is used to acquire multi-source sensing data at the current moment and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid of each sensing data. The construction module 403 is used to determine the credible characteristics of each of the sensing data and the intention to update the historical attribute information of the target raster, and to construct update candidates for each of the sensing data by combining the source type of each of the sensing data and the target raster to which it is applied. Calculation module 404 is used to determine the update magnitude of the historical attribute information of the target raster based on each of the update candidates acting on the target raster; The update module 405 is used to update the historical attribute information of the target raster based on the update magnitude and the update intention, to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

[0070] In the apparatus of this invention, the construction module 403 is used for: The reliable feature is determined based on at least one of the following in the perception data: observation distance, number of observations, observation density, target category, target speed, source confidence, and algorithm confidence. Based on the source type of each of the sensing data and the current observation characteristics at the target grid, determine the intention of each of the sensing data to update the historical attribute information of the target grid.

[0071] In the apparatus of this invention, the multi-source sensing data includes at least one of main sensing data, near-field blind spot sensing data, semantic sensing data, and dynamic sensing data, and the process of determining the update intention includes at least one of the following: When the main sensing data indicates that an echo signal reflected from an obstacle has been detected, it is determined that the update intention is to improve the historical attribute information; otherwise, it is determined that the update intention is to reduce the historical attribute information. When the near-field blind spot sensing data indicates that an echo signal reflected from a nearby obstacle has been detected, it is determined that the update intention is to improve the historical attribute information; otherwise, it is determined that the update intention is to reduce the historical attribute information. When the semantic category identified by the semantic perception data belongs to a preset inaccessible category, the update intention is determined to improve the historical attribute information; otherwise, the update intention is determined to reduce the historical attribute information. When the dynamic sensing data indicates that there is a dynamic obstacle in the target grid, the update intention is determined to be to improve the historical attribute information; when the historical attribute information indicates that there is a dynamic obstacle in the target grid, but the dynamic sensing data indicates that there is no dynamic obstacle in the target grid, the update intention is determined to be to reduce the historical attribute information.

[0072] In the device of this invention, the historical attribute information includes historical observation features, which are determined by the historical sensing data set corresponding to each grid. The calculation module 404 is used to: input the historical observation features of the target grid and the information of each update candidate other than the target grid into a preset update model, and output the update magnitude of each update candidate on the historical attribute information of the target grid; The update module 405 is configured to: fuse the update magnitude and update intention of the historical attribute information of the target raster based on the update candidates to obtain a total update magnitude; update the historical attribute information of the target raster based on the total update magnitude to obtain the current attribute information of the target raster.

[0073] In the apparatus of the present invention, the calculation module 404 is further configured to: Determine whether the historical observation features of the target grid include the current observation features of each of the sensing data at the target grid; If the judgment result indicates that the update magnitude of the historical attribute information of the target raster by each of the update candidates is increased according to the preset boosting strategy; otherwise, the update magnitude of the historical attribute information of the target raster by each of the update candidates is reduced according to the preset decay strategy.

[0074] In the device of this invention, the historical attribute information includes historical observation features, the historical sensing data set used in the last update, and the last update time; The device also includes a forgetting module for: Identify the unupdated graticules in the local graticule map that have not been updated based on the multi-source sensing data, input the historical sensing data set and historical observation features of the unupdated graticules into a preset forgetting period model, and output the forgetting period of the unupdated graticules; Calculate the time difference between the current time and the last update time of the unupdated raster, and perform forgetting processing on the unupdated raster in response to the time difference being greater than the forgetting period of the unupdated raster.

[0075] In the apparatus of this invention, the forgetting module is used for: Based on the historical observation features of the unupdated raster, a forgetting strategy is matched to adjust the historical attribute information of the unupdated raster through the forgetting strategy, so as to obtain the current attribute information of the unupdated raster.

[0076] Furthermore, the specific implementation details of the device described in the embodiments of the present invention have been described in detail in the above-described method, so the details will not be repeated here.

[0077] Figure 5 An exemplary system architecture 500 to which embodiments of the present invention can be applied is shown, including terminal devices 501, 502, 503, network 504, and server 505 (this is merely an example).

[0078] Terminal devices 501, 502, and 503 can be various electronic devices with displays and web browsing capabilities, and have various communication client applications installed. Users can use terminal devices 501, 502, and 503 to interact with server 505 via network 504 to receive or send messages, etc.

[0079] Network 504 is a medium used to provide a communication link between terminal devices 501, 502, 503 and server 505. Network 504 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0080] Server 505 can be a server providing various services, such as a backend management server supporting shopping websites browsed by users using terminal devices 501, 502, and 503 (this is just an example). The backend management server can analyze and process received data such as product information query requests, and feed back the processing results (e.g., target push information, product information—this is just an example) to the terminal devices. It should be noted that the method provided in this embodiment of the invention is generally executed by server 505, and correspondingly, the apparatus is generally set in server 505.

[0081] It should be understood that Figure 5 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0082] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer system 600 suitable for implementing a terminal device of the present invention. Figure 6 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0083] like Figure 6As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the system 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0084] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed into storage section 608 as needed.

[0085] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the functions defined above in the system of this invention.

[0086] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0087] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0088] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor can be described as including a determining module, a matching module, a building module, a calculating module, and an updating module. The names of these modules do not necessarily limit the module itself; for example, the matching module can also be described as a "locating module."

[0089] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to perform any of the aforementioned local raster map update methods.

[0090] The computer program product of the present invention includes a computer program that, when executed by a processor, implements the local grid map update method in the embodiments of the present invention.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for updating a local raster map, characterized in that, include: Determine the historical attribute information of each grid in the latest historical local grid map; Acquire multi-source sensing data at the current moment, and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid for each sensing data. Determine the credible characteristics of each of the sensing data and the intention to update the historical attribute information of the target raster, and construct update candidates for each of the sensing data by combining the source type of each of the sensing data and the target raster to which it is applied; Based on each of the update candidates acting on the target raster, the update magnitude of the historical attribute information of the target raster is determined; Based on the update magnitude and the update intention, the historical attribute information of the target raster is updated to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

2. The method according to claim 1, characterized in that, The determination of the credible features of each of the perceived data and the intention to update the historical attribute information of the target raster includes: The reliable feature is determined based on at least one of the following in the perception data: observation distance, number of observations, observation density, target category, target speed, source confidence, and algorithm confidence. Based on the source type of each of the sensing data and the current observation characteristics at the target grid, determine the intention of each of the sensing data to update the historical attribute information of the target grid.

3. The method according to claim 2, characterized in that, The multi-source sensing data includes at least one of main sensing data, near-field blind-filling sensing data, semantic sensing data, and dynamic sensing data. The process of determining the update intention includes at least one of the following: When the main sensing data indicates that an echo signal reflected from an obstacle has been detected, it is determined that the update intention is to improve the historical attribute information; otherwise, it is determined that the update intention is to reduce the historical attribute information. When the near-field blind spot sensing data indicates that an echo signal reflected from a nearby obstacle has been detected, it is determined that the update intention is to improve the historical attribute information; otherwise, it is determined that the update intention is to reduce the historical attribute information. When the semantic category identified by the semantic perception data belongs to a preset inaccessible category, the update intention is determined to improve the historical attribute information; otherwise, the update intention is determined to reduce the historical attribute information. When the dynamic sensing data indicates that there is a dynamic obstacle in the target grid, the update intention is determined to be to improve the historical attribute information; when the historical attribute information indicates that there is a dynamic obstacle in the target grid, but the dynamic sensing data indicates that there is no dynamic obstacle in the target grid, the update intention is determined to be to reduce the historical attribute information.

4. The method according to claim 1 or 2, characterized in that, The historical attribute information includes historical observation features, which are determined by the historical sensing data set corresponding to each of the grids. The step of determining the update magnitude of the historical attribute information of the target grid based on each of the update candidates acting on the target grid includes: inputting the historical observation features of the target grid and the information of each update candidate other than the target grid into a preset update model, and outputting the update magnitude of each update candidate on the historical attribute information of the target grid; The step of updating the historical attribute information of the target raster based on the update magnitude and the update intention to obtain the current attribute information of the target raster includes: fusing the update magnitude and update intention of the historical attribute information of the target raster by each update candidate to obtain a total update magnitude; and updating the historical attribute information of the target raster based on the total update magnitude to obtain the current attribute information of the target raster.

5. The method according to claim 4, characterized in that, After outputting the update magnitude of the historical attribute information of each of the update candidate pairs for the target raster, the method further includes: Determine whether the historical observation features of the target grid include the current observation features of each of the sensing data at the target grid; If the judgment result indicates that the update magnitude of the historical attribute information of the target raster by each of the update candidates is increased according to the preset boosting strategy; otherwise, the update magnitude of the historical attribute information of the target raster by each of the update candidates is reduced according to the preset decay strategy.

6. The method according to claim 1, characterized in that, The historical attribute information includes historical observation characteristics, the historical sensing data set used in the last update, and the last update time; The process of generating a local raster map at the current moment also includes: Identify the unupdated graticules in the local graticule map that have not been updated based on the multi-source sensing data, input the historical sensing data set and historical observation features of the unupdated graticules into a preset forgetting period model, and output the forgetting period of the unupdated graticules; Calculate the time difference between the current time and the last update time of the unupdated raster, and perform forgetting processing on the unupdated raster in response to the time difference being greater than the forgetting period of the unupdated raster.

7. The method according to claim 6, characterized in that, The forgetting process for the unupdated raster includes: Based on the historical observation features of the unupdated raster, a forgetting strategy is matched to adjust the historical attribute information of the unupdated raster through the forgetting strategy, so as to obtain the current attribute information of the unupdated raster.

8. A local raster map updating device, characterized in that, include: The determination module is used to determine the historical attribute information of each grid in the latest historical local grid map; The matching module is used to acquire the multi-source sensing data at the current moment and determine the grid in the historical local grid map that matches the coordinates of each sensing data as the target grid of each sensing data. A construction module is used to determine the credible characteristics of each of the sensed data and the intention to update the historical attribute information of the target raster, and to construct update candidates for each of the sensed data by combining the source type of each sensed data and the target raster to which it is applied; A calculation module is used to determine the update magnitude of the historical attribute information of the target raster based on each of the update candidates acting on the target raster; The update module is used to update the historical attribute information of the target raster based on the update magnitude and the update intention, to obtain the current attribute information of the target raster, so as to generate a local raster map at the current moment.

9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images