A global unique positioning method for symmetric passageway of unmanned vehicle

Abstract

The application discloses a kind of unmanned vehicle symmetric passage global unique positioning method, belong to mobile robot navigation and positioning technical field.The method is by with passage level semantics as center to build positioning constraint, and with the direction of passage identification, node identification and traffic composition directed triple, and with anchor point coding explicit four binding, form only in node trigger identity check field, avoid generating new identity in long straight section;For each mobile type, respectively establish state branch and interval constraint, with mobile label and stability degree gate control extrapolation and switching;Only around current passage and adjacent passage of shared boundary establish minimum scale multi-hypothesis;According to passage boundary hard constraint and semantic conflict driving survival degree, maintain a small amount of continuous feasible identity;Topological reachability, mobile alignment, orientation, timing window and geometric endorsement are carried out at node / anchor point Single-pass verification, and with multiplicative gate and one-time solidification determine dominant identity, prohibit backflow.

Description

Technical Field

[0001] This invention relates to the field of mobile robot navigation and positioning technology, specifically to a globally unique positioning method for unmanned vehicles in symmetrical channels. Background Technology

[0002] In existing technologies, AGV positioning for indoor passageway scenarios often employs geometric matching and map relocalization methods based on LiDAR or vision. A common approach is to construct a global grid map or feature map, maintaining pose consistency through ICP / NDT, loop closure detection, and global optimization. In areas with repetitive structures, multi-sensor fusion filtering with tightly coupled odometer and IMU is used to reduce short-term drift. For large-scale deployments, systems often introduce sparse markers or low-density anchor points such as QR codes, UWB / Bluetooth, etc., triggering relocalization at specific locations to correct accumulated errors and update the path status.

[0003] In terms of functional division, the localization front end is typically responsible for point cloud preprocessing, feature extraction, and local matching, outputting a relative pose with continuous mileage. The back end uses graph optimization or extended Kalman filter frameworks to fuse odometry, IMU, and anchor point observations to achieve a smooth trajectory under global reference. To adapt to channelized environments, some solutions use topological maps to represent traversable connectivity, with nodes representing intersections or key points and edges representing passageways. The navigation module plans paths based on the topology and drives local tracking. Anchor point-related methods generally associate observations with map anchor points by recognizing encoded information, performing relocalization and correcting drift after entering the recognition range; in sparse anchor point regions, continuity is maintained by inertial navigation and wheel speed sensors. For vehicle maneuvers, common implementations use fixed state machines to describe straight-ahead, turning, and U-turn modes, switching based on threshold judgments, and enhancing alignment accuracy at turning nodes using angular velocity and geometric matching. The overall process emphasizes geometric consistency and the stability of measurement fusion, combined with periodic correction of sparse anchor points, to complete navigation and localization in channelized environments.

[0004] like Figure 4 As shown, existing technologies employ a geometrically hypothetical matching method, which, when triggered by nodes, leads to anchor points not forming closed loops and is prone to premature / lagging associations. This amplifies ambiguity, causing multi-branch diffusion and an increase in the number of hypotheses, making it prone to channel jumps and drifts. Furthermore, it is difficult to achieve unique and stable post-correction outputs and backtracking fine-tuning. Therefore, the problems of positioning drift, channel jumps, and high anchor point dependence in existing technologies restrict the large-scale application of equipment. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a globally unique positioning method for symmetrical channels of unmanned vehicles, in order to solve the problems of positioning drift, channel jump, and high anchor point dependence in the existing technology.

[0006] To achieve the above objectives, the method of the present invention includes the following steps:

[0007] S1. Assign a unique channel identifier to each channel in the site where the symmetrical channel is located; assign a unique node identifier to the key nodes of the site, including the entrance, fork decision points, and the end; map the passage direction of the site to a directed attribute; and form a channel-level semantic skeleton using a triplet of the directed relationship between the channel identifier, node identifier, and passage direction; configure a unique coded anchor point in the visible sub-area of ​​each key node, and explicitly bind the anchor point to the channel identifier, node identifier, and passage direction to form an anchor point semantic quadruple; establish a channel semantic graph with the anchor point as the graph node and the passable segment as the directed edge, and limit the channel identity verification to only be performed at the triggered graph node;

[0008] S2. Define the maneuver types of unmanned vehicles in a symmetrical channel. From the set of directed edges and key nodes of the channel semantic graph, set allowed subsets of edges or nodes for each type of maneuver and limit attitude constraints. Establish independent state branches for each type of maneuver and output the pose evolution and branch confidence for each branch. Select the maneuver type corresponding to the branch with the highest branch confidence as the current maneuver and calculate the stability of the current maneuver. Write the maneuver identifier of the current maneuver and the stability identifier of the stability into the current edge identifier or node identifier of the channel semantic graph to form an instruction on how to move along the channel at this moment.

[0009] S3. Generate channel identity hypotheses around the current channel in which the unmanned vehicle is located and all adjacent channels that share a boundary and are semantically reachable. For each channel identity hypothesis, advance only along the directed edge set of the corresponding channel to obtain the pose advancement result of each channel identity hypothesis. Based on the current maneuver identifier obtained in S2 and the corresponding limited attitude constraints, restrict the attitude advancement of the unmanned vehicle, and use the channel geometric boundary as a hard constraint to obtain a continuous trajectory consistent with the channel semantics under the current maneuver of the unmanned vehicle. By checking the consistency between the pose advancement result of each channel identity hypothesis and the continuous trajectory consistent with the channel semantics, identify semantic conflicts of the channel identity hypotheses, and retain only a first set of channel identity hypotheses that pass the consistency check and meet the continuity requirements as the current valid set. Use this current valid set as the sole candidate pool for anchor point events in the subsequent S4 step.

[0010] S4. Determine whether the unmanned vehicle has entered the node's neighborhood based on its projection on the channel semantic graph, and filter visible candidate anchor points by combining the current maneuver tag, the node's allowed maneuver, and the anchor point's field of view; set a maneuver consistency window for entry and exit based on the cumulative mileage along the edge, and impose temporal constraints on the candidate anchor points; prune each channel identity hypothesis and candidate anchor point in the unique candidate pool from step S3 by single-pass verification based on topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement, and select a second set number of association pairs with a comprehensive score of multiplicative gating and additive endorsement; implement serial adjudication of dominant determination, semantic exclusion, path integrity, and maneuver alignment based on the associations, perform one-time weighted solidification of dominant hypotheses, and reduce the weight of non-dominant hypotheses to a non-returnable state; output the solidified channel identifier and pose as the current identity, and bind this identity to the node.

[0011] Its beneficial effects are as follows: This invention publishes the channel semantic graph as the sole basis for channel identity, explicitly allowing channel identity verification only within triggered graph nodes, and strictly prohibiting the generation of new identities in non-graph node regions, thus avoiding channel jumps and identity drifts in geometrically similar long straight segments. When generating channel identity assumptions, this invention generates channel identity assumptions only for channels that may be entered, by using the current channel and all adjacent channels that share a boundary and are semantically reachable. That is, it limits the channel identity assumptions to the minimum necessary scale, avoiding unconstrained expansion in geometrically similar regions. This invention performs single-pass verification of topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement at nodes / anchors, and determines the dominant identity with multiplicative gating and one-time solidification, prohibiting backflow.

[0012] Furthermore, step S2 also includes: setting the stability identifier as a switching threshold driving quantity; not triggering maneuver switching when the stability does not meet the preset range and the duration is less than the preset first time; performing a switch from the current maneuver to another maneuver when the stability does not meet the preset range and the duration is greater than the preset second time; limiting the switching operation to a subset of nodes or boundary conditions allowed by the channel semantic graph; and writing the switching result back to the maneuver label and stability identifier; wherein the preset second time is greater than or equal to the preset first time.

[0013] This invention limits switching operations to a subset of nodes or boundary conditions allowed by the channel semantic graph, avoiding misjudgments of turning in or out due to minor disturbances in long straight sections. The switching results are written back to the maneuver semantic label and stability identifier, ensuring that long straight sections maintain a single maneuver propulsion, and achieving clear and traceable maneuver transitions at key nodes.

[0014] Further, in step S3, the pose advancement result of each channel identity hypothesis is obtained by advancing only along the corresponding channel directed edge set. This includes: binding a unique channel identifier to each channel identity hypothesis, binding an independent pose trajectory container identifier and mileage boundary identifier to each channel identity hypothesis, constraining the longitudinal advancement of each channel identity hypothesis to a subset of directed edges with the same direction of travel, and updating the pose trajectory container identifier only within the channel directed edges to obtain the pose advancement result of each channel identity hypothesis.

[0015] Further, in step S3, identifying semantic conflicts of channel identity assumptions by checking the consistency between the pose advancement results of each channel identity assumption and the continuous trajectory consistent with the channel semantics includes: by checking the consistency between the pose advancement results of each channel identity assumption and the continuous trajectory consistent with the channel semantics, cases of non-node turning in or out, contradicting the direction of travel, or crossing the channel boundary are marked as semantic conflicts.

[0016] Further, in step S3, retaining only a first set number of channel identity assumptions that pass the consistency check and meet the continuity requirements as the current valid set includes: setting a survival indicator for each channel identity assumption; lowering the survival indicator corresponding to a channel identity assumption that is identified as having a semantic conflict; raising the survival indicator corresponding to the remaining channel identity assumptions; forming the current valid set of channel identity assumptions that pass the consistency check and whose survival indicators reach a set threshold; sorting each channel identity assumption in the current valid set according to the satisfaction level of the channel identifier and trajectory continuity; and selecting the first set number of channel identity assumptions with the highest satisfaction level in the satisfaction ranking as the current valid set.

[0017] This invention constructs and advances a minimum necessary set of multiple assumptions about channel identity through step S3, which facilitates the accuracy and efficiency of subsequent selection based on the channel identity assumptions determined in S3.

[0018] Furthermore, in step S4, the single-pass verification of each channel identity assumption and candidate anchor point in the unique candidate pool of step S3 is pruned according to topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement, including:

[0019] Using topological reachability verification as the entry point, the current pose of each channel identity assumption is projected onto its set of directed edges, limiting entry into the node neighborhood only under the conditions of not crossing the channel boundary and not crossing closed edges; channel identity assumptions that do not meet the requirements are directly eliminated.

[0020] The maneuver alignment verification is performed under the premise of topological reachability. The current maneuver is aligned with the set of maneuvers allowed by the node. Only the channel identity assumption that matches one of the maneuver types (turn in, turn out, go straight or turn around) is retained to avoid interpreting the slow-moving and slipping of the maneuver types as channel switching.

[0021] The orientation verification is performed by mirroring the symmetrical layout. Each channel identity assumption is checked for consistency between the visible side and the anchor installation side under the current flight direction. Only the combination of channel identity assumption and visible candidate anchor point that is visible in the visible side corner domain is retained.

[0022] The timing window verification is performed based on the mileage relationship along the edge. The entry and exit windows of each combination and its candidate anchor point are aligned. Combinations not in the window are marked as unassociative.

[0023] Geometric endorsement checks will only be used for final confirmation after the first four levels have been passed.

[0024] This invention uses geometric endorsement checks only for final confirmation after passing the first four levels, avoiding the use of geometric similarity as an early decision and preventing being misled under strong symmetry conditions. The above five verifications are implemented in a one-time pass manner, ensuring that each hypothesis exits immediately if any early constraint is not met, without entering subsequent checks, and avoiding the introduction of irrelevant branches in long straight segments.

[0025] Furthermore, in step S4, based on the association, a series of decisions are made regarding dominance determination, semantic exclusion, path integrity, and maneuver alignment. This includes a one-time weighted solidification of the dominant hypothesis and a reduction of the non-dominant hypothesis to a non-returnable state, including:

[0026] The semantic exclusion rule is implemented first. For the combination of the pruned channel identity assumption and the visible candidate anchor point, the consistency check of the channel identity assumption, node identity, and travel direction of the visible candidate anchor point is performed. Any inconsistency is immediately marked as elimination.

[0027] The path integrity rule is executed after exclusion. The sequence of traveled edges assumed by the remaining channel identity is checked against the allowed inbound and outbound edges of the key nodes. Entry from non-allowed inbound edges or passing through in reverse is not allowed. If the condition is not met, the node is directly eliminated.

[0028] The maneuver alignment rule is used as the third adjudication, aligning the current maneuver from step S2 with the set of allowed maneuvers of the node, and directly eliminating channel identity assumptions that need to be associated under non-allowed maneuvers.

[0029] This invention prioritizes the semantic exclusion rule, executes the path integrity rule after exclusion, and uses the maneuver alignment rule as a third ruling to eliminate false turns induced by slow-moving vehicles and skidding. These three rules are used only as strong pruning to preserve the dominant assumption, ensuring convergence at each node without retaining buffer branches.

[0030] This invention achieves selective strong constraint association and one-time solidification at key nodes and anchor points through step S4, which completes the accurate transformation of the channel identity to determine the dominant identity and prohibit backflow.

[0031] Furthermore, it also includes:

[0032] S5. Implement channel locking and controlled spare tire strategy in long straight sections with sparse anchor points: Gating the channel identity assumption based on the stability of the maneuver, locking the propulsion to the set of directed edges of the current channel under stable cruise marking; generating only a unique spare tire assumption that shares the boundary with the current channel and is in the same direction, and shrinking its allowed maneuver to slow passing and slight slippage; suppressing identity switching when not entering the node neighborhood, the trajectory under the current channel is continuously accumulated while the unique spare tire assumption retains only the minimum necessary segment; applying piecewise function constraints such as adjacency filtering, restricted propulsion, dual capping of mileage and time, and cross-edge or reverse trend monitoring to the unique spare tire assumption, and clearing it in real time when entering the node neighborhood, reaching the cap, or showing a tendency to cross the boundary; when reaching the preset trigger point, calling step S4 to make a decision and restoring channel locking and controlled propulsion of the single spare tire.

[0033] Furthermore, it also includes:

[0034] S6. Implement the task, grant closed-loop authorization for the channel, and conduct on-site verification: Compare the mapping between the task point and the channel identifier with the current identity output in step S4. If they match, enter the queue for authorization. If they do not match, trigger on-site verification. Prioritize selecting the node corresponding to the nearest visible anchor point or planning to the nearest strongly constrained node under the direction of travel and the confirmed direction. After calling step S4 to complete the identity convergence again, return to the comparison and authorize execution when the ternary confirmation is consistent.

[0035] Furthermore, it also includes:

[0036] S7. Execute freeze and rollback convergence under continuous anomalies where the channel identity is uncertain: When the cumulative duration or mileage of the abnormal maneuver meets the threshold and no strong constraint is triggered at the node, freeze the set of channel identity assumptions as read-only, prohibiting expansion and switching; advance along the travel direction and the confirmed direction with set parameters to the visible anchor point or strong constraint node, call step S4 to complete convergence, retain the dominant assumption consistent with the convergence result and clear inconsistent branches, reset the freeze flag and continue to advance on the convergence channel.

[0037] In long straight sections, this invention uses a dominant channel to lock the advance and only allows the single adjacent spare tire in the same direction to exist for a short time under restricted maneuverability, mileage / duration double caps, and trend monitoring; when entering the switching preparation window or reaching the trigger point, the node strong constraint adjudication is called, thus realizing the controlled transfer and convergence of identity.

[0038] The method of the present invention has the following advantages:

[0039] Compared to existing technologies that rely on geometric matching and periodic anchor point relocation, this application uses channel semantics as the unique identity decision domain. It employs a triplet of channel ID, node ID, and travel direction, along with a quadruple of anchor point encoding, to bind identities. Verification is triggered only at nodes, reducing identity generation and drift in long straight segments. For maneuver estimation, an IMM (Interacting Multiple Model Filtering) is used to construct the maneuver semantic flow, with stability-gated switching, ensuring that extrapolation is strictly constrained by travel direction and node permission. In identity management, only the minimum necessary multiple assumptions for the current channel and adjacent channels on the shared boundary are maintained. At nodes / anchor points, the dominant identity is determined through multiplicative gating and one-time solidification of topological reachability, maneuver alignment, orientation, and temporal windows, suppressing backflow expansion. Long straight segments employ a restricted strategy of locking the dominant channel and using a single adjacent backup in the same direction, achieving controlled continuous advancement in sparse anchor point regions.

[0040] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the overall process of the global unique positioning method for unmanned vehicles in symmetrical channels according to the present invention;

[0042] Figure 2 This is a schematic diagram of the multi-hypothesis minimum set construction process used in the method of this invention;

[0043] Figure 3 This is a schematic diagram of the stable cruise-dominated and single spare tire process used in the method of this invention;

[0044] Figure 4 This is a comparative diagram of the present invention and the prior art. Detailed Implementation

[0045] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. However, those skilled in the art should understand that the embodiments described below are only for illustrating the present invention and should not be regarded as limiting the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example of a globally unique positioning method for symmetrical channels of unmanned vehicles

[0047] The innovation of this application lies in the following: This application constructs positioning constraints centered on channel-level semantics, forming a directed triplet by combining the channel identifier, node identifier, and travel direction, and explicitly binding them with anchor point encoding to create an identity verification domain triggered only at nodes, thus avoiding the generation of new identities on long straight segments. Unlike geometry-driven methods, this scheme uses the semantic graph as the sole basis for identity, with geometry only serving as an end-point endorsement.

[0048] Furthermore, this application proposes an interactive multi-model filtering method to form a maneuver semantic flow, which establishes state branches and interval constraints for going straight, turning in, turning out, slowing down when meeting oncoming traffic, making a U-turn, and skidding, respectively. It uses maneuver labels and stability gating to extrapolate and switch, so that the pose advancement is subject to dual restrictions of the direction of travel and node permission.

[0049] In addition, this application establishes only minimal-scale multiple hypotheses around the current channel and adjacent channels of the shared boundary, drives survivability based on channel boundary hard constraints and semantic conflicts, maintains a small number of continuous feasible identities, performs single-pass verification of topological reachability, maneuver alignment, orientation, temporal window and geometric endorsement at nodes / anchors, and determines the dominant identity with multiplicative gating and one-time solidification, and prohibits backflow.

[0050] During long straight-line operation, the main channel is locked for advancement, and only the single adjacent spare tire in the same direction is allowed to exist for a short time under restricted maneuverability, mileage / duration double caps, and trend monitoring; when entering the switching preparation window or reaching the trigger point, the node strong constraint adjudication is called, thus realizing the controlled transfer and convergence of identity.

[0051] like Figure 1 The globally unique positioning method for symmetrical channels of unmanned vehicles in this embodiment includes the following steps:

[0052] S1. Construct a channel-level semantic skeleton and channel semantic graph: Divide the site into a set of channels and assign a unique channel identifier to each channel; abstract the entrance, fork decision points, and end points as key nodes and assign unique node identifiers; map the travel direction as a directed attribute, and form a channel-level semantic skeleton using a triplet of directed relationship between channel identifier, node identifier, and travel direction; configure a unique coded anchor point in the visible sub-area of ​​each key node, and explicitly bind the anchor point code with the channel identifier, node identifier, and travel direction to form an anchor point semantic quadruple; establish a channel semantic graph with anchor points as graph nodes and passable segments as directed edges, and limit identity verification to only be performed at the triggered graph nodes.

[0053] S2. Establishing a maneuver semantic flow based on interactive multi-model filtering: Define six types of maneuvers: straight-line cruise, channel entry, channel exit, slow-moving when meeting oncoming traffic, U-turn at the end, and low-grip skidding. Set allowable edge / node domains and interval constraints for heading, lateral offset, and speed for each maneuver. Establish an independent state branch for each type of maneuver and output the pose evolution and branch confidence. Select the maximum confidence as the current maneuver semantic label and calculate the label stability. Based on the permissibility of the maneuver label and its corresponding edge / node, pose extrapolation is only allowed within the permissible domain. Lateral, heading, and longitudinal advances are restricted to the corresponding intervals and travel direction sets, respectively. Use stability as a switching threshold. Maneuver switching is only performed when the threshold and permissible conditions on the graph are met, and the maneuver label and stability are written back.

[0054] S3. Construct and advance a minimum necessary set of multiple hypotheses for channel identity: Generate channel identity hypotheses around the current channel and its semantically reachable neighboring channels that share a boundary with it. Each hypothesis is bound to a unique channel identifier and an independent pose trajectory container, and advances only along the set of directed edges of that channel. Based on the maneuver semantic label, the longitudinal advancement is limited to a subset of edges consistent with the direction of travel, and the lateral and heading changes are restricted to the pre-defined intervals. The channel geometric boundary is used as a hard constraint to prevent boundary crossings and crossing channels. Semantic conflicts are identified through consistency checks with channel semantics, and the hypothesis survival rate is adjusted accordingly. Only a small number of hypotheses that pass the checks and have good continuity are retained as the current valid set.

[0055] S4. Perform selective strong constraint association and one-time solidification at key nodes and anchor points: Determine whether a vehicle enters the node's neighborhood based on its projection on the channel semantic graph, and filter visible candidate anchor points by combining the current maneuver label, the node's allowed maneuver, and the anchor point's view. Set a maneuver consistency window for entry and exit based on the cumulative mileage along the edge, and impose temporal constraints on the candidates. Prune each channel identity hypothesis and candidate anchor point according to a single-pass verification based on topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement, and select a small number of association pairs with a comprehensive score with multiplicative gating and additive endorsement. Based on the passing association, implement serial adjudication of dominant determination, semantic exclusion, path integrity, and maneuver alignment, perform one-time weighted solidification of dominant hypotheses, and reduce the weight of non-dominant hypotheses to a non-returnable state. Output the solidified channel identifier and pose as the current identity, and bind this identity to the node.

[0056] S5. Implement channel locking and controlled spare tire strategy in long straight sections with sparse anchor points: Gating the set of multiple hypotheses with the stability of the maneuver semantic label, locking the propulsion to the set of directed edges of the dominant channel under stable cruise marking; generating only a unique spare tire hypothesis that shares the boundary with the dominant channel and is in the same direction, and shrinking its allowed maneuvers to slow passing and slight slippage; suppressing identity switching when not entering the switching preparation window, with the dominant trajectory continuously accumulating while the spare tire retains only the minimum necessary segments; applying piecewise function constraints to the spare tire, including adjacency filtering, restricted propulsion, dual capping of mileage and duration, and cross-edge / reverse trend monitoring, and clearing them immediately when entering the switching window, reaching the cap, or experiencing an out-of-bounds trend; when reaching the preset trigger point, calling step S4 to make a decision and restoring channel locking and controlled propulsion of the single spare tire.

[0057] S6. Implement the task, authorize the closed-loop channel and verify it locally: Compare the mapping between the task point and the channel identifier with the current channel identity output in step S4. If they match, enter the queue to be authorized. If they do not match, trigger local verification. Prioritize selecting the node corresponding to the nearest visible anchor point or planning to the nearest strongly constrained node under the direction of travel and the confirmed direction. After calling step S4 to complete the identity convergence again, return to the comparison and authorize execution when the three-element confirmation is consistent.

[0058] S7. Execute freeze and rollback convergence under continuous anomalies with uncertain identity: When the cumulative duration or mileage of the abnormal maneuver meets the threshold and no strong constraint is triggered at the node, freeze the channel identity hypothesis set to read-only and prohibit expansion and switching; advance along the travel direction and the confirmed direction with conservative parameters to the visible anchor point or strong constraint node, call step S4 to complete convergence, retain the dominant hypothesis consistent with the convergence result and clear inconsistent branches, reset the freeze flag and continue to advance on the convergence channel.

[0059] Step S1 includes the following sub-steps:

[0060] S11. Divide the site into a set of passages, denoted as the passage index set C, and assign a unique passage identifier to each passage. The entrance, fork decision points, and end points of each channel are abstracted into a set of key nodes N, and a unique node identifier is assigned to each key node. Map the direction of travel to a set of directed attributes. The direction of travel is Using directed relation triples By binding channel identifiers, node identifiers, and travel directions, a channel-level semantic skeleton is formed. The above ternary relationships are then sketched as a directed sketch set G0 in a graph structure, recording only topological existence without embedding geometric details.

[0061] S12. Design identity anchor point partitioning and sparse coding rules. Each key node n is designated as a visible sub-region of anchor points, forming a visible set V of anchor points. Only one coded anchor point is configured for each visible sub-region. The anchor point set is denoted as A, and the anchor point identifier is... The coded anchor points are mapped to channel identifiers, and a binding relationship is used to explicitly map a and c, while simultaneously binding them to node identifier n and travel direction d, forming an anchor point semantic quadruple. Anchor points are distributed at entrances, intersections, and ends using a low-density, uniform strategy to prevent the addition of new anchor points on long straight sections. The visible range and orientation of the anchor points are matched with the direction of travel to form an anchor point visual constraint set F, which limits false triggering under unexpected orientations.

[0062] S13. Treat anchor points as graph nodes, and abstract traversable segments between graph nodes as directed edges to form a channel semantic graph G. Bidirectionally bind each graph node to its channel identifier, node identifier, and anchor point identifier, using a binding mapping set B to record the graph node to its corresponding node. The unique relationship is established; each directed edge is associated with the allowed mobility set U, and the direction of the edge is consistent with the direction of travel. The channel semantic graph is published externally as the sole basis for identity verification, and identity verification is only allowed in triggered graph nodes. The generation of new identities is strictly prohibited in non-graph node intervals to avoid channel jumps and identity drifts in geometrically similar long straight segments.

[0063] Step S2 includes the following sub-steps:

[0064] S21. Let the set of maneuver semantics be denoted as the maneuver set U, which includes six types of maneuvers: straight-line cruising, turning into a lane, turning out of a lane, slow-moving when meeting oncoming traffic, U-turn at an end of a road, and low-grip skidding, each identified by a maneuver identifier. Let E represent each type of maneuver. Let E be the set of directed edges in the channel semantic graph G, and N be the set of key nodes. For each type of maneuver u, apply constraint binding: only bind to the allowed subset of edges. or a subset of nodes The above is permitted. The attitude constraints for each type of maneuver are recorded in interval form: the allowed heading change interval. Allowable lateral offset range Permissible speed range .

[0065] S22. Establish an independent state branch set S for each type of maneuver u. u Output pose evolution trajectory identifier p for each branch u With branch credibility identifier k u The maneuver identifier with the highest confidence level among all branches at the current moment will be selected. * As the current maneuver semantic label, calculate the stability identifier of the current maneuver semantic label. To characterize the continuity of recent maneuvers. (U) * and Synchronously write the current edge identifier or node identifier to the channel semantic graph to form instructions on how to move along the channel at this moment, and perform service consistency extrapolation and switching gating.

[0066] S23, Set the current maneuver semantic tag u * Perform an allowability check against the directed edge identifier or node identifier, only for those belonging to E. u* or N u* Position advancement and open extrapolation. Lateral offset is limited to... Inside, the course change is restricted to Within the system, the longitudinal displacement direction is aligned with the set of travel directions along the edge, and the generation of reverse or cross-edge extrapolation is prohibited. Short-term obstacle avoidance or slippage is interpreted as a disturbance within the channel, maintaining the continuity of the trajectory within the same directed edge or the same node neighborhood, and not triggering channel jumps.

[0067] S24. Mark the stability. The switching threshold is set so that fluctuations below the short-term threshold (i.e., stability does not meet the preset range and the duration is less than the preset time) do not trigger maneuver switching; only when the target maneuver is continuously met (i.e., stability does not meet the preset range and the duration is greater than the preset second time) will the maneuver switch from the current maneuver be executed. * Switching to candidate maneuvers. The switching operation is limited to a subset of nodes allowed by the channel semantic graph or within boundary conditions to avoid misjudging as entering or exiting a maneuver due to minor disturbances in long straight sections. The switching result is written back to the maneuver semantic label and stability identifier to ensure that long straight sections maintain a single maneuver propulsion, and that clear and traceable maneuver transitions are achieved at key nodes.

[0068] Among them, such as Figure 2 Step S3 includes the following sub-steps:

[0069] S31. Denote the current channel identifier as c0, and the set of adjacent channels that share a boundary and are semantically reachable from c0 as C1. Denote the channel identity hypothesis set as H, which is only the set of adjacent channels that share a boundary with c0 and each Generate a hypothesis and for each Bind a unique channel identifier c h Bind each h to an independent pose trajectory container identifier. Mileage boundary markers Agreement Set only along its channel directed edges To advance. Limiting the size to the minimum necessary size avoids unconstrained expansion in geometrically strong similar regions.

[0070] S32, Denote the semantic tag from the current maneuver in S2 as u. * The allowable range of course changes is denoted as Let its allowed lateral offset range be denoted as Each hypothesis The vertical progression constraint is applied to the set of directed edges that align with its travel direction, only when... Internal update Limit lateral and heading changes to and Inside, the channel's geometric boundary is applied as a hard constraint. This ensures that the pose does not exceed the boundary or cross the channel, forming a continuous trajectory consistent with the channel semantics.

[0071] S33. Compare the pose advancement result of each hypothesis h with u * Perform consistency checks, marking non-node transitions (entry or exit), contradictions with the travel direction, or crossing of channel boundaries as semantic conflicts. The hypothesis survival score is denoted as... For h that is in conflict Immediately reduce or directly remove H; for those satisfying u * Constrained h-lifting .Will The changes are driven solely by semantic conflicts, ensuring that long straight segments still maintain a small number of highly feasible identities even without anchor points.

[0072] S34, will pass the consistency check and The hypotheses with higher survival rates (i.e., survival indicators reaching a set threshold) constitute the current valid set to be selected, denoted as H. * H * According to channel identifier c h Pairs with trajectory continuity The satisfaction levels are sorted, and the set size is constrained within a preset upper limit identifier Y (i.e., the first set quantity), retaining only those related to the dynamic semantic u.* The few most consistent and trajectory-continuous assumptions are used to determine the current valid set, denoted as H. H is used as the sole candidate pool for subsequent anchor point events to maintain the continuity of identity in sparse anchor point intervals and prevent misjudgments of track skipping.

[0073] Step S4 includes the following sub-steps:

[0074] S41. Label the projection of the vehicle's current position onto the channel semantic graph as x, and denote the set of key nodes as N. Denote the node neighborhood radius threshold as r, and associate x with each node identifier. The distance is compared to determine whether it enters the node's neighborhood. The current maneuver semantic tag from S2 is denoted as u. * Let D be the set of travel directions. Let A be the set of anchor points bound to node n. n , using u * with direction subset For A n Filter the candidates to obtain the visible candidate set. .Will As the sole candidate for subsequent association, it avoids channel jumps caused by indiscriminate matching under symmetric geometry.

[0075] S42. Mark the cumulative mileage along the current directed edge as... Let the length of the edge be L. For each candidate anchor point... The edge sequence identifier is q a , for q a and Sort the relationships between the windows to generate an arrival order sequence. Denote the entry window and exit window as follows: and , using u * The allowable propulsion rate range and the reachable range given by the node geometry and topology are paired. , Configure the settings to form a set of mobile consistency windows W. n Mark 'a' that is not in the corresponding window as unassociatable to suppress early or late association between forks and ends.

[0076] S43, the technical approach for this sub-step is topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement. It employs a single-pass pruning strategy, ensuring that each channel identity hypothesis is verified only within the semantic domain allowed by the current node and the current maneuver. Let H be the set of channel identity hypotheses, n be the current node identifier, and u be the current maneuver label from S2. * Let U be the set of allowed maneuvers for a node. n Let the set of visible anchor points of a node be denoted as Let W be the set of mobility consistency windows. nThe candidate set that passes verification is gradually reduced until only the minimum necessary set that can establish a strong constraint association at the current node is retained.

[0077] A. Using topological reachability verification as the entry point, each hypothesis... The current pose is projected onto its set of directed edges in the channel, and entry into the node neighborhood is restricted to the condition that the node does not cross the channel boundary and does not cross the closed edge; h that does not meet the conditions is directly removed.

[0078] B. Perform the maneuver alignment verification under the premise of topological reachability, and u * with U n Alignment: Only retain the h that matches one of the following: turn in, turn out, go straight, or make a U-turn, to avoid interpreting slow-moving or slipping vehicles as lane changes.

[0079] C. Perform mirror confusion on the symmetrical layout for orientation verification, and check the consistency of each h on the visible side and the anchor installation side under the current flight direction, retaining only the h visible within the visible side angle domain. The combination of .

[0080] D. The timing window verification is performed based on the mileage relationship along the edge, and each combination is associated with its entry window. With Exit Window Alignment: Combinations not within a window are marked as unassociative.

[0081] E. Use geometric endorsement checks only for final confirmation after passing the first four levels, avoiding using geometric similarity as an early decision and preventing being misled under strong symmetry conditions. Implement the above five-stage verification in a one-pass manner, ensuring that each hypothesis exits immediately if any early constraint is not met, without proceeding to subsequent checks, and avoiding the introduction of irrelevant branches in long straight segments.

[0082] The above verification process is integrated into a single scoring function to construct a comprehensive score with multiplicative gating and additive endorsement:

[0083] ;

[0084] in, For comprehensive scoring, it is assumed that h and anchor point a are at node n and the maneuver is u. * The relevance score at that time; This is a topology reachability gate; if h passes the topology reachability check of the node's neighborhood, then it is considered topology reachable. Otherwise take ; For mobile positioning gating, if If the incoming / outgoing edges of h are consistent with the requirements, then take it. Otherwise take ; For orientation gating, if 'a' is visible on the current flight path h and the angular domain is reasonable, then 'a' is selected. Otherwise take ; For timing window gating, if the advancement from h to a is located in and In between, take Otherwise take ; The anchor point identity consistency endorsement weight represents the weighted amount when the code of a is consistent with the channel identifier, node identifier and travel direction of h. The path continuity endorsement weight represents the weighting amount when the sequence of traveled edges of h is consistent with the allowed incoming / outgoing edges of node n. For the weight of the maneuvering alignment endorsement, it means u * The weighted amount corresponding to the matching strength of the set of allowed maneuvers of the node.

[0085] The multiplicative part of the scoring function is used as a strong gating mechanism to perform rigid semantic verification at the first four levels; the additive part is used as a weak endorsement, only boosting the dominant hypothesis after passing the strong gating mechanism. The final associative set is defined as... The highest-scoring small number of pairs in K (i.e., the second set number of association pairs) are selected as input candidates for S44, which restricts the search domain of the next strongly constrained association.

[0086] Based on the above, this application avoids being induced by geometric repetition under strong symmetry and sparse anchor points, and guides subsequent convergence at the nodes to a unique channel identity.

[0087] S44. Let the subset of the associative hypothesis passed by S43 be denoted as Record the current trigger anchor point identifier as Limit association attempts to one-to-one or one-to-few (please explain "one-to-one" and "one-to-few" separately), using... The ordered pairs are matched, and for each candidate pair, a second check for orientation and temporal consistency is performed. Strong constraints are applied only to candidate pairs that satisfy both consistency. The selected hypothesis is denoted as h. * , for h * The channel identifier and pose are fixed, giving it an advantage at the current node and guiding subsequent convergence.

[0088] One-to-one refers to the current trigger anchor point a. * Given the established conditions, only H is allowed. ok The highest-scoring single-channel identity hypothesis h that has undergone double consistency verification. * With a * Forming an orderly pair And enter strong constraints; the rest No longer establish strong constraints with this anchor point (here, "\" is the set difference operator, representing the remaining set after removing the specified element from the set), and only consider it as an object to be subsequently eliminated or downgraded. A pair of lesser operators allows for more than one feasible hypothesis when symmetry or insufficient observation leads to this. and No more than a second predetermined number of candidates are used to form an ordered set of pairs in parallel. The system then undergoes a second check and scoring to ensure alignment and timing. However, after the sequential determination of S45's dominance, semantic exclusivity, path integrity, and maneuver alignment, only the single dominant pair remains. The remaining candidates are uniformly downgraded to a non-returnable state, which limits the scale of concurrency while maintaining the principle of one-time solidification.

[0089] S45, this sub-step proposes to determine the dominant pair, implement triple exclusion, and solidify it once, using strong adjudication to avoid backflow expansion after critical nodes. Successful one-to-one or one-to-few-associations that pass S44 are recorded as the dominant pair. , where h * Assuming the identity of the channel, a * To trigger the anchor point, denote the set of all candidate hypotheses at node n as H, and assign a weight to each hypothesis as follows: (Uniquely corresponding to each h), the minimum weight lower bound is denoted as .

[0090] Implement semantic exclusion rules first, for all (Here, "\" is the set difference operator, representing the remaining set after removing a specified element from the set.) Execute the operation with a. * The consistency of channel identifiers, node identifiers, and travel directions is checked, and any inconsistencies are immediately marked as eliminated and ordered to be removed. .

[0091] The path integrity rule is executed after exclusion. The sequence of traveled edges in the remaining h nodes is checked against the allowed incoming and outgoing edges of node n. Entry from non-allowed incoming edges or passage in reverse direction is not allowed; if this condition is not met, the node is directly eliminated. The maneuver alignment rule is used as the third decision, taking the current maneuver tag u from S2. * Aligning with the set of allowed maneuvers for each node, nodes h that need to complete the association under non-allowed maneuvers are directly eliminated to eliminate false turns induced by slow-moving vehicles and skidding. The above three rules are used only as strong pruning to retain the dominant assumption, ensuring convergence at the node in one go and not retaining buffer branches.

[0092] One-time weighting is used for updating the weights of the dominant hypothesis, while non-dominant hypotheses are uniformly downweighted to a non-returnable state. The single-time weighting at the node is constructed as follows:

[0093] ;

[0094] in: The new weights after node adjudication uniquely correspond to hypothesis h; Let h be the indicator function; if h is the dominant hypothesis, then take... Otherwise take ; Weighted for identity consistency, representing a * Anchor point encoding and h * The weighted value that is consistent with the channel identifier, node identifier, and direction of travel; For continuous weighting of paths, h represents * The weighted sum of the traveled edge sequence and the allowed incoming / outgoing edges of node n. For the weighted alignment of the maneuver, it indicates that it comes from u * A weighted amount of matching strength with the set of allowed maneuvers of a node; Let h be the indicator function; if h is not the dominant hypothesis, then take... Otherwise, take 0; The minimum weight lower bound represents the uniform weight of a eliminated or non-returnable state.

[0095] The dominant channel weighting and non-dominant channel weighting in the above formulas are executed in a one-time write operation, prohibiting gradual decay and subsequent backflow. The execution order is fixed as follows: first selected by S44. Then, the process is pruned sequentially according to the three rules of exclusivity, path, and maneuver, and finally, irreversible convergence is achieved by fixing the weights.

[0096] The aforementioned limited adjudication domain in this application is determined solely by node semantics, path legality, and maneuver priors, avoiding retention driven by geometric similarity. This approach blocks delay ambiguity in symmetric arrays from a construction perspective, ensuring that the channel identity after a node returns to a unique and continuous advancement channel.

[0097] S46. The remaining single or dominant hypothesis h after strong constraint fusion. * The channel identifier and pose are output as the current identity, and the identity is bound to node n. h * Write back to the channel to advance the process, so that advancement only follows the path with h. * The corresponding set of directed edges continues. This anchor point event is registered as an identity verification event to provide a stable identity benchmark for subsequent long straight segments, avoiding misjudgment as an adjacent channel in merging or parallel segments.

[0098] Among them, such as Figure 3 Step S5 includes the following sub-steps:

[0099] S51, The stability of the maneuver semantic label from S2 is denoted as Let T be the set of channel types. For each type... Set a stability threshold With abnormal threshold and guarantee .Will Mark as stable cruise, Mark it as an abnormal maneuver, and mark the intermediate region as a transitional state.

[0100] The gating strategy selection of the channel identity assumption set H, which uses the three-state label as the driving input S3, limits subsequent tightening or loosening to be performed only based on this semantic signal, without introducing heuristic triggers that are irrelevant to the business.

[0101] S52. This sub-step is constructed as a four-segment serial process: channel locking, single spare tire generation, switching suppression, and continuous accumulation, which only takes effect under stable cruise conditions. The stability flag is denoted as... The dominant channel identifier is denoted as c. * Let the dominant hypothesis be denoted as h. * Let its set of directed edges be denoted as E. c* Let the set of adjacent channels be denoted as Let the current maneuver semantic label be u. * Let U be the set of allowed maneuvers for a node. n The switch preparation window will be marked as .

[0102] Implement channel locking first, and h * The advancement is limited to E c* Inside, for any pointer not to E c* Expansion requests are directly rejected to ensure pose progression along a single directed edge sequence. Single backup plan generation is performed after locking, starting from... Only select those related to c * The only channel sharing the boundary, denoted as c. + Generate the unique backup hypothesis h + The permitted maneuverability is set to a contraction zone, only interpreting slow-moving and slight slippage when meeting oncoming traffic, disallowing turning in or out. Switching suppression is executed as the third phase, only occurring near the node and satisfying... And in the preparation window Initiate handover preparation within the specified time; increase the handover threshold during purely linear intervals to prevent identity switching. Perform continuous accumulation as the final execution segment for h. * The pose and odometer along the edge are continuously recorded and denoted as the dominant trajectory container; for h + Only the minimum necessary fragments are retained, and the identity is not output to the outside world. It does not participate in external adjudication input.

[0103] Furthermore, this application proposes to uniformly drive the above through a single gating formula, ensuring that the constraint sequences of locking, single spare tire, suppression, and accumulation automatically take effect during stable cruise:

[0104] ;

[0105] in: This is a gating indicator, uniquely corresponding to hypothesis h, used to determine whether to advance or output; As the dominant indicator function, if h is the dominant hypothesis, then take ; This is a channel lock indicator; if the propulsion is only located at E... c* Take the inner rules ; Let h be the backup indicator function; if h is the backup assumption, then take... ; For adjacent legality indication, if c + With c * If the boundary is shared and unique, then take... ; To reduce the maneuverability indicator, if the spare tire's permissible maneuverability only includes slow-moving oncoming traffic and slight slippage, then take... ; To switch the preparation window indicator, if it is near a node and meets the following conditions... And enter Then take .

[0106] The first term in the above formula drives the dominant assumption to propel continuously within a locked single channel; the second term allows only a single, adjacent, retractable spare tire to remain briefly when the switching preparation window is not triggered. Applying this gating result to propulsion and output: only when... When exporting identity and location; Only the minimum necessary fragments are retained internally and their dissemination is strictly prohibited.

[0107] This can be understood as the technical approach of this application being fixed as follows: first, lock in the dominant technology, then the minimum backup, then suppress switching, and finally accumulate coherence. Semantic constraints are used to solve the symmetry confusion problem of long straight segments, ensuring that in the sparse anchor point range, single-channel advancement is still the main approach, and the backup only serves as a last resort and does not replace the original technology.

[0108] S53. Trigger the freeze strategy by marking abnormal maneuvers, suspend identity switching operations, raise the pruning threshold, and allow short-term retention of the assumption h of a single adjacent backup. + Lateral offset and heading fluctuations are interpreted as intra-channel disturbances, allowing only minor adjustments within the current directed edge neighborhood, and prohibiting cross-channel turns. The freeze will continue until... The system will return to a stable state or reach the next strong constraint trigger point, ensuring that short-term fluctuations caused by passing, obstruction, or slippage are not amplified into channel jumps.

[0109] S54. Furthermore, this application proposes and constructs a controlled adjacent spare tire hypothesis to limit the diffusion range.

[0110] This sub-step involves adjacency filtering, restricted advancement, capping and exit, and consistency monitoring, activating only within the range permitted by stability or anomaly gating. The dominant channel is denoted as c. * Let the set of adjacent channels be denoted as The only spare tire passage is designated as Let the dominant hypothesis be denoted as h. * Let the backup plan be denoted as h. + Let the current maneuver semantic label be u. * Let U be the set that allows for maneuverable contraction. shrink (Including only slow-moving traffic and slight slippage), the cumulative mileage will be recorded as follows. The duration of existence is recorded as Record the mileage limit as The maximum duration is recorded as Record the switch preparation window indicator as The consistency indicator of the travel direction along the edge is denoted as X. dir (Taking 1 in the same direction as the dominant trend), the cross-side trend indicator is denoted as X. cross (If a cross-edge trend is detected, set it to 1). Record the reverse trend indicator as X. rev (Reverse trend detected, set to 1).

[0111] The generation of the backup plan is limited to that of c. * The only channel sharing the boundary uses adjacency filtering only when... h is generated at time + And immediately after generation, apply restricted propulsion: only allow propulsion along c * Small-amplitude pose evolution within parallel and unidirectional edges is prohibited, including turning in, turning out, and crossing edges. Mileage and duration are capped as exit conditions; once the cap is reached, the process is automatically cleared. Consistency is monitored throughout the entire process; if crossing edges, reversing, or prematurely triggering a turning trend is detected, the process is immediately terminated and marked as non-compliant, prohibiting repeated generation in the same direction.

[0112] The above four constraints are used to uniformly drive the generation, existence, and exit of the backup plan through piecewise functions:

[0113] ;

[0114] in: The function is a piecewise function for the backup state, where 1 indicates that creation / maintenance is allowed, 0 indicates immediate clearing, and hold indicates maintaining the previous state; c * This serves as the primary channel identifier, uniquely pointing to the channel to which the currently output identity belongs; To be with c * The set of adjacent channels sharing a boundary; c +This is the spare tire lane identifier, uniquely corresponding to the selected adjacent lane; u * The current maneuver semantic tag originates from S2; U shrink To reduce the maneuvering range, it only includes slow passing and slight skidding; X dir For consistency indication, the value is 1 when the main tire and the spare tire are in the same direction along the edge; The mileage accumulated for the spare tire is only accumulated within the limited mileage range of the spare tire. This is the mileage limit; once reached or exceeded, it will be cleared. The duration of the backup plan is determined by a continuous timer starting from the moment it is generated. This is the maximum duration; once reached or exceeded, it will trigger a cleanup. To indicate the switch preparation window, the value is 1 when entering a node and allowing the switch window to be switched; X cross This is a cross-edge trend indicator; it is set to 1 when a cross-edge trend is detected. rev This indicates a reverse trend; it is set to 1 when a reverse trend is detected.

[0115] The piecewise function's allowed segments are subject to four constraints simultaneously: adjacency joining, maneuvering contraction, unidirectional consistency, and failure to reach double caps; this ensures that the backup plan only serves as a short-term fallback and does not generate alternative paths. The clearing segment integrates three types of strong triggers: out-of-bounds trend, reaching double caps, and entering the switching window; this allows for immediate exit at risk or when a point is reached, preventing diffusion and backflow. The hold segment is used to maintain state stability during transitions where neither the allowed conditions are met nor the clearing is triggered, avoiding frequent interruptions.

[0116] This application narrows the control over the generation and withdrawal of backup devices to determinable conditions, ensuring that backup devices only provide the minimum necessary continuity, do not disclose their identity, and do not compete with the dominant identity, thus leaving clear and controllable candidates for the adjudication of the next strongly constrained node.

[0117] S55. Record the delayed adjudication trigger point as the nearest reachable critical node identifier. Or the most recently visible anchor point marker Record the current mileage along the edge from the trigger point to both the primary trajectory container and the backup trajectory container. Use the consistency of arrival time and channel direction as the sole retention condition; for h-axis paths exhibiting reverse or boundary-crossing trends... + Perform timely cleanup. Maintain a neutral stance on identity checks prior to the trigger point, refraining from outputting cross-channel decisions and only maintaining the h-value. * The continuity of one's external identity.

[0118] S56. When the trigger point occurs, invoke the selective strong constraint process of S4, using the node or anchor point as the decision source, and apply it to the dominant hypothesis h. * With the backup assumption h +Perform a round of adjudication. The assumption of a single, strongly constrained channel identity will be solidified, while all other assumptions will be eliminated in one go, with no buffer or return channels retained. The gating strategy will be restored to the S52's stable cruise default settings, bringing propulsion back to a controlled state dominated by a single channel and a single adjacent backup, continuously serving the identity continuity requirements of long, straight sections with sparse anchor points.

[0119] Step S6 includes the following sub-steps:

[0120] S61. Let Q be the set of task points in the channel semantic graph, and let the unique identifier of each task point be denoted as . The mapping relationship between task points and channel identifiers is denoted as follows: The current convergence channel identifier output by S4 is denoted as c. * The upcoming mission point The mapping channel identifier is denoted as .Will With c * Perform a one-to-one comparison: mark consistent cases as candidate authorizations and write them into the pending confirmation queue; mark inconsistent cases as identity conflicts, prohibit the issuance of any action commands, and transfer to the S62 on-site review process.

[0121] S62. Label the projection of the current position onto the channel semantic graph as x, and denote the set of strongly constrained nodes as N. hard Let A be the set of visible anchor points. vis When on-site verification is triggered, the nearest visible anchor node will be searched first, and the nearest candidate will be denoted as... If no visible anchor point is found, then program from x to the nearest strongly constrained node, following only the travel direction and the confirmed direction. The vehicle is driven to the selected node using a conservative strategy, and the anchor point event is actively triggered. The selective strong constraint process of S4 is invoked to complete the channel identity convergence again, and the verified channel identifier is obtained. .Will Write back to the new current identity and return to S61 to re-perform the closed-loop check until it matches... Either the task is completed or it is cancelled.

[0122] S63. Finally bind the channel identifier, which has been consistent through the closed loop of S61 and has undergone necessary verification in S62, with the task point to form a three-element confirmation of task point, channel ID, and current location, denoted as... This confirmation is used as the sole basis for authorization, and the authorization flag is written into the execution command. Tasks that fail consistency checks or verification are suspended and no actions are issued. The authorization result is sent back to the execution module, and before taking any action, the channel identifier for the current location is checked again to ensure it is still "c". *If a deviation occurs, execution will be suspended and the system will return to S62 to trigger on-site verification, preventing the system from entering the wrong channel or performing operations in the wrong location under conditions of symmetrical layout or long straight section disturbance.

[0123] Step S7 includes the following sub-steps:

[0124] S71. Denote the maneuver semantic tag from S2 as u. * Let U be the set of abnormal maneuver tags. abn The cumulative duration of the anomalies is recorded as T. abn Abnormal cumulative mileage is recorded as L. abn The corresponding thresholds are denoted as T. thr L thr Record the current node's strong constraint trigger state as... (1 for triggered, 0 for not triggered). Set the judgment condition to: when And satisfy or ,and When this occurs, the status is determined to be a persistent anomaly with uncertain identity. The current channel identity hypothesis set is denoted as H, its survival state is frozen to read-only, and expansion and handover operations are prohibited, denoted as the freeze flag. Mark the rollback process as the only executable strategy and proceed to S72.

[0125] S72. Label the projection of the current position onto the channel semantic graph as x, and label the current dominant channel as c. * Let its set of directed edges be denoted as E. c* Let N be the set of strongly constrained nodes. hard Let the subset of nodes corresponding to the visible anchor points be denoted as Limit the search strategy to proceed only along visible or confirmed directions: prioritize N vis Select the target node n with the shortest mileage. * ;like Under the constraints of not crossing rows, not reversing, and not exceeding unlabeled branches, at N hard Select along E c* Nearest reachable target node n * The path is restricted to a sequence of edges allowed in the direction of travel, bound to a conservative set of operating parameters (low speed, limited acceleration, zero in / out maneuver), and any generation of new channel identities and switching requests is prohibited, maintaining the freeze flag. Until n is reached * .

[0126] S73. Proceed the vehicle along the path planned in S72 to the target node n using conservative parameters. * Actively trigger the anchor point event and call the selective strong constraint process of S4 to perform identity convergence, obtaining a unique channel identifier denoted as... and its pose. The channel identity hypothesis set written back to S3 is retained only with... The consistent dominant assumption removes the backup and inconsistent branches from before the freeze and resets the freeze flag. Restore the task plan to its original flow, starting from n * along The directed edge continues to advance; this rollback is registered as a strongly constrained recovery event so that subsequent S6 tasks and channel closed-loop checks can be called to form a controllable closed loop of rollback, anchor point convergence, and recovery under uncertain identity conditions, preventing anomalies from being amplified into cross-channel misoperations.

[0127] like Figure 4 The diagram illustrates a comparison between this application and existing technologies. MHT stands for Multiple Hypothesis Tracking, and IMM stands for Interacting Multiple Model Filtering. Compared to existing technologies that rely on geometric matching and periodic anchor point relocation, this application uses channel semantics as the unique identity decision domain. It employs a triplet of channel ID, node ID, and travel direction, bound to a quadruple of anchor point encoding. Verification is triggered only at nodes, reducing identity generation and drift in long straight segments. For maneuver estimation, IMM is used to construct the maneuver semantic flow, and stability-gated switching ensures that extrapolation is strictly constrained by travel direction and node permission. In identity management, only the minimum necessary multiple hypotheses for the current channel and adjacent channels on the shared boundary are maintained. At nodes / anchor points, the dominant identity is determined through multiplicative gating and one-time solidification of topological reachability, maneuver alignment, orientation, and temporal windows, suppressing backflow expansion. Long straight segments employ a restricted strategy of locking the dominant channel and using a single adjacent backup in the same direction to achieve controlled continuous advancement in sparse anchor point regions.

[0128] This invention reduces misjudgment and rollback costs in strongly symmetric, sparse anchor environments by maintaining unique identities centered on channel semantics and achieving strong constraint convergence at nodes. This facilitates stable execution of upper-layer business processes in large-scale, continuous operation scenarios. Its semantically guided, geometrically endorsed structure facilitates decoupling and integration with existing SLAM / scheduling / warehouse control systems. It can be embedded as an independent identity adjudication module into existing AGV / AMR platforms, reducing reliance on high-density anchors and high-precision global maps, thereby balancing deployment costs and operational reliability. The closed-loop design for anomalies and rollbacks also meets security compliance and traceability requirements, providing an operable identity basis for task authorization and pre-execution verification.

[0129] Beyond data centers and warehouse racking, this method can be extended to long, straight, and repetitive structural scenarios such as airport baggage sorting channels, hospital logistics corridors, library stacks, manufacturing plant production line corridors, mines, and tunnels. In environments with weak visual textures or high occlusion, such as logistics sorting chutes, automated parking garage aisles, and port yard passageways, it can also be used as a channel-level identity layer in conjunction with existing positioning layers to ensure identity verification at critical nodes and controlled movement over long sections. For multi-robot collaborative systems, the continuity of channel identities can be used for conflict resolution, right-of-way determination, and task relay, improving the determinism of system-level scheduling. For devices requiring low-computing-power edge terminals, such as low-cost AMRs or tractor units, a semantically gating-based decision chain is also feasible.

[0130] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A globally unique positioning method for unmanned vehicles in symmetrical channels, characterized in that, Includes the following steps: S1. Assign a unique channel identifier to each channel in the site where the symmetrical channel is located; assign a unique node identifier to the key nodes of the site, including the entrance, fork decision points, and the end; map the passage direction of the site to a directed attribute; and form a channel-level semantic skeleton using a triplet of the directed relationship between the channel identifier, node identifier, and passage direction; configure a unique coded anchor point in the visible sub-area of ​​each key node, and explicitly bind the anchor point to the channel identifier, node identifier, and passage direction to form an anchor point semantic quadruple; establish a channel semantic graph with the anchor point as the graph node and the passable segment as the directed edge, and limit the channel identity verification to only be performed at the triggered graph node; S2. Define the maneuver types of unmanned vehicles in a symmetrical channel. From the set of directed edges and key nodes of the channel semantic graph, set allowed subsets of edges or nodes for each type of maneuver and limit attitude constraints. Establish independent state branches for each type of maneuver and output the pose evolution and branch confidence for each branch. Select the maneuver type corresponding to the branch with the highest branch confidence as the current maneuver and calculate the stability of the current maneuver. Write the maneuver identifier of the current maneuver and the stability identifier of the stability into the current edge identifier or node identifier of the channel semantic graph to form an instruction on how to move along the channel at this moment. S3. Generate channel identity hypotheses around the current channel in which the unmanned vehicle is located and all adjacent channels that share a boundary and are semantically reachable. For each channel identity hypothesis, advance only along the directed edge set of the corresponding channel to obtain the pose advancement result of each channel identity hypothesis. Based on the current maneuver identifier obtained in S2 and the corresponding limited attitude constraints, restrict the attitude advancement of the unmanned vehicle, and use the channel geometric boundary as a hard constraint to obtain a continuous trajectory consistent with the channel semantics under the current maneuver of the unmanned vehicle. By checking the consistency between the pose advancement result of each channel identity hypothesis and the continuous trajectory consistent with the channel semantics, identify semantic conflicts of the channel identity hypotheses, and retain only a first set of channel identity hypotheses that pass the consistency check and meet the continuity requirements as the current valid set. Use this current valid set as the sole candidate pool for anchor point events in the subsequent S4 step. S4. Determine whether the unmanned vehicle has entered the node neighborhood by projecting its image onto the channel semantic map, and filter visible candidate anchor points by combining the current mobility label, the node's allowed mobility, and the anchor point's field of view. Based on the cumulative mileage along the edge, a consistent window for entry and exit maneuvers is set, and temporal constraints are applied to candidate anchor points. For each channel identity assumption and candidate anchor point in the unique candidate pool of step S3, a single-pass verification is performed based on topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement. A second set number of association pairs are selected by a comprehensive score with multiplicative gating and additive endorsement. Based on the associations, a serial decision is made on dominant determination, semantic exclusion, path integrity, and maneuver alignment. The dominant assumption is weighted and solidified in one go, and the non-dominant assumptions are downgraded to a non-returnable state. The solidified channel identifier and pose are output as the current identity, and this identity is bound to the node.

2. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 1, characterized in that, Step S2 further includes: setting the stability flag as a switching threshold driving quantity; not triggering maneuver switching if the stability does not meet the preset range and the duration is less than the preset first time; performing a switch from the current maneuver to another maneuver if the stability does not meet the preset range and the duration is greater than the preset second time; limiting the switching operation to a subset of nodes or boundary conditions allowed by the channel semantic graph; and writing the switching result back to the maneuver tag and stability flag; wherein the preset second time is greater than or equal to the preset first time.

3. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 1, characterized in that, In step S3, the pose advancement result of each channel identity hypothesis is obtained by advancing only along the corresponding channel directed edge set. This includes: binding a unique channel identifier to each channel identity hypothesis, binding an independent pose trajectory container identifier and mileage boundary identifier to each channel identity hypothesis, constraining the longitudinal advancement of each channel identity hypothesis to a subset of directed edges with the same direction of travel, and updating the pose trajectory container identifier only within the channel directed edges to obtain the pose advancement result of each channel identity hypothesis.

4. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 1, characterized in that, In step S3, identifying semantic conflicts of channel identity assumptions by checking the consistency between the pose advancement results of each channel identity assumption and the continuous trajectory consistent with the channel semantics includes: by checking the consistency between the pose advancement results of each channel identity assumption and the continuous trajectory consistent with the channel semantics, cases where non-nodes turn in or out, contradict the direction of travel, or cross the channel boundary are marked as semantic conflicts.

5. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 4, characterized in that, In step S3, retaining only a first set number of channel identity assumptions that pass the consistency check and meet the continuity requirements as the current valid set includes: setting a survival indicator for each channel identity assumption; lowering the survival indicator corresponding to a channel identity assumption that is identified as having a semantic conflict; raising the survival indicator corresponding to the remaining channel identity assumptions; forming the current valid set of channel identity assumptions that pass the consistency check and whose survival indicators reach a set threshold; sorting each channel identity assumption in the current valid set according to the satisfaction level of the channel identifier and trajectory continuity; and selecting the first set number of channel identity assumptions with the highest satisfaction level in the satisfaction ranking as the current valid set.

6. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 1, characterized in that, In step S4, the identity assumptions and candidate anchors of each channel in the unique candidate pool from step S3 are pruned according to the single-pass verification based on topological reachability, maneuver alignment, orientation, temporal window, and geometric endorsement, including: Using topological reachability verification as the entry point, the current pose of each channel identity assumption is projected onto its set of directed edges, limiting entry into the node neighborhood only under the conditions of not crossing the channel boundary and not crossing closed edges; channel identity assumptions that do not meet the conditions are directly eliminated. The maneuver alignment verification is performed under the premise of topological reachability. The current maneuver is aligned with the set of maneuvers allowed by the node. Only the channel identity assumption that matches one of the maneuver types (turn in, turn out, go straight or turn around) is retained to avoid interpreting the slow-moving and slipping of the maneuver types as channel switching. The orientation verification is performed by mirroring the symmetrical layout. Each channel identity assumption is checked for consistency between the visible side and the anchor installation side under the current flight direction. Only the combination of channel identity assumption and visible candidate anchor point that is visible in the visible side corner domain is retained. The timing window verification is performed based on the mileage relationship along the edge. The entry and exit windows of each combination and its candidate anchor point are aligned. Combinations not in the window are marked as unassociative. Geometric endorsement checks will only be used for final confirmation after the first four levels have been passed.

7. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 6, characterized in that, In step S4, based on association, a series of decisions are made regarding dominance determination, semantic exclusion, path integrity, and maneuver alignment. This includes a one-time weighted solidification of the dominant hypothesis and a reduction of the weight of non-dominant hypotheses to a non-returnable state, including: The semantic exclusion rule is implemented first. For the combination of the pruned channel identity assumption and the visible candidate anchor point, the consistency check of the channel identity assumption, node identity, and travel direction of the visible candidate anchor point is performed. Any inconsistency is immediately marked as elimination. The path integrity rule is executed after exclusion. The sequence of traveled edges assumed by the remaining channel identity is checked against the allowed inbound and outbound edges of the key nodes. Entry from non-allowed inbound edges or passing through in reverse is not allowed. If the condition is not met, the node is directly eliminated. The maneuver alignment rule is used as the third adjudication, aligning the current maneuver from step S2 with the set of allowed maneuvers of the node, and directly eliminating channel identity assumptions that need to be associated under non-allowed maneuvers.

8. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 1, characterized in that, Also includes: S5. Implement channel locking and controlled spare tire strategy in long straight sections with sparse anchor points: Gating the channel identity assumption based on the stability of the maneuver, locking the propulsion to the set of directed edges of the current channel under stable cruise marking; generating only a unique spare tire assumption that shares the boundary with the current channel and is in the same direction, and shrinking its allowed maneuver to slow passing and slight slippage; suppressing identity switching when not entering the node neighborhood, the trajectory under the current channel is continuously accumulated while the unique spare tire assumption retains only the minimum necessary segment; applying piecewise function constraints such as adjacency filtering, restricted propulsion, dual capping of mileage and time, and cross-edge or reverse trend monitoring to the unique spare tire assumption, and clearing it in real time when entering the node neighborhood, reaching the cap, or showing a tendency to cross the boundary; when reaching the preset trigger point, calling step S4 to make a decision and restoring channel locking and controlled propulsion of the single spare tire.

9. The globally unique positioning method for unmanned vehicles in symmetrical channels according to claim 8, characterized in that, Also includes: S6. Implement the task, grant closed-loop authorization for the channel, and conduct on-site verification: Compare the mapping between the task point and the channel identifier with the current identity output in step S4. If they match, enter the queue for authorization. If they do not match, trigger on-site verification. Prioritize selecting the node corresponding to the nearest visible anchor point or planning to the nearest strongly constrained node under the direction of travel and the confirmed direction. After calling step S4 to complete the identity convergence again, return to the comparison and authorize execution when the ternary confirmation is consistent.

10. The globally unique positioning method for symmetrical channels of unmanned vehicles according to claim 9, characterized in that, Also includes: S7. Execute freeze and rollback convergence under continuous anomalies where the channel identity is uncertain: When the cumulative duration or mileage of the abnormal maneuver meets the threshold and no strong constraint is triggered at the node, freeze the set of channel identity assumptions as read-only, prohibiting expansion and switching; advance along the travel direction and the confirmed direction with set parameters to the visible anchor point or strong constraint node, call step S4 to complete convergence, retain the dominant assumption consistent with the convergence result and clear inconsistent branches, reset the freeze flag and continue to advance on the convergence channel.

Images