A robot work trajectory planning method, device and medium

Abstract

This invention discloses a robot operation trajectory planning method, device, and medium, belonging to the field of operation trajectory planning technology. The method includes: acquiring visual positioning and mapping data, performing coordinate alignment and generating traversable regions, and outputting a traversable map; performing obstacle semantic classification on the traversable map, fusing dynamic target prediction to construct a time-occupied zone and a time-feasible region; performing global coverage sequence planning on the backfill contract binding map to generate a global trajectory segment sequence; performing queue planning according to the backfill contract binding map based on the backfill queue to generate a backfill trajectory, driving the robot to perform backfilling and acceptance, and outputting an operation report. This invention generates a traversable map through visual positioning and mapping and coordinate alignment, and then uses obstacle semantic classification to fuse dynamic target prediction to construct a time-occupied zone and a time-feasible region, forming a process control capability that runs through planning, execution, and acceptance, improving coverage integrity, continuous operation, and delivery reliability.

Description

Technical Field

[0001] This invention relates to the field of work trajectory planning technology, and in particular to a method, device and medium for robot work trajectory planning. Background Technology

[0002] Trajectory planning technology based on Visual Simultaneous Localization and Mapping (Visual SLAM) has become a core means of realizing environmental perception and path generation. Existing mainstream technical solutions usually rely on depth cameras or LiDAR to collect point cloud data, construct a 3D map of the static environment through feature matching and pose estimation, and then use rasterization methods to divide the passable area. Subsequently, planning algorithms often adopt the full-coverage path planning (CPP) strategy, such as ox-plowing decomposition or spanning tree cover algorithm, to divide the workspace into several sub-regions and generate a global path that traverses all sub-regions based on heuristic rules.

[0003] As application scenarios extend to highly dynamic, unstructured, and complex semantic environments, existing technologies are gradually revealing their limitations in handling coverage integrity under spatiotemporal coupling constraints. In particular, there is room for optimization in the process control logic between high-level decision-making and low-level execution. Traditional planning methods often treat environmental modeling as a static process or simply treat dynamic obstacles as instantaneous no-entry zones, lacking predictive modeling of the movement trends of dynamic targets and mechanisms for constructing feasible domains in the time dimension. When there are moving objects or temporary obstacles in the environment, the process control strategies of existing solutions cannot predict their impact on coverage quality during the planning stage, resulting in frequent local path blockages encountered by the robot during execution. Summary of the Invention

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

[0005] Therefore, this invention provides a robot operation trajectory planning method to solve the problem that coverage integrity is difficult to predict in the planning stage and to stably control in the execution stage.

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

[0007] In a first aspect, the present invention provides a robot operation trajectory planning method, comprising: acquiring visual positioning and mapping data, performing coordinate alignment and generating traversable regions, and outputting a traversable map; performing obstacle semantic classification on the traversable map, fusing dynamic target prediction to construct a time-occupied zone and a time-feasible region; performing coverage partitioning decomposition on the time-feasible region to generate a set of coverage partitions, simultaneously establishing a backfill contract binding graph for the set of coverage partitions, and initializing the coverage quality field target layer; performing global coverage sequence planning on the backfill contract binding graph to generate a global trajectory segment sequence; performing trajectory tracking and local rerouting based on the global trajectory segment sequence, and updating the coverage quality field online to trigger skipped segments and writing them into a queue to be backfilled; performing queueing planning according to the backfill contract binding graph based on the queue to be backfilled to generate a backfill trajectory, driving the robot to perform backfilling and acceptance, and outputting an operation report.

[0008] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the specific steps of collecting visual positioning and mapping data, performing coordinate alignment and generating traversable regions, and outputting a traversable map are as follows:

[0009] The robot uses cameras and inertial measurement to collect continuous images and inertial data, synchronously acquires odometry information, forms visual positioning mapping data, performs time synchronization and abnormal frame removal on the visual positioning mapping data, and extracts positioning feature information to obtain continuous pose estimation results.

[0010] The continuous pose estimation results are registered with the operation coordinate system to form an aligned environment map. Obstacle areas are identified and expanded on the aligned environment map to generate a passable region. The passable region is then fused with the operation boundary to output a passable map.

[0011] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the specific steps of performing obstacle semantic classification on the passable map and fusing dynamic target prediction to construct the time occupancy zone and the time feasible region are as follows:

[0012] Read the passable map and extract the boundaries of the obstacle area to form a candidate obstacle set. Perform geometric feature calculation and obstacle semantic classification on the candidate obstacle set to generate obstacle semantic labels.

[0013] Based on the semantic labels of obstacles, edge-fitting buffers are generated for wall obstacles and detour buffers are generated for column obstacles to form semantic safety constraints. The robot’s current position and velocity information are collected, and the predicted position is recursively calculated by time step in the prediction time domain to generate the predicted occupied area.

[0014] The predicted occupied area and semantic security constraints are jointly mapped to the walkable map to construct the temporal occupied zone. The walkable map is then clipped in the temporal domain based on the temporal occupied zone to output the temporal feasible region.

[0015] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the steps of performing coverage partition decomposition on the time feasible domain to generate a coverage partition set, and simultaneously establishing a backfill contract binding graph for the coverage partition set, are as follows:

[0016] Extract the regional boundary information of the time feasible region to form the region to be decomposed. Perform connectivity analysis on the region to be decomposed and perform coverage partition decomposition to form multiple coverage partitions. After summarizing, form a set of coverage partitions.

[0017] The accessible and exit boundaries are extracted from the set of coverage partitions, and the main coverage direction and turning buffer range are calculated to obtain the coverage partition planning constraints. The adjacency relationship of coverage partitions is established based on the shared boundaries between the sets of coverage partitions to form a coverage partition switching diagram.

[0018] Based on coverage zoning planning constraints, select replenishment entry and exit points on the accessible and exit boundaries of the coverage zoning, set replenishment trigger and termination conditions, generate replenishment contracts, and associate and store the replenishment contracts with the coverage zoning switching diagram to form a replenishment contract binding diagram.

[0019] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the coverage quality field target layer is established in the traversable map coordinate system, which is consistent with the spatial range of the time feasible domain. The target coverage intensity and target overlap level corresponding to the operation type in the operation quality requirements are written into the target quality distribution layer position by position to form the basic target value. Then, combined with the obstacle semantic label, the edge buffer range corresponding to the wall obstacle and the detour buffer range corresponding to the column obstacle are mapped to the target quality distribution layer, and the corresponding target quality requirements are written into the edge buffer range and the detour buffer range.

[0020] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the specific steps for performing global coverage sequence planning on the backfill contract binding graph to generate a global trajectory segment sequence are as follows:

[0021] Based on the coverage partition switching graph in the backfill contract binding graph, the switching cost between coverage partitions is calculated and merged with the expected coverage cost within the coverage partition to form a global planning cost. Starting from the coverage partition where the starting position is located, the access order of coverage partitions is generated by constraining the global planning cost.

[0022] For each coverage partition in the coverage partition access sequence, combined with the main coverage direction, a trajectory segment within the partition is generated. Based on the backfill exit point and backfill entry point, a switching trajectory segment is generated. The trajectory segments within the partition and the switching trajectory segments are concatenated and associated with the corresponding backfill contracts to generate a global trajectory segment sequence.

[0023] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the steps of performing trajectory tracking and local rerouting based on the global trajectory segment sequence, and updating the coverage mass field online to trigger skipped segments and writing them into the queue to be replenished are as follows:

[0024] Read the global trajectory segment sequence and schedule the current trajectory segments to be executed in sequence. Based on the robot's current pose state and the current trajectory segments to be executed, generate motion control commands and perform trajectory tracking.

[0025] During trajectory tracking, environmental observation data is continuously acquired and its consistency with the time occupancy zone is verified to generate a local rerouting trajectory.

[0026] During trajectory execution, the system continuously acquires job feedback data and maps it to the job area to obtain the coverage quality field quality value. The coverage quality field is updated by comparing the target layer of the coverage quality field with the coverage quality field quality value, and the current trajectory segment to be executed is written into the queue to be replenished.

[0027] As a preferred embodiment of the robot operation trajectory planning method of the present invention, the steps of performing queue insertion planning according to the queue to be replenished based on the replenishment contract binding diagram, generating a replenishment trajectory, driving the robot to perform replenishment and acceptance, and outputting an operation report are as follows:

[0028] Read the queue to be replenished and obtain the replenishment entry point and replenishment exit point associated with the skipped segment. At the same time, read the replenishment contract binding graph to obtain the set of overlay partitions and the overlay partition switching graph associated with the skipped segment.

[0029] Based on the contract binding graph, calculate the queueing cost from the robot's current position to the queueing entry point and generate the queueing planning result. Select the skipped segment corresponding to the queueing planning result as the current queueing task.

[0030] Based on the current replenishment task's replenishment entry point and replenishment exit point, a replenishment trajectory is generated, and a global trajectory segment sequence is inserted to form an interleaved execution sequence. This sequence drives the robot to execute the replenishment trajectory and generates an acceptance conclusion based on the coverage quality field and the coverage quality field target layer, outputting a work report.

[0031] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the robot operation trajectory planning method as described in the first aspect of the present invention.

[0032] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the robot operation trajectory planning method as described in the first aspect of the present invention.

[0033] The beneficial effects of this invention are as follows: by generating a passable map through visual positioning mapping and coordinate alignment, a unified expression of the work space and planarable boundary constraints are achieved. Then, by integrating obstacle semantic classification with dynamic target prediction, a time occupancy zone and a time feasible domain are constructed, forming a process control capability that runs through planning, execution and acceptance, realizing spatiotemporal integrated safety constraints and forward collision avoidance, and improving coverage integrity, continuous operation and delivery reliability. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart for robot trajectory planning methods.

[0036] Figure 2 A flowchart for generating the time feasible region.

[0037] Figure 3 A flowchart for creating a backfill contract binding diagram.

[0038] Figure 4 A flowchart for execution tracking and resubmission acceptance.

[0039] Figure 5 This is the histogram of the minimum dynamic target distance distribution.

[0040] Figure 6 A multi-curve comparison chart showing how the percentage of compliant area changes over the prediction time window. Detailed Implementation

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

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

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

[0044] Reference Figures 1-6 As one embodiment of the present invention, this embodiment provides a robot operation trajectory planning method, including the following steps:

[0045] S1. Collect visual positioning and mapping data, perform coordinate alignment and generate passable regions, and output a passable map.

[0046] S1.1. The robot calls the camera and inertial measurement to collect continuous images and inertial data, synchronously acquires odometry information, forms visual positioning mapping data, performs time synchronization and abnormal frame removal on the visual positioning mapping data, and extracts positioning feature information to obtain continuous pose estimation results.

[0047] Specifically, after the robot uses cameras and inertial measurement units to acquire continuous images and inertial data and synchronously obtains odometer information, it aligns the continuous images, inertial data, and odometer information to the same time series according to the acquisition timestamps to form visual positioning mapping data. This visual positioning mapping data then undergoes time synchronization processing. By correcting the correspondence between the timestamps of the continuous images, inertial data, and odometer information, and interpolating missing timestamps, the time-synchronized visual positioning mapping data is obtained. This time-synchronized visual positioning mapping data then undergoes abnormal frame removal processing, which detects abnormalities in the clarity and brightness of the continuous images and detects inertial... Data saturation mutations are used to delete corresponding segments to obtain visual positioning mapping data after abnormal frame removal. The visual positioning mapping data after abnormal frame removal enters the positioning feature information extraction processing. Trackable feature points and feature descriptions are extracted from continuous images to generate feature matching relationships between adjacent images. The positioning feature information is combined with inertial data and odometer information. Pose recursion is achieved by pre-integrating the inertial measurement unit (IMU) data. After combining the positioning feature information and odometer information, extended Kalman filtering is used to update the position, attitude, velocity, and sensor bias to complete the pose correction, and the continuous pose estimation results in the working coordinate system are obtained.

[0048] S1.2. Register the continuous pose estimation results with the operation coordinate system to form an aligned environment map. Identify and dilate the obstacle areas on the aligned environment map to generate a passable region. Merge the passable region with the operation boundary to output a passable map.

[0049] Specifically, the continuous pose estimation results are used as the pose sequence input to transform the environmental representation corresponding to the visual localization mapping data into the operational coordinate system. The continuous pose estimation results and the operational coordinate system are transformed by selecting a reference point or calibration point of the operational coordinate system and performing rigid body registration calculation. The coordinate transformation relationship is applied to the continuous pose estimation results to obtain the pose sequence in the operational coordinate system and generate an aligned environment map accordingly. When identifying obstacle areas in the aligned environment map, the boundary of the obstacle area is extracted based on the occupied grid threshold or point cloud clustering, and the free space and obstacle area are distinguished. When expanding the obstacle area, the boundary of the obstacle area is morphologically expanded using the robot's external dimensions and safety margin as the expansion radius to form an impassable buffer zone and obtain a passable region. When fusing the passable region with the operational boundary, the passable region within the operational boundary is clipped and the area outside the operational boundary is removed to output a passable map.

[0050] S2. Perform obstacle semantic classification on the passable map, and integrate dynamic target prediction to construct the time occupancy zone and the time feasible region.

[0051] S2.1. Read the passable map and extract the boundaries of the obstacle area to form a candidate obstacle set. Perform geometric feature calculation and obstacle semantic classification on the candidate obstacle set to generate obstacle semantic labels.

[0052] Specifically, after reading the passable map, the boundary between the occupied area and the free area is located on the passable map, and boundary tracking is performed along the boundary to obtain the boundary of the obstacle area. The boundary of the obstacle area is separated into independent obstacle areas according to the connected component segmentation and summarized to form a candidate obstacle set. When the candidate obstacle set enters the geometric feature calculation, the perimeter, area, aspect ratio of the circumscribed rectangle, and boundary curvature change of each obstacle area boundary are calculated and the geometric feature results are recorded. When the geometric feature results enter the obstacle semantic classification, the candidate obstacle set is judged according to the aspect ratio and curvature change of the circumscribed rectangle. The members of the candidate obstacle set with a large aspect ratio and continuous boundary extension are classified as wall obstacles. The members of the candidate obstacle set with similar aspect ratios and boundary curvature close to circles or rectangles are classified as columnar obstacles. Then, the corresponding category identifier is written for the members of the candidate obstacle set and obstacle semantic labels are generated.

[0053] S2.2. Generate edge-fitting buffers for wall obstacles and detour buffers for column obstacles based on obstacle semantic labels to form semantic safety constraints. Collect the robot's current position and velocity information, and predict the position by time step in the prediction time domain to generate the predicted occupied area.

[0054] Specifically, after the obstacle semantic labels are processed and entered into the buffer zone, they are expanded outward at equal intervals along the boundary of the obstacle area corresponding to the wall obstacle towards the free space side, and the outward expansion boundary is closed and filled to obtain the edge buffer zone. The boundary of the obstacle area corresponding to the column obstacle is expanded outward at equal intervals in the circumferential direction, and the outward expansion boundary is closed and filled to obtain the detour buffer zone. The edge buffer zone and the detour buffer zone are superimposed in the passable map coordinate system, and the superposition result is written into the restricted area to form semantic safety constraints (referring to mapping the semantic categories of different objects / areas in the environment (such as pedestrians, vehicles, walls, columns, steps, restricted areas, work area boundaries, etc.) to different safety rules). The cost / constraint parameters are then used to apply the rules to the feasible region and cost function of the planning, so that the trajectory planning not only avoids obstacles, but also takes different safety behaviors (deceleration, detour, maintaining distance, prohibiting entry, maintaining distance along the edge, etc.) according to semantic differences. After the robot's current position and velocity information are collected, the prediction position is recursively processed. The prediction time domain is divided into continuous time steps, and the robot's current position is displaced and accumulated based on the velocity information in each time step to obtain the time step prediction position sequence. The time step prediction position sequence is spatially enveloped and expanded in the traversable map coordinate system and merged with the restricted area of ​​semantic safety constraints to generate the predicted occupied area.

[0055] S2.3. Map the predicted occupied area and semantic security constraints together to the walkable map to construct the temporal occupied zone. Based on the temporal occupied zone, perform temporal clipping on the walkable map and output the temporal feasible region.

[0056] Specifically, after the predicted occupied area and semantic safety constraints enter the mapping process, the predicted occupied area is covered to the corresponding spatial location in the traversable map coordinate system, and the covered location is marked as the temporal occupied area. At the same time, the semantic safety constraints are covered to the corresponding spatial location, and the covered location is marked as the semantic occupied area. The temporal occupied area and the semantic occupied area are combined to obtain the temporal occupied zone, which is written into the temporal occupied marker layer of the traversable map. After the temporal occupied zone enters the temporal clipping process, the free space of the traversable map is searched grid by grid or polygon by polygon segment. During the search, the temporal occupied area and the semantic occupied area marked by the temporal occupied zone are used as the elimination conditions to delete the free space parts that overlap with the temporal occupied zone. At the same time, the free space parts that do not overlap with the temporal occupied zone are retained, and the free space boundaries are updated. The updated free space and the corresponding boundaries are summarized to form the temporal feasible region.

[0057] S3. Perform coverage partition decomposition on the time feasible domain to generate a set of coverage partitions. At the same time, establish a backfill contract binding graph for the set of coverage partitions and initialize the target layer of the coverage quality field.

[0058] S3.1. Extract the regional boundary information of the time feasible region to form the region to be decomposed. Perform connectivity analysis on the region to be decomposed and perform coverage partition decomposition to form multiple coverage partitions. After summarizing, form a set of coverage partitions.

[0059] Specifically, after the temporal feasible region enters the region boundary extraction process, boundary tracking is performed along the boundary between the free space and non-free space of the temporal feasible region to generate closed boundary lines. The closed boundary lines are separated according to the connected component segmentation to obtain the boundary segments corresponding to each connected free space, and are summarized to form the region boundary information of the temporal feasible region. The region boundary information of the temporal feasible region is used to fill and generate the region to be decomposed and limit the spatial range of the coverage partition decomposition. After the region to be decomposed enters the connectivity analysis process, the region to be decomposed is used to perform reachability retrieval on a grid-by-grid or polygon-by-polygon segment basis using connected component markers. The reachability retrieval obtains the connected parts through adjacency traversal and outputs the boundaries of the connected parts. After the connectivity analysis results enter the coverage partition decomposition process, the region is segmented along the separation position between the connected part boundary and the obstacle boundary. After segmentation, the boundary of each segmented region is reclosed to obtain multiple coverage partitions. The multiple coverage partitions and the corresponding region boundary information are summarized to form a coverage partition set.

[0060] S3.2. Extract the accessible and exitable boundaries from the coverage partition set and calculate the coverage main direction and turning buffer range to obtain the coverage partition planning constraints. Establish the coverage partition adjacency relationship based on the shared boundaries between the coverage partition sets to form the coverage partition switching diagram.

[0061] Specifically, for each coverage partition in the coverage partition set, boundary segment traversal is performed on the corresponding region boundary information. The boundary segments are matched for overlapping segments with the region boundary information of other coverage partition sets. Overlapping segment matching is used to obtain the shared boundary (the segment that overlaps between the region boundary segments of the two coverage partitions) through endpoint coordinate consistency checks and overlap length calculations, generating candidate enterable boundaries and candidate exitable boundaries. Boundary segments for which no shared boundary is obtained are entered into a neighbor search. The neighbor search supplements the candidate enterable and exitable boundaries through minimum distance calculations and connectivity checks within the time feasible domain. These are then summarized to obtain the enterable and exitable boundaries. The main coverage direction is searched through the candidate parallel sweep direction. The candidate parallel sweep direction is obtained by rotating the main axis direction of the region boundary information and then rotating it back and forth along the main axis direction to form a coverage line. The total length of the back and forth coverage line and the number of end turns are used to select the main coverage direction. The turning buffer range is obtained by searching the region boundary information along the normal direction based on the end position of the back and forth coverage line under the main coverage direction and calculating the available gap width. The main coverage direction, the turning buffer range, the enterable boundary and the exitable boundary are combined to obtain the coverage partition planning constraint. The coverage partition planning constraint is a set of constraints composed of the partition boundary and the back and forth coverage characteristics inside the partition. Based on the shared boundary matching result and the neighbor retrieval result, the connection information is written and associated with the coverage partition set to form a coverage partition switching graph.

[0062] S3.3. Based on coverage zoning planning constraints, select the replenishment entry point and replenishment exit point on the accessible and exit boundaries of the coverage zoning, set the replenishment trigger conditions and replenishment termination conditions, generate replenishment contracts, and associate and store the replenishment contracts with the coverage zoning switching diagram to form a replenishment contract binding diagram.

[0063] Specifically, after the coverage zoning planning constraints are processed in the replenishment contract generation process, multiple sets of candidate replenishment entry points are generated along the accessible boundaries of the coverage zoning, and multiple sets of candidate replenishment exit points are generated along the exit boundaries of the coverage zoning. The candidate replenishment entry points and exit points are filtered based on the access position relationship of the reciprocating coverage lines in the main coverage direction and the available space of the turning buffer range, respectively. The filtering process includes calculating the access distance from the candidate replenishment entry point to the first reciprocating coverage line in the main coverage direction and calculating the disengagement distance from the candidate replenishment exit point to the last reciprocating coverage line in the main coverage direction, and verifying whether the turning buffer range can accommodate the passage of the connecting section. The filtering results are written to the replenishment entry point and replenishment exit point. The replenishment triggering condition is based on the coverage quality field and the coverage... The comparison results of the target layer of the quality field are used to set and record the trigger rules corresponding to the coverage quality deviation. The termination condition is based on the comparison results of the coverage quality field and the target layer of the coverage quality field to set and record the termination rules corresponding to the convergence of the coverage quality deviation. The termination entry point, termination exit point, termination trigger condition and termination condition are summarized to generate a termination contract. The connection information of the termination contract and the coverage partition switching graph is associated and stored, and the correspondence between the coverage partition set identifier and the termination entry point and termination exit point is recorded, thus forming a termination contract binding graph. The termination contract binding graph is an associated storage of the connection relationship between the termination contract (entry point, exit point, trigger condition, termination condition) and the coverage partition switching graph, forming a searchable structure of available termination contracts for partition set identifiers / adjacent relationships.

[0064] Furthermore, the coverage quality field target layer is established under the traversable map coordinate system, with the target quality distribution layer consistent with the temporal feasible domain spatial range. The target coverage intensity and target overlap level corresponding to the operation type in the operation quality requirements are written into the target quality distribution layer position by position to form the basic target value. Then, combined with the obstacle semantic tags, the edge buffer range corresponding to the wall obstacle and the detour buffer range corresponding to the column obstacle are mapped to the target quality distribution layer, and the corresponding target quality requirements are written into the edge buffer range and detour buffer range. The operation quality requirements are obtained by first finding the target coverage intensity and target overlap level according to the operation type (such as cleaning, spraying, disinfection, etc.), then calculating the range of the edge buffer and detour buffer according to the obstacle type and geometric size, and applying the corresponding quality target mapping rules. The basic coverage target edge / detour area targets are summarized into a set of parameters.

[0065] S4. Perform global coverage sequence planning on the backfill contract binding graph to generate a global trajectory segment sequence.

[0066] S4.1. Calculate the switching cost between coverage partitions based on the coverage partition switching graph in the backfill contract binding graph and merge it with the expected coverage cost within the coverage partition to form a global planning cost. Using the coverage partition where the starting position is located as the starting point, generate the access order of coverage partitions through global planning cost constraints.

[0067] Specifically, after the coverage partition switching map in the backfill contract binding diagram enters the switching cost calculation process, the connection information of the coverage partition switching map is used to locate the backfill exit point and backfill entry point corresponding to the adjacent coverage partitions. The backfill exit point and backfill entry point are searched on the passable map to obtain the switching cost between coverage partitions by calculating the length of the passable path and the number of turns. The expected coverage cost within the coverage partition is obtained by calculating the total length of the reciprocating coverage line and the number of end connection segments through the generation result of the reciprocating coverage line corresponding to the main coverage direction. The switching cost and the expected coverage cost are fused and written into the fusion result for adjacent coverage partitions to form the global planning cost. After the coverage partition where the starting position is located is used as the access starting point, the global planning cost is used to perform stepwise selection of unvisited coverage partitions. In the stepwise selection process, the global planning cost from the current coverage partition to each unvisited coverage partition is compared at each step, and the coverage partition with the smaller global planning cost (for example, the entry point of partition A is 2m away from the current robot, and the exit point can be directly connected to the shared boundary of the next partition) is selected as the next access coverage partition. After the entire set of coverage partitions is included in the access sequence, the access order of the coverage partitions is output.

[0068] It should be noted that the expression for calculating the global planning cost fusion is as follows:

[0069] ;

[0070] in, Indicates from the overlay partition To cover partition Path cost, Indicates from the overlay partition To cover partition The cost of switching at that time Indicates covering partition The estimated cost of internal coverage. This represents the number or index of a specific covering partition within the set of covering partitions. This indicates the number or index of a specific overlay partition within the overlay partition set.

[0071] S4.2. For each coverage partition in the coverage partition access sequence, generate a trajectory segment within the partition based on the main coverage direction. Based on the backfill exit point and backfill entry point, generate a switching trajectory segment. Connect the trajectory segments within the partition and the switching trajectory segment and associate them with the corresponding backfill contracts to generate a global trajectory segment sequence.

[0072] Specifically, after inputting the coverage partition access order, the main coverage direction is read for each coverage partition in the access order, and a reciprocating coverage line parallel to the main coverage direction is generated within the free space defined by the area boundary information of the coverage partition. The reciprocating coverage line is obtained by projecting the coverage partition boundary along the main coverage direction and maintaining the coverage spacing between adjacent reciprocating coverage lines. The end of the reciprocating coverage line generates a connecting segment based on the turning buffer range, and the reciprocating coverage line and the connecting segment are combined to obtain the trajectory segment within the partition. The backfill exit point of the previous coverage partition and the backfill entry point of the next coverage partition are read between adjacent coverage partitions, and a travel path is searched on the traversable map. The travel path search avoids the occupied area marked by time occupation and outputs the travel path from the backfill exit point to the backfill entry point. Then, the travel path is subjected to turning smoothing processing to form a switching trajectory segment. The trajectory segments within the partition and the switching trajectory segments are concatenated end to end according to the coverage partition access order, and the trajectory segments within each partition are associated with the corresponding backfill contract according to the backfill contract binding map to generate a global trajectory segment sequence.

[0073] S5. Perform trajectory tracking and local rerouting based on the global trajectory segment sequence, and update the coverage quality field online to trigger skipped segments and write them to the queue to be replenished.

[0074] S5.1. Read the global trajectory segment sequence and schedule the current trajectory segments to be executed in sequence. Based on the robot's current pose state and the current trajectory segments to be executed, generate motion control commands and perform trajectory tracking.

[0075] Specifically, the first element of the global trajectory segment sequence is extracted as the current trajectory segment to be executed, according to the order of the global trajectory segment sequence. In each control cycle, the correspondence between the robot's current pose state and the current trajectory segment to be executed is updated. The robot's current pose state is projected onto the current trajectory segment to be executed to obtain the target tracking point and the target heading. Then, the linear velocity command and angular velocity command are calculated from the lateral deviation between the robot's current pose state and the target tracking point and the heading deviation between the robot's current pose state and the target heading, and combined to form motion control commands. After the motion control commands are issued, the robot moves along the current trajectory segment to be executed and continuously refreshes the robot's current pose state to correct the deviation in a closed loop. When the robot's current pose state reaches the neighborhood of the termination position of the current trajectory segment to be executed, the next element of the global trajectory segment sequence is updated as the current trajectory segment to be executed, and the process of generating motion control commands is repeated to achieve trajectory tracking.

[0076] S5.2. During trajectory tracking, continuously acquire environmental observation data and verify its consistency with the time occupancy zone to generate a local rerouting trajectory.

[0077] Specifically, during trajectory tracking, environmental observation data is continuously collected and converted into an occupied area representation in the collapsible map coordinate system. The occupied area representation corresponding to the environmental observation data is spatially overlapped with the time-occupied zone to complete consistency verification. Consistency verification is achieved by statistically analyzing the overlap range between the occupied area representation corresponding to the environmental observation data (2D / 3D LiDAR point cloud or scan line, depth camera depth map, millimeter-wave radar target point, ultrasonic ranging, etc.) and the occupied area marked by the time-occupied zone, and comparing it with the collapsible area ahead of the current trajectory segment to be executed to obtain the conflict position. After the conflict position is output, the robot's current pose is used as the starting point and the reconnection position of the current trajectory segment to be executed after the conflict position is used as the ending point. Under the constraints of the collapsible map, a collapsible path search is performed around the conflict position. During the collapsible path search, the occupied area marked by the time-occupied zone is removed and the non-occupied area is retained as the optional collapsible space. The collapsible path output by the collapsible path search is smoothed by turning to form a local rerouting trajectory.

[0078] It should be noted that, as Figure 5 As shown, the minimum distance between the robot and the dynamic target is statistically distributed with the minimum dynamic target distance (meters) as the horizontal axis and the count as the vertical axis. This is used to characterize the overall level and fluctuation range of the robot maintaining a safe distance during the execution phase after constructing the predicted occupied area—temporal occupied zone—temporal feasible region and performing conflict detection / avoidance. When the distribution is generally biased towards larger distances and the counts are fewer in the low distance interval, it can statistically reflect the spatiotemporal integrated safety constraints and forward collision avoidance.

[0079] S5.3. During trajectory execution, continuously acquire job feedback data and map it to the job area to obtain the coverage quality field quality value. Update the coverage quality field by comparing the target layer of the coverage quality field and the coverage quality field quality value, and write the current trajectory segment to be executed into the queue to be replenished.

[0080] Specifically, during trajectory execution, continuous collection of operation feedback data (trajectory mileage and pose projection onto the operation grid, "arrival / passage" records; visual / depth estimations of ground traces, wetness, spray texture, cleanliness, etc.) is performed, and the robot's current pose state and the position of the current trajectory segment to be executed are read simultaneously. The operation feedback data is aligned with the robot's current pose state according to the timestamp and then projected onto the spatial position of the operation area. The projected position within the operation area accumulates and normalizes the operation feedback data to obtain the coverage quality field quality value. The coverage quality field target layer and the coverage quality field quality value are compared position by position at the same spatial position. The comparison result is used to replace or incrementally update the coverage quality field quality value at the same spatial position in the coverage quality field to complete the coverage quality field update. The current trajectory segment to be executed and the supplementary contract identifier associated with the current trajectory segment to be executed in the supplementary contract binding diagram are written into the supplementary queue.

[0081] It should be noted that the expression for calculating the coverage quality difference is:

[0082] ;

[0083] in, Indicates the location Coverage quality deviation Indicates the location of the target layer of the covering mass field. The target quality value on the surface, Indicates the location The actual quality value achieved Indicates the position index.

[0084] S6. Based on the queue to be replenished, perform queue insertion planning according to the replenishment contract binding diagram, generate replenishment trajectory, drive the robot to perform replenishment and acceptance, and output a work report.

[0085] S6.1. Read the queue to be replenished and obtain the replenishment entry point and replenishment exit point associated with the skipped segment. At the same time, read the replenishment contract binding graph to obtain the set of overlay partitions and the overlay partition switching graph associated with the skipped segment.

[0086] Specifically, after the pending replacement queue enters the parsing process, the skipped segment identifier is read for each skipped segment in the pending replacement queue, and the replacement contract record in the replacement contract binding graph is retrieved according to the skipped segment identifier. The replacement contract record contains the replacement entry point and the replacement exit point, and the replacement entry point and the replacement exit point are written into the skipped segment association information. After the replacement contract binding graph enters the association reading process, the cover partition set identifier corresponding to the skipped segment in the replacement contract binding graph is located based on the skipped segment identifier, and the regional boundary information of the cover partition in the cover partition set is read. The connection information associated with the cover partition set identifier in the replacement contract binding graph is read and a cover partition switching graph is formed.

[0087] S6.2. Calculate the queueing cost from the robot's current position to the queueing entry point based on the queueing contract binding graph and generate the queueing planning result. Select the skipped segment corresponding to the queueing planning result as the current queueing task.

[0088] Specifically, after the robot's current position is input, it traverses each skipped segment in the queue to be replenished and reads the replenishment entry point corresponding to the skipped segment from the replenishment contract binding graph and the coverage partition switching graph connection information. The insertion cost from the robot's current position to the replenishment entry point is obtained by searching for a passable path on the passable map. The passable path search takes the occupied area marked by the time occupation zone as the impassable area and outputs the passable path from the robot's current position to the replenishment entry point. The passable path length and the number of passable path turns are used to calculate the insertion cost and merge it with the coverage partition switching cost recorded in the coverage partition switching graph connection information to obtain the skipped segment insertion cost. All skipped segment insertion costs are sorted according to their numerical values ​​to generate the insertion planning result. The skipped segment with the highest insertion planning result is selected as the current replenishment task.

[0089] S6.3. Based on the current replenishment task's replenishment entry point and replenishment exit point, generate a replenishment trajectory, insert a global trajectory segment sequence to form an interleaved execution sequence, drive the robot to execute the replenishment trajectory, generate an acceptance conclusion based on the coverage quality field and the coverage quality field target layer, and output a work report.

[0090] Specifically, after the current recovery task is input, the recovery entry point and recovery exit point corresponding to the current recovery task are read, and a path search is performed on the passable map to generate a path from the recovery entry point to the recovery exit point. The path search marks the occupied area of ​​the time occupation zone as the impassable area and outputs the passable path in the non-occupied area. After the passable path enters the turning smoothing process, the recovery trajectory is formed. The recovery trajectory and the global trajectory segment sequence are inserted at the current execution position. The global trajectory segment sequence insertion operation is obtained by placing the recovery trajectory before the current trajectory segment to be executed and keeping the order of the other trajectory segments unchanged. After the queued execution sequence enters the execution process, the recovery trajectory is scheduled according to the queued execution sequence and motion control instructions are generated to drive the robot to execute the recovery trajectory. At the same time, the operation feedback data is collected and mapped to the operation area to update the coverage quality field. The coverage quality field and the coverage quality field target layer are compared position by position in the operation area and the comparison results are summarized to obtain the acceptance conclusion. The acceptance conclusion and the queued execution sequence execution record are summarized to generate an operation report.

[0091] It should be noted that the expression for calculating the cost of the backtracking path is:

[0092] ;

[0093] in, This represents the path cost from the replenishment entry point to the replenishment exit point. The weighting coefficient represents the length of the recovery path. Indicates the length of the backfill path. The weighting coefficient representing the number of turns. This indicates the number of turns in the path.

[0094] Furthermore, such as Figure 6 As shown, with the prediction time window (seconds) as the horizontal axis and the percentage of compliant area as the vertical axis, the curves of the percentage of compliant area under different parameter groups (parameter groups 1 to 4) are given. For example, the discrete time of parameter group 1 (the baseline group) is 0.5s and the expansion radius is 0.25m; the discrete time of parameter group 2 (a finer time granularity) is 0.2s and the expansion radius is 0.25m; the discrete time of parameter group 3 (a more conservative safety margin) is 0.5s and the expansion radius is 0.4m; and the discrete time of parameter group 4 (a finer granularity + a more conservative) is 0.2s and the expansion radius is 0.4m. One group is used to compare the time discrete granularity, the other to compare the safety margin, and the other to compare the difference in the percentage of compliant area caused by the superposition of the two. The upper figure is an overview, and the window interval with more obvious differences is marked with red dashed rectangles. The lower figure is a magnified comparison of the corresponding window, and the difference points and feature points are marked with dashed lines and arrows in the magnified figure to make it easier to intuitively present the trend and difference position of the percentage of compliant area change with parameter setting changes.

[0095] This embodiment also provides a computer device applicable to robot trajectory planning methods, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the robot trajectory planning method proposed in the above embodiment.

[0096] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0097] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the robot trajectory planning method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0098] In summary, this invention generates a passable map through visual positioning and coordinate alignment, achieving a unified expression of the work space and planarable boundary constraints. Furthermore, it constructs a time-occupied zone and a time-feasible domain by integrating obstacle semantic classification with dynamic target prediction, forming a process control capability that runs through planning, execution, and acceptance. This enables spatiotemporal integrated safety constraints and forward-looking collision avoidance, improving coverage integrity, continuous operation, and delivery reliability.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for planning robot operation trajectories, characterized in that, include: Collect visual positioning and mapping data, perform coordinate alignment and generate traversable regions, and output a traversable map; Obstacle semantic classification is performed on the passable map, and dynamic target prediction is integrated to construct the time occupancy zone and the time feasible region; The time feasible domain is decomposed into coverage partitions to generate a set of coverage partitions. At the same time, a backfill contract binding graph is established for the set of coverage partitions, and the target layer of the coverage quality field is initialized. The steps for decomposing the time feasible domain into covered partitions to generate a set of covered partitions, and simultaneously establishing a backfill contract binding graph for the set of covered partitions, are as follows: Extract the regional boundary information of the time feasible region to form the region to be decomposed. Perform connectivity analysis on the region to be decomposed and perform coverage partition decomposition to form multiple coverage partitions. After summarizing, form a set of coverage partitions. The accessible and exit boundaries are extracted from the set of coverage partitions, and the main coverage direction and turning buffer range are calculated to obtain the coverage partition planning constraints. The adjacency relationship of coverage partitions is established based on the shared boundaries between the sets of coverage partitions to form a coverage partition switching diagram. Based on coverage zoning planning constraints, select replenishment entry points and replenishment exit points on the accessible and exit boundaries of the coverage zoning, set replenishment trigger conditions and replenishment termination conditions, generate replenishment contracts, and associate and store the replenishment contracts with the coverage zoning switching diagram to form a replenishment contract binding diagram. The coverage quality field target layer is established in the traversable map coordinate system, which is consistent with the spatial range of the time feasible domain. The target coverage intensity and target overlap level of the corresponding operation type in the operation quality requirements are written into the target quality distribution layer position by position to form the basic target value. Then, combined with the obstacle semantic label, the edge buffer range corresponding to the wall obstacle and the detour buffer range corresponding to the column obstacle are mapped to the target quality distribution layer, and the corresponding target quality requirements are written into the edge buffer range and the detour buffer range. Perform global coverage sequence planning on the contract binding graph to generate a global trajectory segment sequence; The specific steps for performing global coverage sequence planning on the backfill contract binding graph to generate a global trajectory segment sequence are as follows: Based on the coverage partition switching graph in the backfill contract binding graph, the switching cost between coverage partitions is calculated and merged with the expected coverage cost within the coverage partition to form a global planning cost. Starting from the coverage partition where the starting position is located, the access order of coverage partitions is generated by constraining the global planning cost. For each coverage partition in the coverage partition access order, combined with the main coverage direction, a trajectory segment within the partition is generated. Based on the backfill exit point and backfill entry point, a switching trajectory segment is generated. The trajectory segments within the partition and the switching trajectory segments are concatenated and associated with the corresponding backfill contracts to generate a global trajectory segment sequence. Trajectory tracking and local rerouting are performed based on global trajectory segment sequences, and the overlay mass field is updated online to trigger skipped segments and write them to the queue to be replenished; The specific steps for performing trajectory tracking and local rerouting based on the global trajectory segment sequence, and updating the coverage mass field online to trigger skipped segments and write them to the queue to be replenished are as follows: Read the global trajectory segment sequence and schedule the current trajectory segments to be executed in sequence. Based on the robot's current pose state and the current trajectory segments to be executed, generate motion control commands and perform trajectory tracking. During trajectory tracking, environmental observation data is continuously acquired and its consistency with the time occupancy zone is verified to generate a local rerouting trajectory. During trajectory execution, the operation feedback data is continuously acquired and mapped to the operation area to obtain the coverage quality field quality value. The coverage quality field is updated by comparing the target layer of the coverage quality field and the coverage quality field quality value, and the current trajectory segment to be executed is written into the queue to be replenished. Based on the queue to be replenished and the replenishment contract binding diagram, the robot is used to plan the queue insertion, generate the replenishment trajectory, drive the robot to perform replenishment and acceptance, and output the operation report. The steps are as follows: Based on the queue to be replenished and the replenishment contract binding diagram, queue planning is performed to generate a replenishment trajectory, and the robot is driven to perform replenishment and acceptance, outputting a work report. Read the queue to be replenished and obtain the replenishment entry point and replenishment exit point associated with the skipped segment. At the same time, read the replenishment contract binding graph to obtain the set of overlay partitions and the overlay partition switching graph associated with the skipped segment. Based on the contract binding graph, calculate the queueing cost from the robot's current position to the queueing entry point and generate the queueing planning result. Select the skipped segment corresponding to the queueing planning result as the current queueing task. Based on the current replenishment task's replenishment entry point and replenishment exit point, a replenishment trajectory is generated, and a global trajectory segment sequence is inserted to form an interleaved execution sequence. This sequence drives the robot to execute the replenishment trajectory and generates an acceptance conclusion based on the coverage quality field and the coverage quality field target layer, outputting a work report.

2. The robot trajectory planning method as described in claim 1, characterized in that, The specific steps for collecting visual positioning and mapping data, performing coordinate alignment and generating traversable regions, and outputting a traversable map are as follows: The robot uses cameras and inertial measurement to collect continuous images and inertial data, synchronously acquires odometry information, forms visual positioning mapping data, performs time synchronization and abnormal frame removal on the visual positioning mapping data, and extracts positioning feature information to obtain continuous pose estimation results. The continuous pose estimation results are registered with the operation coordinate system to form an aligned environment map. Obstacle areas are identified and expanded on the aligned environment map to generate a passable region. The passable region is then fused with the operation boundary to output a passable map.

3. The robot operation trajectory planning method as described in claim 2, characterized in that, The specific steps for performing obstacle semantic classification on the passable map and fusing dynamic target prediction to construct the time occupancy zone and time feasible region are as follows: Read the passable map and extract the boundaries of the obstacle area to form a candidate obstacle set. Perform geometric feature calculation and obstacle semantic classification on the candidate obstacle set to generate obstacle semantic labels. Based on the semantic labels of obstacles, edge-fitting buffers are generated for wall obstacles and detour buffers are generated for column obstacles to form semantic safety constraints. The robot’s current position and velocity information are collected, and the predicted position is recursively calculated by time step in the prediction time domain to generate the predicted occupied area. The predicted occupied area and semantic security constraints are jointly mapped to the walkable map to construct the temporal occupied zone. The walkable map is then clipped in the temporal domain based on the temporal occupied zone to output the temporal feasible region.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the robot operation trajectory planning method according to any one of claims 1 to 3.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the robot operation trajectory planning method according to any one of claims 1 to 3.

Images