A picking-transporting robot cooperative dynamic path planning system and method for hilly disorder orchard

Abstract

The application discloses a picking-transportation robot cooperative dynamic path planning system and method for hilly disorder orchard, and belongs to the technical field of agricultural robots. The method comprises the following steps: based on the estimated yield of fruit trees and the upper limit of the carrying capacity of a transport vehicle, yield matching clustering is carried out, so that the total yield of each operation cluster is close to the carrying capacity threshold, and energy consumption and time waste caused by return without full load are avoided; the position of a unloading point is optimized through a dynamic weighted shortest path model, and completed clusters are excluded along with the operation progress; a cooperative sequential planning strategy is adopted, the fruit trees in the cluster are traversed in the cluster in the order of minimum distance and slope energy consumption, and the nearest switching or automatic vehicle replacement is adopted according to the carrying state between clusters. In addition, a slope danger edge exclusion mechanism is introduced, the maximum safe climbing angle and the maximum safe roll angle of the tracked chassis transport vehicle are set, and the over-limit path is directly disabled. The application significantly improves the fuel economy and time efficiency of the picking-transportation cooperation in the hilly orchard, and is suitable for complex terrain and yield uncertainty.

Description

Technical Field

[0001] This invention belongs to the field of agricultural robot technology, specifically relating to a collaborative dynamic path planning system and method for picking vehicles and transport vehicles in hilly and disordered orchards. It is particularly suitable for collaborative harvesting and transporting operations in orchards where mangoes, citrus fruits, and other fruits are planted in hilly areas and the fruit trees are irregularly distributed. Background Technology

[0002] In hilly areas, orchards often have irregularly distributed fruit trees, narrow roads, steep slopes, small turning radii, and road conditions that change drastically with the weather. Taking mango orchards as an example, mango trees have large canopies, uneven fruit distribution, and significant differences in yield per tree. When transport vehicles are fully loaded, the fruit is heavy and has a high center of gravity, making them extremely sensitive to safety indicators such as slope and lateral tilt along the transport path. Existing harvesting and transportation route planning schemes have the following main shortcomings:

[0003] (1) Existing solutions are mostly based on the design of row orchards in plains or terraced orchards, which cannot adapt to the scattered characteristics of disordered orchards in hilly areas. The path planning of row orchards in plains relies on the regular structure of fruit trees arranged in rows, while terraced orchards use elevation stratification for serial planning. In disordered orchards in hilly areas, fruit trees do not have fixed row spacing or tree spacing, and the passable area is affected by slope, ditches, tree canopy obstruction, temporary piles of materials and slippery soil, making the above-mentioned regularization methods difficult to apply.

[0004] (2) The capacity configuration of transport vehicles is out of sync with the fruit tree yield distribution, resulting in serious waste on full-load return trips. Existing solutions usually assume that the yield of each fruit tree is uniform or adopt a fixed picking order, without taking into account the yield differences of individual fruit trees into clustering and route planning. In hilly mango orchards, the yield of different trees can vary by several times due to factors such as sunlight, tree age, and variety. If picking is carried out in a fixed order, transport vehicles often have to return before being fully loaded or are forced to transport in multiple trips due to overloading, resulting in a significant increase in empty mileage and energy consumption.

[0005] (3) The unloading point is fixed and cannot be dynamically optimized with the progress of the operation. The existing solution usually sets the unloading point (drone hoisting point or ground unloading point) at a fixed position on the edge of the orchard. As the harvest progresses, the return distance of the transport vehicle with a full load gradually increases, and energy consumption and efficiency continue to deteriorate. Summary of the Invention

[0006] This invention addresses the technical challenges of irregular fruit tree distribution, high energy consumption of transport vehicles returning fully loaded in hilly orchards, and significant safety hazards when driving on slopes. It provides a collaborative dynamic path planning system and method for harvesting and transport vehicles in hilly orchards. By organically combining yield matching and clustering, dynamic unloading point optimization, collaborative sequence planning, and slope hazard edge elimination, collaborative harvesting and transport path planning is achieved.

[0007] To solve the above technical problems, the specific technical solution adopted by the present invention is as follows:

[0008] A collaborative dynamic path planning method for picking and transporting robots in hilly, unordered orchards:

[0009] Obtain orchard semantic map information, which includes at least the location of fruit trees, terrain elevation, estimated yield of each fruit tree, passable area information, and candidate locations for unloading points;

[0010] The hilly operation map is constructed based on vehicle passability constraints, and each edge is assigned a passability cost related to the load. The passability constraints include at least road width and slope. Edges in the hilly operation map that exceed the maximum safe climbing angle and the maximum safe side tilt angle are marked as impassable.

[0011] Yield-matched clustering is based on graph distances between fruit trees and the full load capacity of transport vehicles. Unharvested fruit trees are dynamically divided into several work clusters, and the sum of the cumulative estimated yield upper limit of fruit trees in each work cluster is an integer multiple of the full load capacity of the transport vehicle.

[0012] Dynamic planning of unloading points optimizes the location of unloading points across the entire orchard, minimizing the sum of weighted costs from the meeting point of all currently unharvested work clusters to their optimal unloading point, where the weights are the expected total yield of each work cluster; once a work cluster has completed harvesting, it is automatically excluded from subsequent unloading point optimizations.

[0013] Collaborative sequence planning aims to minimize total travel distance and total energy consumption. It plans the fruit tree harvesting sequence within a work cluster and the switching sequence between work clusters, and automatically triggers the replacement process when the transport vehicle is fully loaded.

[0014] Furthermore, the output matching clustering specifically refers to:

[0015] Calculate the graph distance between all unpicked fruit trees, that is, the shortest path length of the unpicked fruit trees on the hilly operation map;

[0016] Initial clustering is performed based on the graph distance, ensuring that the graph distance between any two fruit trees within the same cluster is less than the intra-cluster distance threshold. This yields the initial candidate clusters;

[0017] Calculate the sum of the estimated yield caps of all fruit trees within each candidate cluster. ;

[0018] Determine the minimum number of transport vehicles required for the cluster. ;

[0019] like The cluster is retained, by One transport vehicle or the same vehicle The harvesting is completed in a relay within the cluster; among them The maximum number of transport vehicles allowed;

[0020] like The cluster is divided into several sub-clusters according to its spatial distribution, such that each sub-cluster has... Value not exceeding ;

[0021] For production less than Clusters whose distance to neighboring graphs is less than the inter-cluster distance threshold are considered. Cluster merging, the output after merging does not exceed .

[0022] Furthermore, the yield matching clustering also considers the operating efficiency of the harvesting vehicle: if the estimated harvesting time of a cluster exceeds a preset threshold... If the cluster is split into smaller sub-clusters, the estimated harvesting time is estimated by dividing the sum of the maximum yield by the average harvesting efficiency.

[0023] Furthermore, the dynamic programming of the unloading point determines the optimal unloading point location by solving the following weighted distance optimization problem:

[0024]

[0025] in: For the set of candidate unloading points, This is a collection of harvesting clusters that have not yet been completed. For the first The geometric center of each job cluster As weight, For the hilly operation map, from the candidate unloading point To the cluster center The shortest path distance.

[0026] Furthermore, the passage cost is:

[0027]

[0028] in, Let be the side length. , These are the weighting coefficients. For the empty load weight of the transport vehicle, It is the acceleration due to gravity. For the current load, This represents the elevation change on the uphill slope.

[0029] Furthermore, the fruit tree harvesting order within the job cluster is planned as follows: taking the meeting point of the job cluster as the starting point, calculate multiple path sequences for traversing all fruit trees from the current point, and select the sequence with the minimum cumulative travel cost.

[0030] Furthermore, real-time monitoring of the remaining capacity of transport vehicles. When deciding on the next harvest target, priority should be given to selecting from the unharvested trees in the current work cluster, with an estimated yield ceiling not exceeding [a certain value]. The fruit trees; if there are multiple fruit trees that meet the conditions, select the one with the lowest passage cost; if the estimated lower limit of the yield of all unharvested fruit trees is greater than If the current transport vehicle is determined to be unable to continue operating, a replacement process is triggered: the current transport vehicle proceeds to the unloading point, and an idle transport vehicle takes over to continue harvesting the remaining fruit trees in the current cluster.

[0031] Furthermore, the switching order between the work clusters is planned as follows: when a work cluster is completed, if the current following transport vehicle is not fully loaded, the adjacent work cluster with the shortest distance and the lowest total passage cost is selected as the next target; if it is fully loaded, the takeover process is triggered immediately.

[0032] Furthermore, the maximum safe climbing angle and the maximum safe side tilt angle are determined according to the ground conditions: for soft soil or slippery road surfaces, they are set to 15° and 10° respectively; for hardened road surfaces or dry hard soil, they are set to 20° and 15° respectively.

[0033] A collaborative dynamic path planning system for picking and transporting robots in hilly, unordered orchards includes:

[0034] The orchard semantic map module stores terrain elevation and passable areas, fruit tree locations and estimated yields, and candidate unloading points;

[0035] The online planning module includes a hilly operation map construction module, a slope hazard edge elimination module, a yield matching clustering module, a loading point dynamic planning module, a collaborative sequence planning module, and a transport vehicle dynamic docking module. The hilly operation map construction module is used to construct a graph structure with passage costs based on vehicle passability constraints. The slope hazard edge elimination module is used to directly disable dangerous road sections that exceed the limits based on safety angle thresholds. The yield matching clustering module is used to divide fruit trees into operation clusters with appropriate total yield according to graph distance and load limit. The loading point dynamic planning module is used to dynamically optimize the location of loading points according to the operation progress using a weighted shortest path model. The collaborative sequence planning module is used to plan the optimal traversal order of picking within a cluster and the switching order between clusters. The transport vehicle dynamic docking module is used to intelligently make decisions and dispatch idle vehicles to replace fully loaded vehicles based on real-time load status.

[0036] The execution and feedback module is used to execute the job instructions issued by the system and upload the status in real time.

[0037] Compared with existing technologies, the present invention has the following significant advantages:

[0038] (1) Yield-matched clustering to avoid half-load waste. By matching the estimated yield of fruit trees with the upper limit of the transport vehicle's load capacity, the total yield of each working cluster is close to an integer multiple of the load capacity, minimizing the return trip of transport vehicles that are not fully loaded. Simulation results show that compared with fixed-sequence harvesting, the empty mileage is reduced by 40% to 60%.

[0039] (2) Dynamic unloading point optimization to reduce return trip energy consumption. The unloading point location is dynamically optimized by using a weighted shortest path model. The weight is taken from the cluster output, so that high-yield clusters are close to the unloading point, and completed clusters are excluded as the operation progresses. Compared with fixed boundary unloading points, the total return trip mileage is reduced by 15%~25% when fully loaded.

[0040] (3) Eliminate dangerous slope edges to ensure safe driving on slopes. Directly disable slope edges that exceed the limit, rather than just imposing weighted penalties, to completely prevent transport vehicles from entering dangerous sections. Set a 20° climbing and 15° side tilt safety threshold for tracked chassis transport vehicles, and adaptively reduce them on soft and slippery surfaces to eliminate the risk of rollover and slippage from the path planning level.

[0041] (4) Collaborative sequence planning takes into account both efficiency and load status. Within a cluster, traversal is performed with the energy consumption minimized by distance and slope. Between clusters, intelligent decision-making is made based on load status - if not fully loaded, the nearest vehicle is switched; if fully loaded, the vehicle is automatically switched back to the warehouse, achieving seamless connection between collection and transportation. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0043] Figure 2 This is a schematic diagram of the operation cluster division and optimal unloading point planning of the present invention;

[0044] Figure 3 This is a flowchart of the overall operation method of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0046] I. System Hardware Configuration and Preliminary Preparations

[0047] This example uses a mango orchard in a hilly area of ​​southern my country as an example. The orchard covers a total area of ​​approximately 50 mu (about 3.3 hectares), with about 200 mango trees scattered irregularly. The terrain within the orchard is significantly undulating, with a maximum slope of about 25°, and some areas have transverse slopes (lateral inclines). The orchard's road network consists of dirt roads between rows and access paths, with road widths ranging from 1.5 to 3.0 meters.

[0048] Each transport vehicle (i.e., transport robot) is equipped with a fruit storage box, a weight sensor, and a tracked chassis that serves as the walking mechanism. The weight sensor is used to monitor the cumulative load inside the fruit storage box in real time. The transport vehicle is 1.2m wide, has a maximum safe climbing angle of 20°, a maximum safe side tilt angle of 15°, and a full load capacity of the fruit storage unit. Average harvesting efficiency of harvesting robots The transport vehicle has a maximum load capacity of only 50kg. Assuming an average yield of 10kg per mango tree, a return trip is required after harvesting every 5 trees, placing extremely high demands on clustering accuracy and succession efficiency. In this embodiment, the estimated yield range for each mango tree is set at 5-15kg, with a desired yield of 10kg.

[0049] Before data collection, a lidar sensor must be securely installed on the drone or relevant mapping platform, and an RTK-GNSS base station must be set up in an open area of ​​the orchard to provide differential signals. Simultaneously, an RTK-GNSS rover receiver should be configured on the mapping equipment or handheld marker pole. Before the operation begins, a drone equipped with lidar will perform a 3D scan of the orchard to obtain a digital elevation model (DEM), the location of fruit trees, and the projection range of the tree canopy. RTK-GNSS will be used to accurately locate each fruit tree, and the estimated yield (including lower limit, expected value, and upper limit) for each tree will be obtained through historical yield data or a visual yield estimation model. Impassable areas (ditches, waterlogged areas, temporary storage areas) and candidate unloading points (flat terrain, locations easy for drone take-off and landing or vehicle parking) will be manually marked. All information will be stored in the orchard semantic map module.

[0050] Number of transport vehicles configured in the initial stage of operation The platform is parked in the waiting area for the shuttle, with empty fruit crates ready to go.

[0051] like Figure 1As shown, the collaborative dynamic path planning system of the present invention includes an orchard semantic map module, an online planning module, and an execution and feedback module. The orchard semantic map module stores terrain elevation and passable areas, fruit tree locations and estimated yields, and candidate unloading points. The online planning module includes a hilly operation map construction module, a slope hazard edge elimination module, a yield matching clustering module, an unloading point dynamic planning module, a collaborative sequence planning module, and a transport vehicle dynamic connection module. Specifically, the hilly operation map construction module is used to construct a graph structure with passage costs based on vehicle passability constraints; the slope hazard edge elimination module is used to directly disable dangerous road sections exceeding limits based on safety angle thresholds; and the yield matching module... The clustering module divides fruit trees into operational clusters with appropriate total yield based on map distance and load limit. The unloading point dynamic planning module dynamically optimizes the unloading point location according to the operation progress using a weighted shortest path model. The collaborative sequence planning module plans the optimal traversal order of picking within a cluster and the switching order between clusters. The transport vehicle dynamic docking module makes intelligent decisions based on real-time load status and dispatches idle vehicles to replace fully loaded vehicles. The execution and feedback module executes the operation instructions issued by the system and uploads the status in real time. Specifically, it controls the picking robot to pick, controls the docking and transfer of transport vehicles, and obtains the cumulative weight of the fruit in the fruit storage box through weight sensors and provides real-time feedback.

[0052] II. Construction of Hilly Operation Map

[0053] The system discretizes the orchard area into a terrain grid with a resolution of 0.5m, overlaying fruit tree service points, obstacles, unloading points, and candidate meeting points. For any two adjacent nodes... , A walkable edge is established when the following conditions are met simultaneously. :

[0054] Road width m (vehicle width 1.2m, safety margin factor 1.2);

[0055] slope (Unloaded state);

[0056] transverse slope angle (Unloaded state);

[0057] Surface adhesion coefficient (Not in a soft and slippery state).

[0058] Based on the maximum safe climbing angle and maximum safe tilt angle of the transport vehicle, edges exceeding the threshold on the hilly operation map are directly marked as impassable (red mark) and will not participate in subsequent path planning; the maximum safe climbing angle and maximum safe tilt angle can be calibrated according to the actual tracked chassis model and ground conditions: for the soft soil or slippery road surface involved in this embodiment, the angles are reduced to 15° and 10° respectively; for hardened road surface or dry hard soil, they are restored to 20° and 15°.

[0059] The travel cost of each walkable edge (including distance cost and slope-related energy cost) is calculated using the following formula:

[0060]

[0061] in, Side length (m), weight coefficient Unloaded weight of transport vehicle kg, gravitational acceleration m / s², The current load (kg) The elevation change is for uphill slopes (m, 0 for downhill slopes).

[0062] This formula allows a fully loaded transport vehicle to obtain a much higher travel cost on uphill sections than when unloaded, thus automatically favoring gentler routes during path searching.

[0063] III. Yield-matched clustering

[0064] The system calculates the graph distance (i.e., the shortest path length on the hilly operation map) between all 200 unharvested fruit trees, performs initial clustering, and ensures that the graph distance between any two fruit trees within the same cluster is less than the cluster distance threshold. After obtaining the initial candidate clusters, calculate the sum of the estimated upper limits of yield for all fruit trees within each candidate cluster. Matched in units of 50kg, see [link / reference]. Figure 2 Clustering example:

[0065]

[0066] Table In this embodiment, the maximum number of transport vehicles allowed is set to 3.

[0067] For clusters A, B, and C, the following conditions are met: If so, the cluster is retained, and its yield is approximately The following is by One transport vehicle (or the same vehicle) (The harvesting is completed in a relay within the cluster)

[0068] For cluster D (270kg, equivalent to 5.4 times the capacity), it satisfies Therefore, it is divided into 3 sub-clusters according to spatial distribution, which makes the minimum number of transport vehicles required for each sub-cluster... Value not exceeding And the outputs of each sub-cluster are as close as possible. Specifically: cluster D1 (approximately 90 kg), cluster D2 (approximately 90 kg), and cluster D3 (approximately 90 kg).

[0069] For production less than Clusters that are less than the inter-cluster distance threshold are attempted to connect with neighboring graphs. Cluster merging, ensuring the merged output does not exceed [a certain value]. (In this embodiment, it is 150kg) and the loading rate should be increased as much as possible.

[0070] Yield-matched clustering also considers the operational efficiency of the harvesting vehicles: if the estimated harvesting time for a cluster is calculated by dividing the sum of the maximum yield by the average harvesting efficiency... (Estimated) Exceeds the preset threshold (the preset threshold) (This is based on a comprehensive evaluation of the effective battery life of the picking robot and the maximum turnover cycle of the transport vehicle to and from the unloading point.) In this case, the cluster is forcibly split into smaller sub-clusters to avoid difficulties in scheduling replacement due to excessively long continuous operation time of the picking vehicle.

[0071] This ultimately forms several work clusters, each corresponding to a safe rendezvous point (a flat, easily accessible location), and the minimum number of transport vehicles required for that cluster is recorded. (1-3 times).

[0072] IV. Dynamic Planning of Unloading Points

[0073] The system is based on the set of harvesting clusters that have not yet been completed. Solve the weighted distance optimization problem to determine the optimal unloading point location:

[0074]

[0075] in, For weight, for the first The estimated total yield (kg) of a cluster is the sum of the estimated yield caps of all trees in that cluster; For the first The geometric center of a job cluster or a pre-specified rendezvous point of that cluster. For the hilly operation map, from the candidate unloading point To the cluster center Shortest path distance, set of candidate unloading points Based on terrain flatness and drone take-off and landing requirements, it is pre-generated.

[0076] Weight The significance of using the projected total output is that higher-output work clusters require more round trips for transport vehicles to unload. Increasing their weight forces unloading points to shift towards higher-output clusters, thereby reducing the full-load return distance of transport vehicles from higher-output clusters and lowering overall energy consumption. Then, from all candidate unloading points, the one that best utilizes the projected total output is selected. The point with the smallest value is taken as the optimal unloading point. This optimal unloading point guides the fully loaded return routes of all transport vehicles, and during the operation, whenever a harvested cluster is completely picked, the system removes it from the harvester. Remove the clusters from the optimization and recalculate the optimal unloading point for the remaining clusters, thus excluding the completed clusters from the optimization scope.

[0077] Optimization Example: In the initial stage, all 6 incomplete task clusters (total output approximately 460 kg) participate in the optimization. High-yield clusters (such as cluster B, 130 kg) have a larger weight, and the system selects the candidate point that minimizes the weighted distance as the unloading point. This point is located near the dense area of ​​high-yield clusters, approximately 35 m from the high-yield area and approximately 80 m from the low-yield area, with a weighted average distance of approximately 50 m. This saves approximately 37.5% of the return journey distance compared to the fixed boundary unloading point (80 m from the high-yield area).

[0078] After cluster A (45kg) is harvested, the system removes it from... The remaining 5 clusters were removed and their unloading point locations were re-optimized. As the operation progressed, completed clusters were gradually eliminated, and the unloading points gradually shifted towards the remaining high-yield clusters, continuously approaching the optimal location.

[0079] V. Coordination Sequence Planning

[0080] (1) Planning of harvesting sequence within the cluster

[0081] Starting from the meeting point of the job cluster, a greedy algorithm is used to traverse all fruit trees within the cluster to minimize the cumulative travel cost. Taking cluster C (8 fruit trees, yield approximately 75kg, requiring 2 transport trips) as an example:

[0082] Starting from the meeting point, the system calculates multiple path sequences for traversing all fruit trees from the current point and selects the sequence with the lowest cumulative travel cost (e.g., tree 5 → tree 3 → tree 8 → tree 1 → tree 9 → tree 2 → tree 7 → tree 6). When the cumulative cost of the candidate sequences differs by no more than 5%, the sequence that minimizes the number of times the transport vehicle reverses or makes large-angle turns is prioritized.

[0083] During the harvesting process, the system monitors the remaining capacity of the transport vehicle in real time, as obtained from the weight sensor. When deciding on the next harvest target, priority should be given to selecting from the unharvested trees in the current work cluster, with an estimated yield ceiling not exceeding [a certain value]. The fruit trees; if there are multiple fruit trees that meet the conditions, the principle of minimizing the passage cost shall be followed; if the estimated lower limit of the yield of all unharvested fruit trees is greater than If any tree cannot be fully loaded with the remaining capacity, the current transport vehicle is deemed unable to continue operating, triggering a replacement process: the current transport vehicle proceeds to the unloading point, and an idle transport vehicle takes over to continue harvesting the remaining fruit trees in the current cluster; this avoids harvesting the same fruit tree in stages (i.e., harvesting a portion first, and then harvesting the remaining portion after the transport vehicle returns from unloading), as staged harvesting would cause the harvesting vehicle to repeatedly stop at the same fruit tree, interrupting the work rhythm and causing the transport vehicle to wait ineffectively. For example: after harvesting the first 3 trees (cumulative yield of approximately 48kg), if the remaining capacity (2kg) is less than the estimated minimum yield of the next fruit tree (5kg), the system triggers the first replacement: the current transport vehicle returns fully loaded (approximately 48kg, close to the 50kg upper limit), and an idle transport vehicle takes over to continue harvesting the remaining 5 trees (approximately 27kg).

[0084] (2) Inter-cluster handover sequence planning

[0085] After completing cluster C, the current following transport vehicle is fully loaded (the second trip is loaded with about 27kg, which is insufficient to harvest the minimum yield of any fruit tree). The system determines to trigger the replacement process - the fully loaded transport vehicle returns to the unloading point along the safe path planned by the full load cost map, and at the same time, an idle transport vehicle is dispatched to the rendezvous point of the next target cluster.

[0086] When switching between clusters, the system starts from the rendezvous point of the current cluster and selects the adjacent working cluster with the lowest total travel cost as the next target.

[0087] Figure 3 This invention demonstrates the collaborative dynamic path planning process for harvesting and transporting robots in hilly, disordered orchards.

[0088] The above embodiments are merely preferred embodiments of the present invention. Equivalent substitutions and parameter adaptations made by those skilled in the art for other orchard scenarios such as citrus and apples without departing from the core concept of the present invention all fall within the protection scope of the present invention.

Claims

1. A collaborative dynamic path planning method for picking-transporting robots in hilly, disordered orchards, characterized in that: Obtain orchard semantic map information, which includes at least the location of fruit trees, terrain elevation, estimated yield of each fruit tree, passable area information, and candidate locations for unloading points; The hilly operation map is constructed based on vehicle passability constraints, and each edge is assigned a passability cost related to the load. The passability constraints include at least road width and slope. Edges in the hilly operation map that exceed the maximum safe climbing angle and the maximum safe side tilt angle are marked as impassable. Yield-matched clustering is based on graph distances between fruit trees and the full load capacity of transport vehicles. Unharvested fruit trees are dynamically divided into several work clusters, and the sum of the cumulative estimated yield upper limit of fruit trees in each work cluster is an integer multiple of the full load capacity of the transport vehicle. Dynamic planning of unloading points optimizes the location of unloading points across the entire orchard, minimizing the sum of weighted costs from the meeting point of all currently unharvested work clusters to their optimal unloading point, where the weights are the expected total yield of each work cluster; once a work cluster has completed harvesting, it is automatically excluded from subsequent unloading point optimizations. Collaborative sequence planning aims to minimize total travel distance and total energy consumption. It plans the fruit tree harvesting sequence within a work cluster and the switching sequence between work clusters, and automatically triggers the replacement process when the transport vehicle is fully loaded.

2. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 1, characterized in that, The specific method of output matching clustering is as follows: Calculate the graph distance between all unpicked fruit trees, that is, the shortest path length of the unpicked fruit trees on the hilly operation map; Initial clustering is performed based on the graph distance, ensuring that the graph distance between any two fruit trees within the same cluster is less than the intra-cluster distance threshold. This yields the initial candidate clusters; Calculate the sum of the estimated yield caps of all fruit trees within each candidate cluster. ; Determine the minimum number of transport vehicles required for the cluster. ; like The cluster is retained, by One transport vehicle or the same vehicle The harvesting is completed in a relay within the cluster; among them The maximum number of transport vehicles allowed; like The cluster is divided into several sub-clusters according to its spatial distribution, such that each sub-cluster has... Value not exceeding ; For production less than Clusters whose distance to neighboring graphs is less than the inter-cluster distance threshold are considered. Cluster merging, the output after merging does not exceed .

3. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 2, characterized in that, The yield matching clustering also considers the operating efficiency of the harvesting vehicle: if the estimated harvesting time of a cluster exceeds a preset threshold... If the cluster is split into smaller sub-clusters, the estimated harvesting time is estimated by dividing the sum of the maximum yield by the average harvesting efficiency.

4. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 1, characterized in that, The dynamic programming approach for the unloading point determines the optimal unloading point location by solving the following weighted distance optimization problem: in: For the set of candidate unloading points, This is a collection of harvesting clusters that have not yet been completed. For the first The geometric center of each job cluster As weight, For the hilly operation map, from the candidate unloading point To the cluster center The shortest path distance.

5. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 1, characterized in that, The passage cost is: in, Let be the side length. , These are the weighting coefficients. For the empty load weight of the transport vehicle, It is the acceleration due to gravity. For the current load, This represents the elevation change on the uphill slope.

6. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 5, characterized in that, The fruit tree harvesting order within the job cluster is planned as follows: starting from the meeting point of the job cluster, calculate multiple path sequences for traversing all fruit trees from the current point, and select the sequence with the minimum cumulative travel cost.

7. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 6, characterized in that, Real-time monitoring of the remaining capacity of transport vehicles When deciding on the next harvest target, priority should be given to selecting from the unharvested trees in the current work cluster, with an estimated yield ceiling not exceeding [a certain value]. The fruit trees; if there are multiple fruit trees that meet the conditions, the one with the lowest passage cost shall be selected. If the lower limit of the estimated yield of all unharvested fruit trees is greater than If the current transport vehicle is determined to be unable to continue operating, a replacement process is triggered: the current transport vehicle proceeds to the unloading point, and an idle transport vehicle takes over to continue harvesting the remaining fruit trees in the current cluster.

8. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 6, characterized in that, The switching order between the work clusters is planned as follows: when a work cluster is completed, if the current following transport vehicle is not fully loaded, the adjacent work cluster with the shortest graph distance and the lowest total passage cost is selected as the next target. If the system is fully loaded, the takeover process will be triggered immediately.

9. The collaborative dynamic path planning method for harvesting-transporting robots according to claim 6, characterized in that, The maximum safe climbing angle and the maximum safe side tilt angle are determined according to the ground conditions: for soft soil or slippery road surfaces, they are set to 15° and 10° respectively; for hardened road surfaces or dry hard soil, they are set to 20° and 15° respectively.

10. A system for implementing the cooperative dynamic path planning method for harvesting-transporting robots according to any one of claims 1-9, characterized in that, include: The orchard semantic map module stores terrain elevation and passable areas, fruit tree locations and estimated yields, and candidate unloading points; The online planning module includes a hilly operation map construction module, a slope hazard edge elimination module, a yield matching clustering module, a loading point dynamic planning module, a collaborative sequence planning module, and a transport vehicle dynamic docking module. The hilly operation map construction module is used to construct a graph structure with passage costs based on vehicle passability constraints. The slope hazard edge elimination module is used to directly disable dangerous road sections that exceed the limits based on safety angle thresholds. The yield matching clustering module is used to divide fruit trees into operation clusters with appropriate total yield according to graph distance and load limit. The loading point dynamic planning module is used to dynamically optimize the location of loading points according to the operation progress using a weighted shortest path model. The collaborative sequence planning module is used to plan the optimal traversal order of picking within a cluster and the switching order between clusters. The transport vehicle dynamic docking module is used to intelligently make decisions and dispatch idle vehicles to replace fully loaded vehicles based on real-time load status. The execution and feedback module is used to execute the job instructions issued by the system and upload the status in real time.

Images