Multi-level collaborative planning robot navigation control method and system

Abstract

The application relates to the technical field of robot navigation control, and discloses a robot navigation control method and system based on multi-level cooperative planning, which comprises the following steps: firstly, acquiring environment sensing data of various nodes such as global planning, local obstacle avoidance and execution control, extracting navigation features and matching to obtain a level identification pair, and then constructing a navigation model according to the level identification pair to generate standard control instructions through layered instruction encapsulation; meanwhile, acquiring environment topology data, constructing a three-dimensional environment map, dividing regional units, classifying the regional units into types such as core navigation areas, iteratively optimizing an initial navigation path by using an adaptability evaluation system combined with an ant colony algorithm, and screening the level identification pair through bidirectional matching. A controller of the system is used for executing the above operation. The method and system realize multi-level cooperative planning, and improve the efficiency, accuracy and reliability of robot navigation in a complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot navigation control, in particular to a multi-level cooperative planning robot navigation control method and system. BACKGROUND

[0002] In the development process of robot technology, navigation control, as one of its core technologies, faces challenges in adaptability to complex environments, path planning efficiency, and control instruction accuracy. With the continuous expansion of robot application scenarios, from industrial production to service fields, from structured environments to unstructured complex scenarios, traditional navigation control methods gradually reveal many shortcomings.

[0003] Traditional robot navigation control often uses a single level of planning, which is difficult to effectively handle environment information of different scales and complexities. For example, at the global planning level, traditional methods may not be able to accurately adjust in combination with real-time changes in the local environment, resulting in a lack of flexibility in the actual execution process when encountering obstacles, requiring frequent re-planning, which seriously affects navigation efficiency. At the local obstacle avoidance level, if only relying on local sensor data, it may lack overall awareness of the global environment, fall into a local optimal trap, and be unable to find a globally optimal obstacle avoidance path, or even cause the robot to lose its way in a complex environment.

[0004] At the same time, in terms of processing of environment perception data, traditional methods have limited ability to fuse multi-source heterogeneous data. Different types of sensors, such as lidar, vision cameras, ultrasonic sensors, etc., collect data with different characteristics and precision, and traditional methods are difficult to efficiently extract navigation features from these data and accurately match and fuse them, thereby affecting the construction accuracy of the navigation model. In addition, the traditional instruction packaging method is relatively fixed and cannot be flexibly adjusted according to different environmental scenarios and task requirements, resulting in control instructions generated in the actual execution process that may not match the hardware characteristics and motion capabilities of the robot, affecting the motion stability and control accuracy of the robot.

[0005] In terms of path planning, traditional methods lack path optimization capability in complex environments. When faced with complex obstacle distribution and multiple target areas, traditional path planning algorithms often require a long calculation time and are difficult to achieve a good balance between path smoothness, obstacle avoidance difficulty, and energy cost. For example, some algorithms may focus too much on obstacle avoidance, resulting in a long path and increased energy consumption; or in order to pursue the shortest path, the smoothness of the path is ignored, causing large vibrations during robot movement, affecting equipment life and task execution effect.

[0006] In addition, the traditional robot navigation control system lacks an effective hierarchical coordination mechanism when responding to environmental dynamic changes. The information exchange and coordination between different nodes such as global planning, local obstacle avoidance, and execution control are inefficient, and the navigation strategy cannot be adjusted in time according to environmental changes, resulting in poor adaptability of the robot in dynamic environments. For example, when a new obstacle suddenly appears in the environment, the traditional system may not be able to quickly pass this information to the global planning layer, causing the global path planning to be disconnected from the actual environment and affecting the navigation safety of the robot.

[0007] The traditional robot navigation control adopts a single-level planning method, which is difficult to effectively process environmental information of different scales and complexities. The global planning layer cannot adjust the path in real time according to local environmental changes, resulting in a lack of flexibility when encountering obstacles in actual execution and the need for frequent re-planning, which seriously affects navigation efficiency. The local obstacle avoidance layer only relies on local sensor data and is prone to fall into a local optimal trap due to lack of global environmental awareness, making it difficult to find a globally optimal obstacle avoidance path, and even losing direction in complex environments. At the same time, the traditional method has limited multi-source heterogeneous data fusion capability. The data characteristics and precision collected by different types of sensors are different, making it difficult to efficiently extract navigation features and accurately match and fuse, affecting the accuracy of the navigation model. In addition, the traditional instruction packaging method is fixed and cannot be flexibly adjusted according to environmental scenarios and task requirements, resulting in control instructions that may not match the robot's hardware characteristics and motion capabilities, affecting motion stability and control accuracy. In terms of path planning, the traditional method lacks optimization capability in complex environments. In the face of complex obstacle distribution and multiple target areas, the calculation time is long, and it is difficult to achieve a good balance among path smoothness, obstacle avoidance difficulty, and energy cost, etc. It may be too focused on obstacle avoidance, resulting in a longer path and increased energy consumption, or it may ignore smoothness in pursuit of the shortest path, causing the robot to move with large vibrations, affecting equipment life and task execution effectiveness. In addition, the traditional system lacks an effective hierarchical coordination mechanism when responding to environmental dynamic changes. The information exchange and coordination between different nodes such as global planning, local obstacle avoidance, and execution control are inefficient, and the navigation strategy cannot be adjusted in time according to environmental changes, resulting in disconnection between global path planning and the actual environment, affecting navigation safety. SUMMARY

[0008] The present application aims to provide a multi-level collaborative planning robot navigation control method and system to solve the problems raised in the background art.

[0009] To achieve the above-mentioned purpose, the present application provides the following technical solution: a multi-level collaborative planning robot navigation control method, the method comprising:

[0010] Obtain environment perception data collected by multiple nodes, extract navigation features according to the environment perception data, and obtain a pair of hierarchical identifiers by matching the extracted navigation features; wherein the multiple nodes include a global planning node, a local obstacle avoidance node, and an execution control node; wherein obtaining a pair of hierarchical identifiers by matching the extracted navigation features includes: extracting a hierarchical descriptor at the navigation feature, assigning a time sequence label to the navigation feature, finding an associated point according to the hierarchical descriptor, and thereby obtaining a matching pair of hierarchical identifiers; wherein the hierarchical descriptor is a sequence of attributes describing the navigation feature;

[0011] Construct a navigation model according to the environment perception data and the pair of hierarchical identifiers, and generate standard control instructions by encapsulating the navigation model according to the environment perception data, including:

[0012] Obtain the pair of hierarchical identifiers as a reference point by continuous association to obtain a navigation model, use a region segmentation-based method, and obtain a complete or partial navigation path model covering the target region by expanding and screening based on the sparse path obtained by matching the navigation features between multiple nodes;

[0013] Use hierarchical instruction encapsulation, encapsulate instructions through two levels, the first level encapsulates from the original perception space to an intermediate mode that can be expressed by available strategies; the second level encapsulates the instructions of the intermediate mode to the target space that needs to be executed, and generates standard control instructions.

[0014] Preferably, the method further includes the following contents:

[0015] Obtain environment topology data required for navigation, generate a navigation path according to the environment topology data; and multiple nodes collect environment perception data according to the navigation path;

[0016] The environment topology data includes target region, obstacle distribution, and node state data of the region required for navigation, and the navigation path is generated according to the environment topology data, including the following contents:

[0017] Construct a navigation region according to boundary recognition of the target region, and plan a navigation path according to the navigation region;

[0018] And, establish a navigation reference according to the obstacle distribution and the node state data, and collect environment perception data according to the navigation reference.

[0019] Preferably, planning a navigation path according to the navigation region includes:

[0020] Construct a three-dimensional environment map in the target region to be navigated, and divide the target region to be navigated into a plurality of region units with a preset interval as the side length according to the three-dimensional environment map;

[0021] According to the target region, obstacle distribution and node state data, the three-dimensional region unit is classified, including three types of core navigation area, general path area and obstacle avoidance area, and the core navigation area is marked with priority;

[0022] A plurality of initial navigation paths are randomly generated to form an initial navigation path set;

[0023] An adaptability evaluation system is constructed, including a path smoothness evaluation item, an obstacle avoidance difficulty evaluation item and an energy consumption cost evaluation item;

[0024] The path smoothness evaluation item is the cumulative value of the turning angle between each node in the initial navigation path;

[0025] The obstacle avoidance difficulty evaluation item is the sum of the obstacle density in each region unit in the initial navigation path;

[0026] The energy consumption cost evaluation item is the number of node power and turning parameters that need to be adjusted in the initial navigation path;

[0027] The initial navigation path set is taken as the initial population, and iterative optimization is performed according to the adaptability evaluation system to obtain the optimal navigation path, thereby completing the planning of the navigation path.

[0028] Preferably, taking the initial navigation path set as the initial population and performing iterative optimization according to the adaptability evaluation system includes:

[0029] According to the adaptability evaluation system, the evaluation value of each initial navigation path is calculated, and the initial navigation paths are sorted from low to high according to the three evaluation items to obtain a sorting set of the three initial navigation paths;

[0030] According to the evaluation value, a plurality of initial navigation paths are selected from the three sorting sets by probability, forming three sub-populations in different directions, and the selected probability is inversely proportional to the evaluation value;

[0031] Each sub-population is updated in position, pheromone is adjusted and individual learning is performed according to the ant colony optimization algorithm, and the inadmissible path scheme is eliminated, and each is iterated for a preset number of times to form three sub-populations after iteration;

[0032] According to the evaluation value, individuals are randomly selected from the three sub-populations to form a basic pheromone library, and three groups of cross sub-populations are formed by crossing between the three sub-populations, for each group of cross sub-populations, the three evaluation values of each individual are calculated, the two evaluation values involved in the cross sub-population are sorted, and the position is updated, the pheromone is adjusted and the individual learning is performed according to the sorting result, and then the inadmissible path scheme is eliminated;

[0033] After the three cross sub-populations iterate for a preset number of times, the three sub-populations are combined together, a comprehensive evaluation value is obtained by weighted summation of the three evaluation values, position updating, pheromone adjustment and individual learning are performed according to the comprehensive evaluation value, and after iterating for a preset number of times, an individual with the minimum evaluation value is obtained as the optimal navigation path.

[0034] Preferably, the two evaluation values involved in the cross sub-population are subjected to multi-objective sorting, and in the position updating, pheromone adjustment and individual learning according to the sorting result, before the cross of the cross sub-population, a dynamic adjustment mechanism is used for selection of the cross objects, wherein the selection probability of the cross objects is related to the position of the individual in the sorting, and the earlier the position of the individual in the sorting, the higher the selection probability.

[0035] Preferably, the method further comprises: assigning a selection probability to the individual to be crossed according to the sorting order of the individual, and the earlier the position of the individual in the sorting, the greater the selection probability value.

[0036] Preferably, the method further comprises the following contents:

[0037] The matched level identifier pairs are screened, the mis-matched level identifier pairs are screened out, and when the navigation model is constructed, the construction is performed according to the screened level identifier pairs.

[0038] Preferably, the screening of the matched level identifier pairs is: the bidirectional matching and the uniqueness verification are used to remove the mis-matched level identifier pairs.

[0039] Preferably, the application further comprises a multi-level collaborative planning robot navigation control system using the above-mentioned multi-level collaborative planning robot navigation control method, and the system comprises a controller:

[0040] The controller is used for acquiring environment perception data collected by a plurality of nodes, performing navigation feature extraction according to the environment perception data, and obtaining level identifier pairs according to the extracted navigation features.

[0041] The controller is further used for constructing a navigation model according to the environment perception data and the level identifier pairs, and generating standard control instructions by encapsulating the navigation model according to the environment perception data.

[0042] Preferably, the controller is further used for generating a navigation path according to the environment topology data, and the plurality of nodes collect and acquire the environment perception data according to the navigation path.

[0043] Compared with the prior art, the application has the following beneficial effects:

[0044] The multi-level collaborative planning robot navigation control method and system significantly improves the navigation control ability of the robot in a complex environment through the collaborative work of multiple nodes and the multi-level planning mechanism.In terms of environment perception and feature processing, the environment perception data collected by multiple nodes are obtained and navigation feature extraction and matching are performed, which can fully utilize the advantages of multi-source data and improve the understanding and expression ability of environment information.The hierarchical descriptor is extracted, the time sequence label is assigned, and the associated point is found to obtain the hierarchical identification pair, which provides an accurate reference point for subsequent navigation model construction, so that the navigation model can more accurately reflect the environmental features and the motion requirements of the robot.

[0045] In terms of navigation model construction and instruction packaging, the navigation model is obtained through continuous association based on the hierarchical identification pair as the reference point, and the sparse path is expanded and screened based on the regional segmentation method to obtain a complete or partial navigation path model covering the target area, effectively improving the integrity and adaptability of the path planning.The layered instruction packaging mechanism gradually packages the information of the original perception space into the target execution space, ensuring the accuracy and executability of the control instructions, so that the robot can more accurately execute the navigation task according to the instructions.

[0046] In the path planning process, by constructing a three-dimensional environment map and classifying the regional units, and combining the fitness evaluation system to iteratively optimize the initial navigation path, a good balance between path smoothness, obstacle avoidance difficulty and energy cost can be achieved.Ant colony optimization algorithm combined with multi-objective sorting and dynamic adjustment mechanism is adopted to improve the efficiency and accuracy of path planning, so that the robot can quickly find the optimal navigation path in a complex environment.

[0047] In addition, the matched hierarchical identification pairs are screened, and the bidirectional matching and uniqueness verification are used to eliminate false matches, ensuring the accuracy of the navigation model construction and further improving the reliability of the robot navigation.The controller can efficiently process environment perception data, construct navigation models and generate standard control instructions, realize the collaborative work between multiple hierarchical nodes, and enable the robot to adjust the navigation strategy in time when facing dynamic changes in the environment, thereby improving the environmental adaptability and navigation safety.

[0048] The method and system effectively integrate the functions of different nodes such as global planning, local obstacle avoidance and execution control through multi-level collaborative planning, realize the whole process optimization from environment perception, path planning to control instruction generation, significantly improve the efficiency, accuracy and reliability of robot navigation, and better meet the diversified navigation task requirements in complex environments. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 is a working principle diagram of the multi-level collaborative planning robot navigation control method.

[0050] Figure 2 designing a navigation path for the environmental topology data;

[0051] Figure 3 designing a navigation path for the navigation area;

[0052] Figure 4 designing a fitness evaluation and iterative optimization. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0054] Please refer to Figures 1-4 The robot navigation control method of multi-level cooperative planning involves the following specific implementation steps:

[0055] Obtain environmental perception data collected by multiple nodes, including a global planning node, a local obstacle avoidance node and an execution control node. According to the environmental perception data, navigation features are extracted, hierarchical descriptors are extracted at the navigation features, the hierarchical descriptors are attribute sequences describing the navigation features, time sequence labels are assigned to the navigation features, and associated points are found according to the hierarchical descriptors, so as to obtain a matching hierarchical identifier pair.

[0056] According to the environmental perception data and the hierarchical identifier pair, a navigation model is constructed, and a standard control instruction is generated by encapsulating the navigation model according to the environmental perception data. The hierarchical identifier pair obtained by matching is used as a reference point to obtain the navigation model by continuous association. A region segmentation-based method is used, and a sparse path obtained by matching navigation features between multiple nodes is used to obtain a complete or partial navigation path model covering the target region by expansion and screening. Hierarchical instruction encapsulation is used, and two levels are used for instruction encapsulation. The first level encapsulates from the original perception space to an intermediate mode expressed by available strategies; and the second level encapsulates the instructions of the intermediate mode to the target space to be executed, to generate a standard control instruction.

[0057] Embodiment 1:

[0058] After obtaining the required navigation environment topology data, a navigation path is generated according to the environment topology data, and then a plurality of nodes collect environment perception data according to the navigation path. The environment topology data here covers the target area of the required navigation area, obstacle distribution and node state data. On the one hand, the target area is used for boundary identification to construct a navigation area, and then a navigation path is planned according to the navigation area; on the other hand, a navigation reference is established according to the obstacle distribution and node state data, so that the environment perception data is collected based on the navigation reference.

[0059] Constructing a three-dimensional environment map in the target area to be navigated is a basic step for planning a navigation path. After constructing a three-dimensional environment map, the target area to be navigated needs to be divided into a plurality of area units with a preset interval as the side length. The setting of this preset interval needs to consider factors such as navigation accuracy and calculation efficiency, for example, a suitable value can be determined according to the size of the robot, the detection range of the sensor and the requirements of the actual navigation task, to ensure that the divided area units can meet the needs of navigation details and will not make the calculation too large.

[0060] When setting the preset interval, the balance between navigation accuracy and calculation efficiency needs to be fully considered. First, combined with the size characteristics of the robot itself, if the robot is large, too small interval will result in too fine area unit, which may not be able to adapt to the actual movement space requirements of the robot, while too large interval may ignore the key environmental details around the robot; at the same time, the detection range of the sensor needs to be referred to, so that the interval matches the effective sensing area of the sensor, ensuring that the environmental information in each area unit can be fully captured by the sensor, avoiding missing obstacles or path features due to the interval exceeding the detection range. In addition, it also needs to be adjusted according to the specific requirements of the actual navigation task. For tasks that require high-precision navigation, such as fine movement in complex and narrow spaces, the interval can be appropriately reduced to retain more environmental details and ensure the accuracy of navigation path planning; while for large-scale navigation tasks that require higher efficiency, the interval can be appropriately increased to reduce the total number of area units, thereby reducing the data processing amount in the subsequent path calculation process, avoiding excessive consumption of computing resources, and ultimately making the divided area units meet the needs of environmental detail perception in the navigation process, ensuring the accuracy of path planning, and avoiding unnecessary calculation burden due to too many units, ensuring the running efficiency of the entire navigation system.

[0061] According to the target region, obstacle distribution and node state data, the three-dimensional region units are classified. The classification results include three types of core navigation area, general path area and obstacle avoidance area, and the core navigation area is marked with priority. The core navigation area is usually directly related to the target region and is essential for the navigation task, such as the main channel to the target point; the general path area is the area connecting the core navigation area or serving as an auxiliary path; the obstacle avoidance area is the area with obstacles that needs to be avoided by the robot. Marking the core navigation area with priority helps to prioritize this area when planning the path later, ensuring that the robot can approach the target more efficiently.

[0062] After completing the classification of region units, a number of initial navigation paths are randomly generated to form an initial navigation path set. The generation of initial navigation paths can be based on certain algorithms or rules, which are based on the region units divided in the three-dimensional environment map, combined with the target region, obstacle distribution and node state data. According to the classification results of region units, i.e. core navigation area, general path area and obstacle avoidance area, the algorithm will preferentially select path points in the core navigation area and general path area, and avoid the obstacle avoidance area to ensure the basic feasibility of the initial path. According to the rules, a series of path points will be randomly selected in the feasible region units from the navigation starting point in the general direction towards the target region. The distribution of these path points will refer to the node state data to ensure that the path points are within the node sensing range and meet the requirements of the node working state. According to the order of path points, straight lines or curves that meet the kinematic constraints of the robot are used to connect these path points in sequence to form an initial navigation path from the starting point to the target region. During the connection process, it is simply checked whether the path segment passes through the obstacle avoidance area. If there is a path segment passing through the obstacle avoidance area, the corresponding path point is fine-tuned or reselected to generate a number of initial navigation paths with different forms, and finally form an initial navigation path set, providing a variety of basic path samples for subsequent iterative optimization based on the fitness evaluation system.

[0063] A fitness evaluation system needs to be constructed, which includes a path smoothness evaluation item, an obstacle avoidance difficulty evaluation item, and an energy cost evaluation item. The path smoothness evaluation item is the cumulative value of the turning angles between nodes in the initial navigation path. The smaller the cumulative value of the turning angles, the smoother the path, and the robot does not need to change direction frequently during travel, which is beneficial to improve the stability and efficiency of travel. The obstacle avoidance difficulty evaluation item is the sum of the obstacle density in each region unit in the initial navigation path. The higher the obstacle density, the greater the difficulty of obstacle avoidance for the robot in that area, and more operations and energy may be needed to avoid obstacles, so the value of the obstacle avoidance difficulty evaluation item is also larger. The energy cost evaluation item is the number of nodes and turning parameters that need to be adjusted in the initial navigation path. The more parameters that need to be adjusted, the more energy the robot needs to consume to change the power and direction during travel, so the value of the energy cost evaluation item is also larger.

[0064] The initial navigation path set is used as the initial population, and iterative optimization is performed according to the fitness evaluation system to obtain the optimal navigation path, completing the planning of the navigation path. During the iterative optimization process, the initial navigation path needs to be continuously evaluated, selected, updated, and screened to gradually improve the quality of the path. For example, in each iteration, the evaluation value of each initial navigation path is calculated according to the fitness evaluation system, and then the paths are selected according to the evaluation value, with the paths with better evaluation values being retained and the paths with worse evaluation values being eliminated. At the same time, the retained paths are updated and optimized to generate better paths. Through multiple iterations, the path with the smallest evaluation value obtained finally is the optimal navigation path.

[0065] During the planning of the navigation path, various factors need to be considered. For example, the shape and size of the target area will affect the construction of the navigation area and the planning of the path; the distribution and type of obstacles will affect the value of the obstacle avoidance difficulty evaluation item and the selection of the path; the robot performance parameters in the node state data, such as maximum speed and minimum turning radius, will affect the values of the path smoothness evaluation item and the energy cost evaluation item. Therefore, when planning the navigation path, these factors need to be considered comprehensively to ensure that the planned path can meet the requirements of the navigation task and the robot can maintain stability, efficiency, and energy saving during travel.

[0066] When generating the initial navigation path set, it is necessary to ensure that the number of initial paths is sufficient to cover various possible path situations, thereby improving the effect of iterative optimization. At the same time, during the iterative optimization process, the preset number of iterations also needs to be reasonably set according to the actual situation. If the number of iterations is too small, the optimal path may not be obtained; if the number of iterations is too large, the computational load and time cost will increase. Therefore, the appropriate number of iterations needs to be determined through experiments or experience to balance the computational efficiency and path quality.

[0067] In determining the appropriate number of iterations, adjustments can be made in combination with specific navigation scenarios and experimental results. For example, in a warehouse environment containing multiple shelves and aisles, the initial navigation path set contains 50 paths. First, try 20 iterations, at which time the optimized path has certain improvements in smoothness and obstacle avoidance difficulty, but some paths still have high energy consumption problems. Then increase to 40 iterations, and find that the comprehensive evaluation value of the path is significantly reduced, the path smoothness is improved, and the energy consumption during obstacle avoidance is also reduced. Continue to iterate to 60 times, and the comprehensive evaluation value of the path decreases slightly, and the calculation time increases by nearly 30% compared to 40 times. Further iteration to 80 times, the evaluation value hardly changes, but the calculation time significantly increases. Through multiple experiments and comparisons, it is found that 40 iterations can achieve a good balance between path smoothness, obstacle avoidance difficulty and energy consumption cost, and will not affect the real-time navigation due to excessive calculation time. Therefore, the number of iterations in this scenario is determined to be 40.

[0068] When classifying three-dimensional region units, it is necessary to ensure the accuracy and reasonableness of the classification. For example, for obstacle distribution, it is necessary to accurately obtain information such as the position, shape and size of the obstacle, and determine the range of obstacle avoidance area according to these information. For node state data, it is necessary to understand the motion ability of the robot and the detection range of the sensor, etc., to determine the division standard of the core navigation area and the ordinary path area.

[0069] When determining the division standard of the core navigation area and the ordinary path area, the motion ability of the robot and the sensor detection range need to be determined first. For example, the minimum turning radius of the robot is 0.8 meters, the maximum travel speed is 1.5 meters / second, and the effective detection range of the laser radar is 8 meters in front, 60 degrees on the left and right. The division of the core navigation area needs to ensure that the channel width in the area is not less than 1.6 meters, which meets the robot's turning action at the maximum speed, and the area needs to be within the effective detection range of the laser radar throughout the journey to ensure that the robot can obtain real-time information about the obstacles in front and on both sides, ensuring navigation safety. At the same time, the core navigation area needs to directly connect the key nodes of the starting point and the target area, which is the main path to complete the navigation task. The ordinary path area, as an auxiliary path connecting the core navigation area, can be set to 1.2 meters in width, slightly smaller than the core navigation area, but still larger than the width of the robot itself (0.6 meters), ensuring that the robot can pass smoothly, and the area needs to be within the partial detection range of the laser radar (such as 5 meters in front, 30 degrees on the left and right), which can meet the basic obstacle sensing needs, and provide alternative paths when the core navigation area is temporarily blocked. The obstacle avoidance area is the area where the laser radar detects more than 5 obstacles per square meter or the channel width is less than 0.8 meters (which cannot meet the robot's turning needs), which is explicitly excluded from the core navigation area and the ordinary path area.

[0070] The navigation path is generated by acquiring environmental topology data, which enables various nodes to collect environmental perception data in a targeted manner, improving the effectiveness and accuracy of data collection. A three-dimensional environmental map is constructed and regional units are divided. The regional units are classified and labeled with core navigation area priorities based on target area, obstacle distribution, and other data. This makes the path planning more in line with the actual environment and highlights the importance of key areas. A set of initial navigation paths is randomly generated and a fitness evaluation system is constructed, including path smoothness, obstacle avoidance difficulty, and energy cost. The optimal path is obtained through iterative optimization, which balances the smoothness of the path, the difficulty of obstacle avoidance, and the energy cost while meeting the navigation requirements. This allows the robot to travel more efficiently and stably in complex environments, reducing unnecessary energy consumption and obstacle avoidance operations.

[0071] Embodiment 2:

[0072] The iterative optimization of the initial navigation path set as the initial population is a key step in generating the optimal navigation path. The specific implementation is as follows: the evaluation value of each initial navigation path is calculated according to the fitness evaluation system, which includes three evaluation items: path smoothness, obstacle avoidance difficulty, and energy cost. For each evaluation item, the initial navigation paths are sorted from low to high according to the evaluation value, resulting in three corresponding initial navigation path sorting sets. For example, the path smoothness evaluation item is based on the cumulative value of the turning angle between nodes. The smaller the cumulative value of the turning angle, the higher the position of the path in the sorting set. The obstacle avoidance difficulty evaluation item is based on the total obstacle density in each regional unit. The smaller the total obstacle density, the higher the position of the path in the sorting set. The energy cost evaluation item is based on the number of nodes that need to be adjusted and the number of turning parameters. The fewer the number of parameter adjustments, the higher the position of the path in the sorting set.

[0073] According to the evaluation values, a number of initial navigation paths are selected from the three ranking sets by probability to form three sub-populations in different directions. The selection probability here is inversely proportional to the evaluation value, that is, the lower the evaluation value, the higher the probability of being selected. For example, in the path smoothness ranking set, the probability of being selected by the paths ranked at the front (low evaluation value) is higher than that of the paths ranked at the back. This selection mechanism aims to preferentially retain paths with better evaluation performance to provide a better foundation for subsequent optimization. After each sub-population is formed, position updating, pheromone adjustment, and individual learning are performed within each sub-population according to the basic principles of the ant colony optimization algorithm. Among them, position updating refers to adjusting the positions of the nodes in the path to find a better path form; pheromone adjustment simulates the behavior of ants leaving pheromones when foraging, and guides subsequent path search by increasing the pheromone concentration on the better path; individual learning refers to the path individuals in the sub-population learning from the structural features of other excellent individuals to optimize their own paths. In this process, infeasible path schemes need to be removed, such as those passing through obstacle areas or exceeding the robot's movement capabilities. Each sub-population is iterated for a predetermined number of times, which needs to be reasonably set according to factors such as the complexity of the navigation task and computing resources, and after iteration, three sub-populations after iteration are formed.

[0074] When setting the predetermined number of iterations for the sub-populations, it can be adjusted according to the complexity of the navigation task and computing resources. For example, in an indoor environment with fewer obstacles and simple region unit division, the navigation task complexity is low, and the robot carries a general-purpose processor with computing power, so the iteration number of each sub-population can be set to 15 times. After 15 iterations, the paths in the sub-population have been well optimized in smoothness and obstacle avoidance difficulty, and further increasing the iteration number does not significantly improve the path quality, but also occupies excessive computing resources and affects real-time performance. In a dense obstacle outdoor park environment, the task complexity is high, and if the robot is equipped with a high-performance processor, the iteration number of each sub-population can be set to 30 times, allowing the paths in the sub-population to be fully optimized through more position updates, pheromone adjustments, and individual learning to adapt to the needs of complex environments, and finally forming three sub-populations after iteration.

[0075] According to the evaluation values, individuals are randomly selected from the three sub-populations to form a basic pheromone library. The basic pheromone library serves as a shared resource of pheromones, providing guidance for subsequent crossover operations. Crossover is performed between each pair of sub-populations, forming three groups of crossover sub-populations. For example, sub-population A and sub-population B are crossed, sub-population B and sub-population C are crossed, and sub-population C and sub-population A are crossed. Each group of crossover sub-populations contains individuals from two original sub-populations. For each group of crossover sub-populations, first calculate the three evaluation values of each individual, then perform multi-objective sorting for the two evaluation values involved in the crossover sub-population. For example, a group of crossover sub-populations is formed by the crossover of sub-population A and sub-population B, which may focus on the path smoothness and obstacle avoidance difficulty evaluation items, and the comprehensive performance of the two evaluation items is sorted. After sorting, position updating, pheromone adjustment and individual learning are performed according to the sorting results, and infeasible path schemes are also removed. The three groups of crossover sub-populations are iterated for a predetermined number of times, promoting gene exchange between different sub-populations through crossover operations, making path characteristics more diverse, and improving the possibility of searching for the optimal path.

[0076] After the iteration of the three groups of crossover sub-populations, the three sub-populations are combined together, and the comprehensive evaluation value is obtained by weighted sum of the three evaluation items. When weighted sum is performed, the weight of each evaluation item can be determined according to the actual navigation requirements, for example, in a scenario with high requirements for path smoothness, the weight of the path smoothness evaluation item can be set relatively large. According to the comprehensive evaluation value, position updating, pheromone adjustment and individual learning are performed, and the iteration is performed for a predetermined number of times. In this stage, the performance of the three evaluation items is comprehensively considered to optimize the path, and finally the individual with the minimum evaluation value is obtained, which is the optimal navigation path.

[0077] During the entire iterative optimization process, attention should be paid to the diversity of the initial navigation path set. If the initial paths are too similar, the iterative optimization may fall into local optimum and fail to find the global optimal path. Therefore, when generating the initial navigation path, a reasonable random generation method should be used to ensure that the path set covers different path forms. At the same time, the preset number of iterations needs to balance the calculation efficiency and path optimization effect. If the number of iterations is too small, the path optimization may not be sufficient; if the number of iterations is too large, it will increase the calculation time and resource consumption. In addition, the strategy of pheromone adjustment in the operation of sub-populations and crossover sub-populations is also crucial, and a reasonable pheromone update rule can guide the search direction and accelerate the convergence speed.

[0078] The reasonable random generation mode can operate as follows: in the area unit of the three-dimensional environment map, in combination with the distribution of the core navigation area and the ordinary path area, 2-3 key path points are randomly selected in the core navigation area from the starting point, and these points are distributed as much as possible in the main direction to the target area; then a plurality of auxiliary path points are randomly selected in the ordinary path area, the positions of the auxiliary path points avoid the obstacle avoidance area, and the auxiliary path points form different connection combinations with the key path points. For example, some initial paths prefer to pass through the left area of the core navigation area, some prefer to pass through the right area, and some select a shorter connection route in the ordinary path area. When connecting the path points, some are connected by straight line segments, and some are connected by small-angle turning according to the shape of the area unit, so as to ensure that the generated initial navigation path presents diversity in path direction and passed area unit combination, and avoids all paths being too similar, thereby providing rich basic samples for subsequent optimization.

[0079] For the elimination of the unreachable path scheme, the kinematics and dynamics constraints of the robot and the obstacle distribution in the environment need to be judged. For example, the minimum turning radius and the maximum climbing angle of the robot limit the feasibility of the path. If there is a turning or climbing operation in the path that exceeds these parameters, the path is an unreachable path and needs to be eliminated. The obstacle distribution information is used to judge whether the path passes through the obstacle area. If the path node is located inside the obstacle or too close to the edge of the obstacle, the robot cannot pass through safely, and the path should also be regarded as an unreachable path.

[0080] When forming the basic pheromone library, the random selection of individuals from the three sub-populations needs to be combined with a certain tendency selection according to the evaluation value, so as to ensure that the basic pheromone library contains sufficient high-quality path information. For example, individuals with a certain proportion of evaluation values in the front can have a higher probability of being selected into the basic pheromone library, so that the basic pheromone library can better guide the subsequent path search.

[0081] According to the fitness evaluation system, the initial navigation path is sorted and selected according to the probability to form a sub-population, which can preferentially retain high-quality path characteristics and lay a good foundation for optimization. The ant colony optimization algorithm is used in the sub-population for iteration, combined with position updating, pheromone adjustment and individual learning, which can continuously optimize the path, eliminate the unreachable scheme and improve the feasibility of the path. The cross between the sub-populations forms a cross sub-population and performs multi-objective sorting optimization, which can promote the fusion of different path characteristics, increase the path diversity and avoid falling into local optimum. Finally, the optimal path is obtained through comprehensive evaluation, which takes into account the needs of each evaluation item and improves the efficiency and accuracy of path planning, so that the robot can quickly find a better navigation path.

[0082] Example 3:

[0083] When the two evaluation values involved in the crossover sub-population are multi-objective sorted and subsequent operations are performed, the crossover object selection mechanism before the crossover of the crossover sub-population is the key link to improve the efficiency of path optimization. The specific implementation is as follows: before the crossover of the crossover sub-population, a dynamic adjustment mechanism is used to select the crossover object. The core of this selection mechanism is to associate the selection probability of the crossover object with the position of the individual in the sorting, where the earlier the individual is in the sorting, the higher the probability of being selected.

[0084] After the two evaluation values involved in the crossover sub-population are multi-objective sorted, each individual has a corresponding position order in the sorting result. For example, assuming that a group of crossover sub-populations focuses on the path smoothness and obstacle avoidance difficulty evaluation items, after multi-objective sorting, individuals are arranged from best to worst according to the comprehensive performance of the two evaluation items, forming an ordered individual list. At this time, the selection probability needs to be assigned according to the sorting order of the individual to be crossed in the list. The earlier the individual is in the sorting, the higher the selection probability value. This assignment method can be implemented in a linear or nonlinear manner, for example, setting the selection probability of the first individual in the sorting as p1, the second as p2, and p1>p2>…>pn (n is the number of individuals in the crossover sub-population). The specific probability value needs to be set according to actual optimization requirements and experience to ensure that individuals with higher sorting have significantly higher selection probability.

[0085] When setting the selection probability of the sorted individual, it can be adjusted according to the actual optimization requirements. For example, in a navigation task that places more emphasis on path smoothness, for the path smoothness sorting set, the probability of the first individual being selected can be set to a relatively high value, the probability of the second individual being selected is slightly lower than the first, the probability of the third individual being selected is lower, and so on. For example, in a sorting set containing 10 individuals, the probability of the first individual being selected is significantly higher than that of the eighth and ninth individuals, ensuring that in the selection of sub-population individuals, high-quality individuals with small turning angle cumulative values and smoother paths can be retained with a higher probability, thereby allowing the sub-population to converge more quickly to a better path in iterative optimization.

[0086] The design principle of this dynamic adjustment mechanism is that individuals with higher sorting perform better in the evaluation items currently focused on and are more likely to carry path features that are beneficial to optimization. By increasing the probability of being selected, the transmission probability of high-quality path features in the crossover operation can be increased, thereby accelerating the convergence of the iterative optimization process to a better solution. For example, in the crossover operation, two high-quality individuals are crossed, which may produce a new individual that inherits the advantages of both, further improving the path quality.

[0087] In practice, the first step is to perform multi-objective ranking of individuals within the crossover subpopulation. This ranking process requires comprehensively considering the values ​​of the two evaluation items involved. For example, the path smoothness evaluation value and the obstacle avoidance difficulty evaluation value can be compared in some way to determine the order of merit of the individuals. This comprehensive comparison can employ methods such as weighted summation or Pareto ranking, depending on the emphasis the navigation task places on each evaluation item. For instance, if the navigation task prioritizes obstacle avoidance safety, the weight of the obstacle avoidance difficulty evaluation item can be set higher in the multi-objective ranking.

[0088] After sorting, a selection probability is assigned to each individual according to a preset probability assignment rule. For example, using a linearly decreasing method, let the selection probability of the individual in the i-th position be...

[0089]

[0090] in: It is the first The probability of each position; It is the total number of positions; It is the current position index (usually starting from 1); the denominator of the expression. It was before The sum of natural numbers. n is the total number of individuals, ensuring that individuals ranked higher have a higher probability, and the sum of the probabilities of all individuals is 1. Alternatively, an exponential decay method can be used to make the increase in probability of individuals ranked higher more significant, strengthening the selection bias towards high-quality individuals.

[0091] When selecting a crossover partner, random sampling is performed based on the selection probability of each individual. For example, random selection methods such as roulette or tournaments can be used to select partners based on probability. In the roulette method, each individual corresponds to a region on the roulette wheel, the size of which is proportional to the selection probability. The individual whose region the pointer lands on is selected when the roulette wheel is spun randomly. In the tournament method, several individuals are randomly selected from the population each time, and the individual with the highest probability is chosen as the crossover partner. These methods ensure that individuals ranked higher have a higher probability of participating in the crossover operation.

[0092] The crossover operation can employ various path crossover strategies, such as single-point crossover, multi-point crossover, or sequential crossover. Taking sequential crossover as an example, a sub-path is selected as the crossover region from two parent paths. This sub-path is then swapped between the two parent paths, and the order of the remaining nodes is adjusted to generate a child path. In this way, the child path after crossover can inherit some structural features of the parent path. Especially when the parent path contains high-ranking, high-quality individuals, the child path is more likely to contain an optimized path structure.

[0093] After the selection and crossover of the cross objects are completed, the generated offspring individuals need to be updated in position, pheromone adjusted, and individual learned. Position updating refers to fine-tuning the node positions according to the path structure after the crossover, to adapt to the environmental constraints and evaluation items requirements; pheromone adjustment is to increase the pheromone concentration on the newly generated high-quality path to guide the subsequent search; individual learning refers to the offspring individuals referring to the path characteristics of other excellent individuals in the population to further optimize their own structure. At the same time, it is necessary to eliminate the infeasible path scheme to ensure the feasibility of the path.

[0094] Throughout the process, attention needs to be paid to the parameter settings of the dynamic adjustment mechanism. For example, the degree of association between selection probability and ranking position, the specific formula of probability assignment, etc. These parameters will directly affect the selection effect of the cross objects. If the difference in selection probability is too small, the advantages of high-quality individuals may not be effectively highlighted, resulting in the crossover operation not being obvious in promoting optimization; if the difference is too large, the algorithm may converge to a local optimum too early, reducing the global search ability. Therefore, these parameters need to be reasonably set according to the complexity of the navigation problem and the quality of the initial population.

[0095] In addition, the accuracy of multi-objective sorting is also crucial. When comprehensively sorting two evaluation items, it is necessary to ensure that the sorting result can truly reflect the comprehensive performance of the individual in these two dimensions. If the sorting rule is unreasonable, high-quality individuals may be incorrectly placed at the back, thereby reducing their selection probability and affecting the optimization effect. Therefore, when designing the multi-objective sorting rule, the physical meaning of the evaluation items and the actual needs of the navigation task need to be fully considered to ensure the scientificity of the sorting result.

[0096] At the same time, the size of the crossover sub-population will also affect the effect of the dynamic adjustment mechanism. If the sub-population size is too small, the number of high-quality individuals is limited, which may lead to insufficient diversity of the crossover operation; if the size is too large, it will increase the computational load. Therefore, the size of the crossover sub-population needs to be reasonably determined according to the computational resources and optimization needs.

[0097] By using the dynamic adjustment mechanism to select the cross objects before the crossover of the crossover sub-population, high-quality individuals with high ranking have a higher probability of being selected, which is conducive to the transmission of high-quality path characteristics in the crossover and accelerates the convergence of the iterative optimization to better solutions. Multi-objective sorting comprehensively considers relevant evaluation items to ensure that the selected cross objects perform better in key dimensions, improving the quality of the offspring paths after the crossover. The dynamic adjustment of the selection probability balances the exploration and utilization capabilities of the algorithm, avoids premature convergence to a local optimum, ensures the effective transmission of high-quality characteristics, further improves the efficiency of path optimization and the applicability of the final path, and makes the optimization process more in line with actual navigation needs.

[0098] Example 4:

[0099] In the multi-level collaborative planning robot navigation control method, screening the matched level identifier pairs is a key step to ensure the accuracy of the navigation model. The specific implementation is as follows: after obtaining the environment perception data collected by multiple nodes and extracting the navigation features to complete the matching, a series of level identifier pairs will be obtained. These identifier pairs may have mis-matching due to environmental interference, feature extraction error or limitations of the matching algorithm. Therefore, it is necessary to remove the mis-matched level identifier pairs through a screening mechanism, so as to construct a navigation model based on accurate level identifier pairs.

[0100] The core of the screening operation is to use the method of bidirectional matching and uniqueness verification. Bidirectional matching refers to verifying the matching relationship of the level identifier pairs from two different directions. For example, after the navigation features of the global planning node and the local obstacle avoidance node are preliminarily matched to obtain a set of level identifier pairs, the consistency of the matching needs to be verified from the direction of the global planning node to the local obstacle avoidance node, and then verified again from the direction of the local obstacle avoidance node to the global planning node, to ensure that the matching relationship in both directions corresponds to each other. Specifically, in the bidirectional matching process, first, the level descriptors and time sequence labels of the global planning node navigation features are extracted, and based on these information, the matching associated points are found in the navigation features of the local obstacle avoidance node, to obtain the forward matching level identifier pairs. Then, the level descriptors and time sequence labels of the local obstacle avoidance node navigation features are extracted, and the matching associated points are found in the navigation features of the global planning node in reverse, to obtain the reverse matching level identifier pairs. If the level identifier pairs obtained by forward and reverse matching are consistent, it is considered that the matching is reliable; if they are not consistent, it is determined as mis-matching and is removed.

[0101] Uniqueness verification is to ensure that each navigation feature only forms an effective match with a unique opposite feature in the matching process, avoiding one-to-many or many-to-many error matching situations. In specific implementation, first, all the level identifier pairs obtained by preliminary matching are combed to establish a mapping relationship table of navigation features and opposite features. For example, for a navigation feature A of the global planning node, it is checked in the mapping relationship table whether it corresponds to only one feature B of the local obstacle avoidance node, and whether feature B also corresponds to only feature A. If feature A corresponds to features B and C at the same time, or feature B corresponds to multiple global planning node features, it indicates that there is a uniqueness conflict, and these conflicting level identifier pairs need to be re-evaluated and screened.

[0102] In the specific operation of bidirectional matching, the hierarchical descriptor, as a key element describing the attribute sequence of navigation features, directly affects the effect of matching verification. The hierarchical descriptor usually contains the geometric attributes (such as coordinates, shape, size, etc.), semantic attributes (such as obstacle type, path point function, etc.), and time attributes (such as the timestamp of feature collection) of navigation features. In forward and reverse matching verification, the consistency of these attribute sequences needs to be compared one by one. For example, the hierarchical descriptor of a feature of a global planning node contains coordinates (x1, y1, z1), obstacle type is wall, and collection time is t1. In reverse verification, the hierarchical descriptor of the opposite feature matched in the local obstacle avoidance node should also contain similar coordinates (allowing a certain error range), the same obstacle type, and the collection time within a reasonable time range, otherwise it is considered as bidirectional matching failure.

[0103] The assignment and comparison of time sequence labels are also important links in bidirectional matching. Time sequence labels are used to identify the collection order and time correlation of navigation features. In bidirectional matching, it is necessary to ensure the logical consistency of the feature time sequence labels in forward and reverse matching. For example, the feature sequence collected in time order by the global planning node is F1, F2, F3, and the matched feature sequence of the local obstacle avoidance node is G1, G2, G3. In reverse verification, the time sequence labels of G1, G2, G3 should form a corresponding time mapping relationship with the time sequence labels of F1, F2, F3. If the time sequence label of G2 is earlier than that of G1, it may indicate a matching error.

[0104] In the uniqueness verification process, for the conflict in the mapping relationship table, further judgment needs to be made in combination with the environmental perception data and the spatial position relationship of navigation features. For example, when the feature A of the global planning node matches the features B and C of the local obstacle avoidance node, it needs to be checked whether the features B and C belong to the same obstacle or different feature descriptions of the same area in space. If features B and C are actually different perspective features of the same obstacle, they may need to be merged into an effective feature for matching; if they belong to different obstacles, it indicates that the matching of feature A is ambiguous, and it needs to be re-extracted or the matching algorithm parameters need to be adjusted to ensure uniqueness.

[0105] In the screening operation, a reasonable error tolerance range also needs to be set. Due to the limitations of sensor data collection accuracy and environmental changes, completely accurate bidirectional matching and uniqueness matching are difficult to achieve in practical applications. Therefore, according to the accuracy of the sensor, the speed of the robot, and the requirements of the navigation task, the allowed error range of each attribute comparison needs to be set. For example, the error range of coordinate comparison can be set to ±5 cm, and the error range of timestamp comparison can be set to ±100 ms. Differences within this range can be considered as valid matching, and differences beyond this range are considered as mismatching.

[0106] The specific implementation steps of the screening process are as follows: first, all the hierarchical identification pairs obtained through preliminary matching are traversed, and bidirectional matching verification is performed for each identification pair. In bidirectional matching verification, hierarchical descriptors and temporal labels are extracted from both directions for comparison, and the comparison results are recorded. If both bidirectional comparisons pass, it enters the uniqueness verification link; if the comparison of any direction does not pass, it is directly marked as a mismatched identification pair. After entering the uniqueness verification link, it checks whether there is a one-to-many or many-to-many mapping conflict for the identification pair. If there is no conflict, the identification pair is retained; if there is a conflict, a secondary judgment is made according to the spatial position relationship and error tolerance range. If it is confirmed as a mismatch, it is removed; if it is a reasonable conflict, the mapping relationship is adjusted.

[0107] After completing the screening of all hierarchical identification pairs, the valid identification pairs retained will be used as the reference points for constructing the navigation model. In constructing the navigation model, based on these valid identification pairs, the sparse path obtained by matching the navigation features between multiple nodes is expanded and screened into a complete or partial navigation path model covering the target area by using a region segmentation-based method through continuous association of navigation features. Since the screened hierarchical identification pairs have high accuracy, the constructed navigation model can more truly reflect the environmental structure and navigation requirements, providing a reliable basis for subsequent instruction packaging and standard control instruction generation.

[0108] In practical applications, the effect of the screening mechanism is also affected by the navigation feature extraction algorithm. If the navigation feature extraction algorithm can accurately capture key features in the environment and generate hierarchical descriptors with unique characteristics, the number of mismatches in the initial matching can be reduced, and the screening efficiency can be improved. Therefore, when implementing the screening operation, it needs to be optimized in coordination with the feature extraction algorithm, such as adjusting the parameters of feature extraction, enhancing the distinguishability and stability of features, thereby reducing the workload and misjudgment rate of subsequent screening.

[0109] In addition, the computational complexity of the screening operation needs to be matched with the computing resources of the robot. For robots with limited computing power, the algorithm process of bidirectional matching and uniqueness verification needs to be optimized, such as using index structure to accelerate feature comparison, or improving screening efficiency through parallel computing, to ensure that the screening operation can be completed within a reasonable time, without affecting the real-time requirements of robot navigation.

[0110] By bidirectional matching and uniqueness verification screening of hierarchical identifier pairs, the false matching identifier pairs can be effectively eliminated, and the reference points used to construct the navigation model are ensured to be accurate and reliable. Bidirectional matching can greatly reduce the matching errors caused by environmental interference or algorithm limitations by verifying the consistency of hierarchical descriptors and time sequence labels in forward and reverse directions; uniqueness verification avoids one-to-many or many-to-many matching conflicts and ensures the uniqueness of each navigation feature matching. The screened hierarchical identifier pairs make the constructed navigation model more truly reflect the environmental structure and navigation requirements, provide a reliable foundation for subsequent instruction encapsulation and control instruction generation, and improve the accuracy and stability of robot navigation.

[0111] Embodiment 5:

[0112] The present embodiment provides a multi-level collaborative planning robot navigation control system, which adopts the above-mentioned multi-level collaborative planning robot navigation control method, and the core component of the system is a controller. The controller undertakes key tasks such as data processing, path planning, model construction and instruction generation in the system, and realizes autonomous navigation control of the robot in a complex environment through collaborative work with various nodes such as global planning nodes, local obstacle avoidance nodes and execution control nodes.

[0113] The primary function of the controller is to obtain environmental perception data collected by various nodes. These nodes include global planning nodes, local obstacle avoidance nodes and execution control nodes, and different nodes have different perception ranges and functional orientations. The global planning node is usually equipped with sensors such as laser radar and visual camera, which is used to obtain global range information of the environment where the robot is located, such as large-scale terrain, building contour, etc.; the local obstacle avoidance node mostly uses ultrasonic sensor, short-distance visual camera, etc., which is used to perceive the distribution of obstacles within a short distance around the robot in real time; the execution control node may integrate encoder, gyroscope, etc., which is used to obtain the motion state data of the robot itself, such as position, speed, acceleration, etc. The controller establishes data transmission channels with these nodes through communication interfaces, receives environmental perception data collected by each node in real time, and pre-processes the data, including data filtering, format conversion, etc., to eliminate noise interference and ensure the accuracy and consistency of the data.

[0114] After obtaining the environmental perception data, the controller extracts navigation features according to these data. For the data of the global planning node, the controller mainly extracts features that can reflect the overall structure of the environment, such as the coordinates of key landmarks, the direction of the main path, etc.; for the data of the local obstacle avoidance node, the focus is on extracting the shape, position, size, etc. of the near-distance obstacles; for the data of the execution control node, the real-time pose information of the robot is extracted. When extracting navigation features, the controller extracts hierarchical descriptors at the navigation features, which are attribute sequences that describe navigation features, including geometric attributes, semantic attributes, and temporal attributes, etc. At the same time, the navigation features are assigned with time sequence labels to identify the time sequence and time correlation of feature collection. After completing feature extraction, the controller finds associated points according to the hierarchical descriptors, thereby obtaining matching hierarchical identification pairs. In the matching process, the controller uses a bidirectional matching and uniqueness verification method to filter the hierarchical identification pairs, eliminating false matches to ensure that the obtained hierarchical identification pairs are accurate and reliable.

[0115] Another important function of the controller is to construct a navigation model according to the environmental perception data and the filtered hierarchical identification pairs. The controller takes the matching hierarchical identification pairs as reference points, and by continuously associating navigation features, it uses a region segmentation-based method to expand and filter the sparse paths obtained by matching navigation features between multiple nodes, thereby obtaining a complete or partial navigation path model covering the target area. When constructing the navigation model, the controller will consider the environmental topology data, including the target area of the required navigation area, obstacle distribution, and node state data, etc. For example, according to the target area, the navigation area is identified and constructed, and the navigation path is planned according to the navigation area; at the same time, according to the obstacle distribution and node state data, the navigation reference is established to ensure that the navigation model can accurately reflect the environmental structure and navigation requirements.

[0116] In addition, the controller also needs to encapsulate the navigation model according to the environmental perception data to generate standard control instructions. The controller uses a hierarchical instruction encapsulation method, which encapsulates instructions through two levels. The first level encapsulates the information of the original perception space into an intermediate mode that can be expressed by the available strategy, which is an abstract representation form that is convenient for strategy processing and conversion, and can convert complex environmental perception information and navigation model information into a format that is easy for the strategy processing unit to understand and operate; the second level encapsulates the instructions of the intermediate mode into the target space that needs to be executed, generating standard control instructions that the robot execution mechanism can recognize and execute, such as motor speed control instructions, steering angle control instructions, etc. This hierarchical encapsulation method improves the flexibility and scalability of instruction generation, and can adapt to the execution mechanism and control requirements of different types of robots.

[0117] The controller also has the function of generating a navigation path according to the environmental topology data. Specifically, the controller first constructs a three-dimensional environmental map in the target area to be navigated, and then divides the target area to be navigated into a plurality of region units with a preset interval as the side length according to the three-dimensional environmental map. Then, the three-dimensional region units are classified according to the target area, obstacle distribution and node state data, including three types of core navigation area, general path area and obstacle avoidance area, and the core navigation area is marked with priority. Then, the controller randomly generates a plurality of initial navigation paths to form an initial navigation path set, and constructs an adaptability evaluation system including path smoothness evaluation item, obstacle avoidance difficulty evaluation item and energy consumption cost evaluation item. Taking the initial navigation path set as the initial population, the adaptability evaluation system is iteratively optimized, and finally the optimal navigation path is obtained. In the process of iterative optimization, the controller calculates the evaluation value of each initial navigation path according to the adaptability evaluation system, and performs sorting, selection, crossover and mutation operations to gradually optimize the path and ensure that the generated navigation path is optimal.

[0118] The nodes collect environmental perception data according to the navigation path generated by the controller. The controller sends the navigation path to each node through the communication link, each node adjusts the perception range and collection frequency according to the path information, and collects the environmental information along the path in a targeted manner, and transmits the collected data to the controller in real time to form a closed-loop feedback control. For example, the global planning node globally perceives the large-scale environment according to the navigation path to provide macro environmental information for path planning; the local obstacle avoidance node perceives the obstacle information around the path in real time during the movement of the robot along the path to provide data support for local obstacle avoidance; the execution control node feeds back the motion state of the robot in real time to ensure that the robot moves accurately according to the navigation path.

[0119] The controller plays a core role in the whole system, and realizes multi-level collaborative planning navigation control through comprehensive processing and analysis of data of each node. Its hardware structure usually includes central processor, memory, communication interface and input / output interface, etc. The central processor is responsible for executing data processing, path planning, model construction and instruction generation algorithms; the memory is used to store program code, environmental map data, navigation path data, etc.; the communication interface realizes data transmission with each node; the input / output interface is used to connect human-computer interaction devices to realize parameter setting and state display functions. On the software level, the controller runs the control program developed based on the above navigation control method, which includes data processing module, feature extraction module, path planning module, model construction module and instruction generation module, etc. Each module cooperates with each other to complete the navigation control task of the robot.

[0120] In practical applications, the performance of the controller directly affects the navigation effect of the robot. In order to improve the processing efficiency and real-time performance of the controller, parallel computing technology is usually used to perform parallel processing on data processing, path optimization and other tasks. At the same time, in order to adapt to complex and variable environments, the controller also has online learning and self-adaptive ability, which can continuously optimize algorithm parameters according to environmental changes and navigation experience, and improve the accuracy and reliability of navigation control.

[0121] By taking the controller as the core hub, the environmental perception data of global planning, local obstacle avoidance and execution control nodes can be efficiently integrated. By accurately extracting navigation features, matching and screening hierarchical identifier pairs, the accuracy of data processing is ensured. Based on these data, a navigation model is constructed, and a regional segmentation method is used to extend the sparse path to obtain a complete or local navigation path model, which can improve the integrity and adaptability of the path. Layered instruction packaging gradually converts raw perception information into executable standard control instructions, ensuring the accuracy and executability of the instructions. The controller generates the optimal navigation path combined with the environmental topology data, and realizes multi-node collaboration through closed-loop feedback, so that the robot can respond to environmental changes in real time, adjust the navigation strategy, and significantly improve the efficiency, accuracy and safety of navigation in complex environments, better meeting the needs of diversified navigation tasks.

[0122] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0123] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A robot navigation and control method based on multi-level collaborative planning, characterized in that, include: The system acquires environmental perception data collected from multiple nodes, extracts navigation features based on the environmental perception data, and matches the extracted navigation features to obtain hierarchical identifier pairs. These nodes include various types such as global planning nodes, local obstacle avoidance nodes, and execution control nodes; The process of obtaining hierarchical identifier pairs by matching the extracted navigation features includes: extracting hierarchical descriptors from the navigation features, assigning temporal labels to the navigation features, finding association points based on the hierarchical descriptors, and thus obtaining matching hierarchical identifier pairs; wherein the hierarchical descriptor is an attribute sequence describing the navigation features; A navigation model is constructed based on environmental perception data and hierarchical identifiers. Furthermore, the navigation model is encapsulated with instructions based on the environmental perception data to generate standard control commands, including: Using the hierarchical identifier pairs obtained through mutual matching as reference points, a navigation model is obtained through continuous association. A region-based segmentation method is adopted, and based on the sparse path obtained by matching navigation features between multiple nodes, a complete or partial navigation path model covering the target area is obtained through expansion and filtering. A layered instruction encapsulation method is adopted, which encapsulates instructions through two levels. The first level encapsulates the instructions from the original perception space into an intermediate mode that can be expressed by a strategy. The second level encapsulates the instructions of the intermediate mode into the target space that needs to be executed, generating standard control instructions. The system acquires the required environmental topology data for navigation and generates a navigation path based on this data; multiple nodes collect environmental perception data according to the navigation path. Environmental topology data includes the target area, obstacle distribution, and node status data of the required navigation area. A navigation path is generated based on this environmental topology data, including: Based on the target area, boundary identification is performed to construct a navigation area, and a navigation path is planned based on the navigation area; A navigation baseline is established based on obstacle distribution and node status data, and environmental perception data is collected based on the navigation baseline. The step of planning a navigation route based on the navigation area includes: A 3D environment map is constructed in the target area to be navigated, and the target area to be navigated is divided into several regional units with a preset spacing as the side length based on the 3D environment map. Based on the target area, obstacle distribution, and node status data, the three-dimensional region units are classified into three types: core navigation area, normal path area, and obstacle avoidance area. The core navigation area is given priority marking. Several initial navigation paths are randomly generated to form an initial navigation path set; An fitness assessment system is constructed, which includes a path smoothness assessment item, an obstacle avoidance difficulty assessment item, and an energy consumption cost assessment item. The path smoothness evaluation term is the cumulative value of the turning angles between each node in the initial navigation path; The obstacle avoidance difficulty assessment item is the sum of the obstacle density in each area unit of the initial navigation path; The energy consumption cost assessment item is the number of node power and steering parameters that need to be adjusted in the initial navigation path; Using the initial set of navigation paths as the initial population, iterative optimization is performed according to the fitness evaluation system to obtain the optimal navigation path, thus completing the navigation path planning, including: Based on the fitness evaluation system, the evaluation value of each initial navigation path is calculated. The initial navigation paths are then sorted from low to high according to the three evaluation items to obtain the sorted set of the three initial navigation paths. Based on the evaluation value, several initial navigation paths are selected from the three sorted sets by probability to form three subpopulations in different directions. The probability of being selected is inversely proportional to the evaluation value. Within each subpopulation, the ant colony optimization algorithm is used to update positions, adjust pheromones, and perform individual learning, while eliminating unreachable path schemes. The process is iterated a preset number of times to form three subpopulations after iteration. Individuals are randomly selected from the three subpopulations based on the evaluation values ​​to form a basic pheromone database. Three cross subpopulations are formed by cross-crossing the three subpopulations in pairs. For each cross subpopulation, three evaluation values ​​are calculated for each individual. The two evaluation values ​​involved in the cross subpopulations are sorted by multiple objectives. Based on the sorting results, position updates, pheromone adjustments, and individual learning are performed. Then, unreachable path schemes are eliminated. After iterating through the three crossover subpopulations a preset number of times, the three subpopulations are combined together, and the three evaluation items are weighted and summed to obtain a comprehensive evaluation value. Based on the comprehensive evaluation value, position updates, pheromone adjustments, and individual learning are performed. After iterating through the preset number of times, the individual with the smallest evaluation value is obtained as the optimal navigation path.

2. The robot navigation control method based on multi-level collaborative planning according to claim 1, characterized in that... In the process of multi-objective ranking of the two evaluation values ​​involved in the crossover subpopulation, and position updates, pheromone adjustments and individual learning based on the ranking results, a dynamic adjustment mechanism is used to select the crossover object before the crossover subpopulation crossover. The selection probability of the crossover object is related to the position of the individual in the ranking. The higher the position of the individual in the ranking, the higher the probability of being selected.

3. The robot navigation control method based on multi-level collaborative planning according to claim 2, characterized in that... It also includes: The selection probability is assigned a value based on the sorting order of the individuals to be crossed; the earlier an individual is in the sorting order, the higher the selection probability value.

4. The robot navigation control method based on multi-level collaborative planning according to claim 1, characterized in that... It also includes: The matching hierarchical identifier pairs are filtered, and mismatched hierarchical identifier pairs are filtered out. When building the navigation model, it is built based on the filtered hierarchical identifier pairs.

5. The robot navigation control method based on multi-level collaborative planning according to claim 4, characterized in that... The matching hierarchical identifier pairs are filtered, including using bidirectional matching and uniqueness verification to remove incorrectly matched hierarchical identifier pairs.

6. A multi-level cooperative planning robot navigation control system, using the multi-level cooperative planning robot navigation control method of any one of claims 1-5, comprising a controller, characterized in that: The controller is used to acquire environmental perception data collected by various nodes, extract navigation features based on the environmental perception data, and match the extracted navigation features to obtain hierarchical identifier pairs. The controller is also used to construct a navigation model based on environmental perception data and hierarchical identifiers, and to encapsulate the navigation model with instructions based on the environmental perception data to generate standard control instructions.

7. The multi-level collaborative planning robot navigation control system according to claim 6, characterized in that: The controller is also used to generate navigation paths based on environmental topology data, and various nodes collect environmental perception data based on the navigation paths.

Images