A fixed route unmanned navigation robot control method and related equipment

Abstract

The application discloses a fixed route unmanned navigation robot control method and related equipment, which can effectively meet the requirements of modern logistics operation. The application comprises the following steps: acquiring perception data of a region in front of a fixed path in real time; performing fusion processing on the perception data to obtain fusion data; generating obstacle information according to the fusion data, wherein the obstacle information comprises existence information about potential obstacles and a perception confidence degree corresponding to the obstacles; calculating a maximum allowed deviation of the robot from the fixed path for safe obstacle avoidance operation in real time according to a current position of the robot and the perception data; acquiring a real-time task priority of a task currently performed by the robot; constructing a decision function; and dynamically selecting and executing an optimal behavior strategy from pre-defined behavior strategies according to an output result of the decision function, wherein the behavior strategies at least comprise emergency stop, maintaining original path driving and deviation driving.

Description

Technical Field

[0001] This application relates to the field of robot navigation technology, and in particular to a control method and related equipment for an unmanned navigation robot operating on a fixed route. Background Technology

[0002] With the rapid development of industrial automation and intelligent logistics, mobile robots that use fixed-route navigation, such as automated guided vehicles and autonomous mobile robots, have been widely used in warehousing logistics, intelligent manufacturing workshops and other scenarios. These unmanned navigation robots usually travel along fixed paths formed by preset magnetic strips, QR codes or laser reflectors to perform tasks such as material handling, production line delivery and goods sorting.

[0003] However, in practical applications, when unmanned navigation robots travel on fixed paths, these paths often involve narrow passages and mixed human-robot traffic. In such situations, existing solutions mostly employ a simple detection-stop obstacle avoidance logic. That is, once the sensor detects a suspected obstacle ahead, regardless of the actual risk level of the target or the urgency of the robot's current task, the system usually triggers an emergency stop command. This simple response mode often leads to frequent and unnecessary emergency stops when faced with common interferences such as ground reflections, stationary goods, and sensor noise. This reduces the transportation efficiency and effective working time of a single robot, failing to meet the requirements of modern logistics operations. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a control method and related equipment for an unmanned navigation robot operating on a fixed route.

[0005] The technical solution provided in this application is described below:

[0006] The first aspect of this application provides a control method for an unmanned navigation robot operating on a fixed route, the method comprising: Real-time acquisition of sensing data of the area ahead along a fixed path; The sensed data is fused to obtain fused data; Obstacle information is generated based on the fused data, and the obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; Based on the robot's current position and the perception data, the maximum permissible deviation of the robot from the fixed path that it can safely deviate from in order to perform obstacle avoidance operations is calculated in real time. Obtain the real-time task priority of the task currently being performed by the robot; Construct a decision function, which is a function with the perceived confidence, the maximum allowable offset, and the real-time task priority as joint inputs; Based on the output of the decision function, the optimal behavior strategy is dynamically selected and executed from the predefined behavior strategies. The behavior strategies include at least emergency stop, maintaining the original path, and deviation.

[0007] Optionally, obstacle information is generated based on the fused data. This obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles, including: Based on the fused data, scene representation information of the area in front of the fixed path is generated, and the scene representation information includes at least an occupied grid map, a set of object detection boxes, or a semantic segmentation map. The scene representation information is analyzed to identify at least one candidate obstacle. The perceptual confidence level of each candidate obstacle is calculated based on the matching between the fused data and the scene representation information. The individual candidate obstacle and its corresponding perception confidence are encapsulated as obstacle information.

[0008] Optionally, based on the robot's current position and the perceived data, the maximum permissible deviation of the robot from the fixed path for obstacle avoidance operations is calculated in real time, including: Based on the robot's current position, the static safety boundaries on both sides of the fixed path at the current position are determined from real-time perception data. The static safety boundaries are determined by the positions of static obstacles on both sides of the path and a preset static safety margin. Calculate the dynamic space occupied by the robot body based on the robot's body geometry parameters and current motion state; Based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, the maximum permissible offset that the robot can safely deviate from to the left or right of the fixed path is calculated.

[0009] Optionally, based on the output of the decision function, the optimal behavioral strategy is dynamically selected and executed from predefined behavioral strategies, including: The output of the decision function is mapped to the corresponding predefined behavior strategy; Based on the output of the decision function and the maximum permissible offset, a set of control parameters corresponding to the mapped behavior strategy is generated. When the mapped behavior strategy is the offset driving, the set of control parameters includes at least one target lateral offset value that is not greater than the maximum permissible offset. The offset driving strategy is to control the robot to temporarily deviate from the fixed path by a distance less than or equal to the maximum permissible offset, and return to the fixed path after passing through the area where the potential obstacle is located. Motion control commands are generated based on the set of control parameters; The robot is driven by the control system to execute the mapped behavioral strategy.

[0010] Optionally, the perceived data is fused to obtain fused data, including: Perception data from different coordinate systems and at different times are mapped to the same spatiotemporal reference system centered on the robot to form registration data; For each detected target in the registration data, feature information based on different physical principles is extracted; Based on preset fusion rules, the feature information of the same detection target is associated and weighted by confidence level to generate the fused data.

[0011] Optionally, based on the robot's current position, the static safety boundaries on both sides of the fixed path at the current position are determined from real-time sensing data, including: Based on the robot's current position, the real-time sensing data is spatially rasterized along the lateral direction of the fixed path to construct a local environmental occupancy raster map. Static obstacle clustering and identification are performed on the local environment occupancy grid map to distinguish between dynamic objects and static obstacles, and the contour information of each static obstacle in the grid map is extracted. Based on the static obstacle contours on the corresponding sides of the fixed path, the preset static safety margin is extended outward in a direction parallel to the fixed path to generate corresponding left and right static safety boundary lines. The left and right static safety boundary lines are used as the two static safety boundaries of the fixed path at the current position.

[0012] Optionally, based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, the maximum permissible offset by which the robot can safely deviate to the left or right of the fixed path is calculated, including: Based on the geometric projection of the dynamically occupied space onto the cross section of the fixed path, one or more key geometric constraint elements are constructed to characterize the robot's motion envelope. For each side of the static safety boundary, calculate the real-time minimum safe distance between the key geometric constraint element and the corresponding side static safety boundary when the robot is along the fixed path; Based on the relationship between the real-time minimum safe distance and the preset safety response threshold, safe offset intervals are determined for the left and right sides of the robot, respectively. From the safe offset intervals on the left and right, the maximum allowable offset value that satisfies the preset offset strategy is selected as the maximum allowable offset of the robot to the left and right.

[0013] A third aspect of this application provides a control device for an unmanned navigation robot operating on a fixed route, the device comprising: The first acquisition unit is used to acquire real-time sensing data of the area ahead of a fixed path; The second acquisition unit is used to perform fusion processing on the perceived data to obtain fused data; The generation unit is used to generate obstacle information based on the fused data, the obstacle information including information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; The computing unit is used to calculate in real time the maximum allowable deviation of the robot from the fixed path for obstacle avoidance operations, based on the robot's current position and the perception data. The third acquisition unit is used to acquire the real-time task priority of the task currently being performed by the robot; A construction unit is used to construct a decision function, which is a function with the perception confidence, the maximum allowable offset, and the real-time task priority as joint inputs; The execution unit is used to dynamically select and execute the optimal behavior strategy from the predefined behavior strategies based on the output of the decision function. The behavior strategies include at least emergency stop, maintaining the original path, and deviation.

[0014] A fourth aspect of this application provides a control device for an unmanned navigation robot operating on a fixed route, the device comprising: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor invokes to perform the method as described in the first aspect and any one of the first aspects.

[0015] A fourth aspect of this application provides a computer-readable storage medium on which a program is stored, which, when executed on a computer, performs the method as described in the first aspect and any one of the first aspects.

[0016] As can be seen from the above technical solutions, this application has the following beneficial effects: 1. The method of this application fuses real-time sensing data and generates obstacle information containing presence information and perception confidence. Compared with single sensor sensing methods, this application can effectively reduce the probability of sensor false detection. By quantifying the detection reliability of potential obstacles through confidence, it can effectively reduce invalid operations caused by false obstacle information.

[0017] 2. This application combines the robot's current position and perception data to calculate the maximum permissible offset in real time, clarifying the safe boundary for the unmanned navigation robot to deviate from the fixed path during obstacle avoidance. This effectively avoids the risk of path loss of control caused by unlimited offset, while ensuring that the offset operation is always within a safe range, preventing the robot from colliding with the surrounding environment due to excessive deviation from the route, and improving the operational safety in fixed-route scenarios.

[0018] 3. This application constructs a decision function with perception confidence, maximum allowable offset, and real-time task priority as joint inputs, which is not significantly different from the decision-making mode of traditional fixed-route robots that "stop upon encountering an obstacle" or "fixed offset". Through the synergistic effect of multi-dimensional parameters, the robot can dynamically select the optimal strategy according to the actual scenario, minimizing interference with task execution and improving overall task completion efficiency while ensuring safety.

[0019] 4. The predefined behavioral strategies in this application can cover different scenarios in fixed-route navigation, and the decision function supports dynamic adjustment of strategy selection based on input parameters. For combinations of different obstacle confidence levels, different allowable offsets, and different task priorities, the robot can output differentiated optimal decisions, solving the problem of poor adaptability of traditional methods in complex dynamic environments, thereby effectively meeting the requirements of modern logistics operations. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of an embodiment of the unmanned navigation robot control method under a fixed route according to this application; Figure 2 This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 3 This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 4This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 5 This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 6 This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 7 This is a schematic diagram of another embodiment of the unmanned navigation robot control method under a fixed route in this application; Figure 8 This is a schematic diagram of an embodiment of the unmanned navigation robot control device under a fixed route according to this application; Figure 9 This is a schematic diagram of another embodiment of the unmanned navigation robot control device under a fixed route according to this application; Figure 10 This is a schematic diagram illustrating an application scenario of the unmanned navigation robot control device under a fixed route, as described in this application. Figure 11 This is a schematic diagram of another application scenario for the unmanned navigation robot control device under a fixed route, as described in this application. Detailed Implementation

[0022] It should be noted that the execution entity of the unmanned navigation robot control method under a fixed route described in this embodiment is not limited and can be flexibly configured according to the needs of the actual application scenario. For example, the execution entity of this control method can be the main controller, microprocessor, or embedded control system mounted on the unmanned navigation robot itself; it can also be a remote control terminal, edge computing node, or cloud server that establishes a communication connection with the unmanned navigation robot; or it can be a distributed control architecture composed of the robot's local control module and a remote control center. As long as the execution entity has the functions of data acquisition, data processing, calculation and analysis, decision generation, and instruction issuance / execution, and can realize the orderly advancement of each step in this method, it falls within the scope of protection of this invention.

[0023] In practical applications, when unmanned navigation robots travel on fixed paths, these paths often involve narrow passages and mixed human-robot traffic. In such situations, existing solutions often employ a simple detection-stop obstacle avoidance logic. This means that once the sensor detects a suspected obstacle ahead, regardless of the actual risk level of the target or the urgency of the robot's current task, the system usually triggers an emergency stop command. This simple response mode is prone to frequent and unnecessary emergency stops when faced with common interferences such as ground reflections, stationary goods, and sensor noise. This reduces the transportation efficiency and effective working time of a single robot, failing to meet the requirements of modern logistics operations.

[0024] Based on this, this application provides a control method and related equipment for an unmanned navigation robot operating on a fixed route, which is not significantly different from the decision-making mode of traditional fixed-route robots that "stops upon encountering an obstacle" or "fixed deviation". Through the synergistic effect of multi-dimensional parameters, the robot can dynamically select the optimal strategy according to the actual scenario, minimizing interference with task execution and improving overall task completion efficiency while ensuring safety.

[0025] Please see Figure 1 This application discloses a control method for an unmanned navigation robot operating on a fixed route, the method comprising: 101. Real-time acquisition of sensing data of the area ahead along a fixed path; 102. Perform fusion processing on the perceived data to obtain fused data; 103. Generate obstacle information based on the fused data, wherein the obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; 104. Based on the robot's current position and the perception data, calculate in real time the maximum permissible deviation of the robot from the fixed path for obstacle avoidance operations; 105. Obtain the real-time task priority of the task currently being executed by the robot; 106. Construct a decision function, wherein the decision function is a function with the perceived confidence, the maximum allowable offset and the real-time task priority as joint inputs; 107. Based on the output of the decision function, dynamically select and execute the optimal behavior strategy from the predefined behavior strategies, wherein the behavior strategies include at least emergency stop, maintaining the original path, and deviation.

[0026] In this embodiment, firstly, perception data of the area ahead of the fixed path is acquired in real time. Then, the perception data is fused to obtain fused data. Next, obstacle information is generated based on the fused data. The obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles. Then, based on the robot's current position and perception data, the maximum allowable deviation of the robot from the fixed path for obstacle avoidance is calculated in real time. Furthermore, the real-time task priority of the robot's current task is obtained. Based on the above, a decision function is constructed. The decision function is a function with perception confidence level, maximum allowable deviation, and real-time task priority as joint inputs. Finally, based on the output of the decision function, the optimal behavior strategy is dynamically selected and executed from predefined behavior strategies. The behavior strategies include at least emergency stop, maintaining the original path, and deviation, thereby realizing the control of the unmanned navigation robot under a fixed route.

[0027] In step 101, during the robot's operation along a preset fixed path, real-time perception data of the area ahead of the fixed path is acquired. Specifically, this data acquisition is achieved through a combination of multiple sensors mounted on the robot body. This combination includes a lidar, a vision camera, a millimeter-wave radar, and an ultrasonic sensor. Each sensor's detection range covers an area at a preset distance ahead of the fixed path. This preset distance is not specifically limited and can be determined based on the robot's speed and braking performance to ensure sufficient reaction time to handle unexpected obstacles. Specifically, the lidar is used to collect 3D point cloud data of objects within the area ahead of the path, obtaining the object's position, outline, and distance information; the vision camera is used to capture image data of the area ahead, identifying the object's shape features, color, and texture information; the millimeter-wave radar is used to stably collect distance and relative speed information of objects in adverse weather conditions such as rain, snow, and fog; and the ultrasonic sensor is used to accurately detect nearby objects, supplementing blind spots in the near-range perception. Each sensor synchronously collects data according to a preset sampling frequency to ensure the real-time nature of the perception data and the consistency of the timestamps.

[0028] In step 102, after acquiring the perception data of the area ahead of the fixed path, the perception data is further fused to obtain fused data. Since different types of sensors have their own advantages and limitations in the data acquisition process, data from a single sensor cannot comprehensively and accurately reflect the environmental conditions of the area ahead of the path. Therefore, a multi-sensor data fusion algorithm is needed to process the raw data collected by each sensor. Specifically, the data fusion algorithm adopts a hierarchical fusion architecture. First, data-level fusion is performed, and the raw data from each sensor are time-registered and spatially registered. Time registration matches data collected by different sensors at the same time through timestamp alignment. Spatial registration, based on the robot's kinematic model and the sensor's installation and calibration parameters, transforms the data collected by each sensor into a unified robot coordinate system, eliminating spatial deviations caused by differences in sensor installation positions.

[0029] Then, feature-level fusion is performed to extract key feature information from the data of each sensor, such as object contour features in lidar point cloud data, object edge features in visual image data, and object velocity features in millimeter-wave radar data. Feature matching algorithms are used to associate the feature information of the same object extracted by different sensors. Finally, decision-level fusion is performed to generate fused data, which can reflect the environmental features of the area ahead of the path.

[0030] In step 103, after acquiring the fused data, obstacle information is generated based on the fused data. This obstacle information includes information about the existence of potential obstacles and the corresponding perception confidence level. Specifically, using environmental feature information from the fused data, an obstacle recognition algorithm detects and classifies objects in the area ahead of the path. The obstacle recognition algorithm employs a deep learning model, trained on numerous obstacle samples from different scenarios, capable of identifying various potential obstacle types such as pedestrians, vehicles, and static obstacles. During the recognition process, the target area of ​​suspected obstacles is first extracted from the fused data. Then, the feature information of the target area is input into the deep learning model to determine whether a potential obstacle exists in the target area, generating information about the existence of potential obstacles, including the obstacle's location, type, size, and relative distance to the robot. Simultaneously, based on the contribution of each sensor during multi-sensor data fusion, data consistency, and feature clarity of the target area, the perception confidence level corresponding to the obstacle is calculated. The perception confidence level ranges from 0 to 1; a higher value indicates a higher reliability of the obstacle's existence. The specific calculation method is as follows: the detection confidence score of each sensor for the obstacle is multiplied by its corresponding weighting coefficient, and then a weighted sum is performed. The weighting coefficients are dynamically adjusted according to the detection performance of the sensors in the current environment. For example, in a strong light environment, the weighting coefficient of the visual camera is reduced, while the weighting coefficients of the lidar and millimeter-wave radar are increased, thereby ensuring that the perception confidence score can truly reflect the reliability of the obstacle detection results.

[0031] In step 104, further, based on the robot's current position and perception data, the maximum permissible deviation of the robot from the fixed path for obstacle avoidance is calculated in real time. The robot's current position is obtained through a positioning module, which includes an inertial navigation unit and a GPS positioning unit. These two components combine to achieve high-precision positioning. First, the position and outline of obstacles ahead are determined based on the perception data, clarifying the lateral distribution range of the obstacles in the robot's coordinate system. Then, combined with the robot's body dimensions, minimum turning radius, and travel speed, a safe boundary is determined to prevent the robot from colliding with obstacles during obstacle avoidance.

[0032] Meanwhile, considering the constraints of the working environment on both sides of the fixed path, such as obstacles like guardrails, equipment, and ditches, the maximum physical limit range of the robot's deviation from the fixed path is determined. Based on this, the lateral offset distance that the robot can safely bypass the obstacle from its current position is calculated in real time using a collision detection algorithm. The maximum value of this lateral offset distance is the maximum allowable offset. The calculation process of the maximum allowable offset needs to be updated synchronously with the robot's real-time position. When the position of the robot or the position of the obstacle changes, the maximum allowable offset is also dynamically adjusted to ensure that the offset always meets the obstacle avoidance safety requirements and does not exceed the physical constraints of the working environment.

[0033] In step 105, after obtaining the maximum permissible deviation of the robot from the fixed path for obstacle avoidance, the real-time task priority of the robot's current task is acquired. The robot's task system pre-stores multiple tasks, each with a preset base priority. The base priority is set based on the task's importance, urgency, and operational requirements. For example, the base priority of an emergency inspection task is higher than that of a regular cleaning task. During task execution, the robot receives instructions from the task scheduling center in real time or dynamically adjusts the task priority based on changes in the operational scenario. These changes include approaching deadlines and unexpected environmental conditions in the work area. The task priority is represented numerically; a higher value indicates a higher priority. Simultaneously, the task system monitors the current task's progress in real time. When the progress falls short of a preset threshold, the task priority is appropriately increased to ensure timely completion. The frequency of acquiring real-time task priorities matches the frequency of collecting sensor data, ensuring that the latest task priority data is used when making obstacle avoidance decisions.

[0034] In step 106, after obtaining the above information, a decision function is constructed based on the above information. It should be noted that the decision function is a function with perception confidence, maximum allowable offset and real-time task priority as joint inputs. The purpose of constructing the decision function is to establish a mapping relationship between input parameters and obstacle avoidance decisions in order to achieve dynamic selection of the optimal behavior strategy.

[0035] Specifically, the quantification indicators for each input parameter are first determined. The perceptual confidence level is quantified as a value between 0 and 1, the maximum permissible deviation is quantified as the lateral distance the robot can deviate from the fixed path, and the real-time task priority is quantified as a preset priority value. Then, the influence weight of each input parameter on the obstacle avoidance decision is analyzed. The higher the perceptual confidence level, the higher the reliability of the obstacle's existence, and the greater its influence on the obstacle avoidance decision. The larger the maximum permissible deviation, the more obstacle avoidance strategies the robot can choose. The higher the real-time task priority, the more likely the robot is to maintain the original path or deviate slightly to ensure task execution efficiency.

[0036] Based on the above, the expression for the decision function is: F = α × C + β × D + γ × P, where F is the output value of the decision function, C is the perception confidence level, D is the normalized value of the maximum allowable offset, P is the normalized value of the real-time task priority, and α, β, and γ are the weight coefficients of each input parameter. The weight coefficients are determined through training with a large amount of experimental data to ensure that the output value of the decision function can accurately reflect the comprehensive influence of each input parameter. At the same time, to adapt to different operation scenarios, the weight coefficients of the decision function can be dynamically adjusted according to the robot's operation type, without specific limitations here.

[0037] In step 107, based on the output of the decision function, the optimal behavior strategy is dynamically selected and executed from predefined behavior strategies. These strategies include at least emergency stop, maintaining the original path, and deviation. The predefined behavior strategies correspond to different intervals of the decision function's output value, and the specific strategy selection rules are determined through experimental calibration. When the output value of the decision function is in the first interval, it indicates extremely high perception confidence and a maximum allowable deviation of 0, meaning there is a clear obstacle ahead and the robot cannot avoid it by deviating from the path. In this case, the emergency stop strategy is selected, and the robot immediately activates its braking system to stop and avoid collision with the obstacle. When the output value of the decision function is in the second interval, it indicates low perception confidence or that the presence of the obstacle has no impact on the robot's movement along the original path, and the real-time task has a high priority. In this case, the maintain the original path strategy is selected, and the robot maintains its current speed and path direction, continues to perform the task, and continuously monitors the perception data of the area ahead. If the perception data changes, the strategy is adjusted promptly. When the output value of the decision function is in the third interval, it indicates that the perception confidence is high and the maximum allowable offset is greater than 0. The robot can avoid obstacles by deviating from the fixed path. At this time, the offset driving strategy is selected. The offset direction and offset distance are determined according to the maximum allowable offset. The robot's steering system is controlled so that the robot can drive along the safe path after the offset. After bypassing the obstacle, it returns to the original fixed path according to the positioning information and continues to perform the task.

[0038] During strategy execution, the robot collects new perception data in real time, recalculates the fused data, obstacle information, maximum allowable offset, and decision function output value to achieve dynamic adjustment of behavior strategy, ensuring safe operation and efficient task execution of the robot in complex environments.

[0039] For a schematic diagram illustrating the application scenarios of this application, please refer to [the relevant documentation / illustration]. Figure 10 and Figure 11 As shown.

[0040] Please refer to Figure 2 According to some embodiments of the present invention, in step 103, obstacle information is generated based on the fused data. The obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles, and may specifically include, but is not limited to, the following: 201. Generate scene representation information of the area in front of the fixed path based on the fused data, wherein the scene representation information includes at least an occupied grid map, a set of target detection boxes, or a semantic segmentation map; 202. Analyze the scene representation information to identify at least one candidate obstacle individual; 203. Calculate the perceptual confidence level corresponding to each candidate obstacle based on the matching between the fused data and the scene representation information; 204. Encapsulate the individual candidate obstacle and the perception confidence corresponding to the individual candidate obstacle into the obstacle information.

[0041] In this embodiment, after fusing the raw data collected by multiple sensors to obtain fused data, scene representation information of the area in front of the fixed path is generated based on the fused data. The scene representation information includes at least an occupied grid map, a set of object detection boxes, or a semantic segmentation map. Specifically, the generation of the occupied grid map involves dividing the environmental space within a preset range in front of the fixed path into several grid units of the same size. Based on information such as the distance and reflection intensity of each location point in the fused data, the occupied state of each grid unit is determined and marked as occupied, idle, or unknown. The occupied state corresponds to the grid with an object, the idle state corresponds to the grid without an object, and the unknown state corresponds to the grid not covered by the sensor or with insufficient data, thereby intuitively presenting the distribution of obstacles in the area in front of the path.

[0042] The generation of the object detection box set is based on the object feature information in the fused data. The object detection algorithm identifies various objects in the area in front of the path and generates a corresponding detection box for each identified object. The position and size of the detection box are determined by the object's coordinates and outline in the robot coordinate system. At the same time, the object category information, such as pedestrian, box, wall, etc., is labeled for each detection box.

[0043] The generation of semantic segmentation maps involves using semantic segmentation algorithms to classify the environmental regions corresponding to the fused data at the pixel level, aggregating pixels of objects of the same category, and clearly distinguishing regions with different semantic meanings, such as path regions, obstacle regions, and background regions. These three types of scene representation information depict the environment in front of the path from different dimensions, providing a multi-dimensional analytical basis for subsequent obstacle recognition.

[0044] After obtaining the scene representation information, further analysis is performed to identify at least one candidate obstacle. Since the scene representation information includes multi-dimensional data such as the spatial distribution of the environment, object outlines, and semantic categories, a comprehensive analysis combining the characteristics of various representation information is necessary. Specifically, firstly, connected component analysis is performed on the occupied grid map, merging adjacent occupied grids to form several connected regions, each corresponding to a potential object. Simultaneously, the set of target detection boxes is filtered, removing boxes that are too small or whose positions exceed the path's influence range, retaining objects corresponding to detection boxes that match the obstacle's size characteristics. Then, combined with the obstacle region in the semantic segmentation map, the potential objects obtained from the connected component analysis are matched with the objects labeled in the target detection boxes, eliminating objects belonging to the background or path regions, ultimately determining at least one candidate obstacle.

[0045] After identifying at least one candidate obstacle, the perceptual confidence score for each candidate obstacle is calculated based on the matching between the fused data and the scene representation information. Specifically, in this process, feature information of each candidate obstacle in the fused data is first extracted, such as the density of the LiDAR point cloud, the texture features of the visual image, and the reflection signal intensity of the millimeter-wave radar. At the same time, the corresponding features of the candidate obstacle in the scene representation information are extracted, such as the number of occupied grids in the occupied grid map, the bounding box selection accuracy in the target detection box set, and the pixel matching degree in the semantic segmentation map.

[0046] Then, the matching degree between the two sets of features is calculated. The matching degree is calculated using the cosine similarity algorithm, which measures the similarity of features by calculating the cosine value of the angle between feature vectors. The higher the matching degree, the more consistent the fused data and scene representation information are in depicting the candidate obstacle. Finally, the matching degree is weighted and corrected by combining the weight coefficients of multiple sensors. For example, for candidate obstacles with clear data from LiDAR, the weight of the corresponding LiDAR features is increased; for candidate obstacles clearly identified by the visual camera, the weight of the visual features is increased. The matching degree after weighting and correction is the perceptual confidence degree corresponding to the candidate obstacle. The perceptual confidence degree ranges from 0 to 1. The higher the value, the higher the reliability of the existence of the candidate obstacle and its related feature information.

[0047] After the above processing, the candidate obstacle individuals and their corresponding perception confidence scores are encapsulated into obstacle information. Specifically, the basic information of each candidate obstacle individual, including its position coordinates in the robot coordinate system, its shape and size, and its object category, is associated and bound with its corresponding perception confidence score. This encapsulation is performed using a preset data structure, which includes an identifier field, an attribute field, and a confidence score field. The identifier field is used to distinguish different candidate obstacle individuals, the attribute field is used to store the position, size, category, and other attribute information of the candidate obstacle individual, and the confidence score field is used to store the corresponding perception confidence score value. The encapsulated obstacle information can completely and clearly present the specific situation of potential obstacles in the area ahead of the path. The encapsulated obstacle information can provide a basis for the robot to calculate the maximum allowable offset, construct the decision function, and select the optimal behavior strategy, ensuring that the robot can fully consider the actual situation of the obstacle and the reliability of the detection results when making obstacle avoidance decisions.

[0048] Please refer to Figure 3 According to some embodiments of the present invention, in step 104, based on the robot's current position and the perception data, the maximum permissible deviation of the robot from the fixed path for obstacle avoidance operations is calculated in real time. This may specifically include, but is not limited to, the following: 301. Based on the robot's current position, determine the static safety boundaries on both sides of the fixed path at the current position from real-time perception data. The static safety boundaries are determined by the positions of static obstacles on both sides of the path and a preset static safety margin. 302. Calculate the dynamic space occupied by the robot body based on the robot's body geometric parameters and current motion state; 303. Based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, calculate the maximum permissible offset by which the robot can safely deviate to the left or right of the fixed path.

[0049] In this embodiment, during the robot's task execution along a fixed path, the static safety boundaries on both sides of the fixed path at the current position are determined from real-time sensing data based on the robot's current position. These static safety boundaries are determined by the positions of static obstacles on both sides of the path and a preset static safety margin. Specifically, the robot's positioning module outputs the current position coordinates in real-time. These coordinates are then matched against the global coordinate system of the fixed path to pinpoint the path segment corresponding to the current position. Subsequently, static obstacle information on both sides of the path segment is filtered from real-time sensing data collected by multiple sensors. It should be noted that static obstacles include, but are not limited to, objects that do not shift, such as guardrails, curbs, and fixed equipment bases along the path edge. Through feature extraction and classification of sensor data, the precise position of each static obstacle in the robot's coordinate system is determined.

[0050] Furthermore, to prevent the robot from scraping against static obstacles due to sensor detection errors or minor vibrations, a preset static safety margin needs to be set between the actual boundary of the static obstacle and the robot's permissible range of movement. The value of this safety margin is determined based on the robot's dimensions, the spatial constraints of the work environment, and the sensor's detection accuracy. For example, in a narrow passage scenario, the static safety margin is 10% of the robot's width, and in an open scenario, it is 15% of the robot's width. This is calculated by offsetting the actual position of the static obstacle towards the center of the fixed path by the distance corresponding to the static safety margin, forming the left and right static safety boundaries of the fixed path at the current position. The area between these two static safety boundaries constitutes the basic safe passage space for the robot when moving along the fixed path at the current position.

[0051] After determining the static safety boundaries on both sides, the dynamic space occupied by the robot body is further calculated based on the robot's body geometry parameters and current motion state. The robot's body geometry parameters include the body length, width, height, and outer contour dimensions of each component. These parameters are pre-stored in the control system when the robot leaves the factory and will not be described in detail here.

[0052] The robot's current motion state includes its real-time driving speed, turning angular velocity, and acceleration. These parameters are collected and fed back in real time by the robot's motion control module. Since the robot is not stationary during movement, its motion state affects the actual space it occupies. For example, when the robot turns at a certain turning angular velocity, it will generate greater space requirements due to centrifugal motion. When the robot is accelerating or decelerating, its inertial motion trend will also affect the reservation of safe space.

[0053] Therefore, when calculating the dynamic occupied space, the static occupied space contour of the robot in a stationary state is first determined based on the robot's body geometry parameters. Then, combined with the current travel speed and steering angular velocity, the real-time positional offset of each part of the robot during movement is calculated using a kinematic model. Simultaneously, a preset dynamic safety margin is considered. This margin is used to compensate for inertial offsets and control response delays during robot movement. Finally, the static occupied space contour is superimposed with the motion offset and dynamic safety margin to form the dynamic occupied space of the robot in its current movement state. The resulting dynamic occupied space accurately reflects the safe passage space actually required by the robot during movement, effectively avoiding obstacle avoidance decision errors caused by considering only the static contour.

[0054] After obtaining the static safety boundaries on both sides and the dynamic occupied space, the maximum permissible offset that the robot can safely deviate from to the left or right of the fixed path is calculated based on the real-time relative positional relationship between the dynamic occupied space and the static safety boundaries on both sides. Specifically, the coordinate information of the dynamic occupied space and the static safety boundaries on both sides is first uniformly transformed into the local coordinate system of the robot's current position to facilitate real-time calculation of the positional relationship. Then, the relative distances between the dynamic occupied space and the left and right static safety boundaries are analyzed respectively.

[0055] When the robot travels along the centerline of a fixed path, a certain safety gap exists between the left and right edges of the dynamically occupied space and the corresponding static safety boundary. When the robot needs to deviate from the fixed path to the left or right to avoid obstacles, the size of this safety gap determines the robot's permissible offset range. When calculating the maximum permissible offset to the left, the real-time distance between the left edge of the dynamically occupied space and the left static safety boundary is used as a benchmark. This ensures that after the robot offsets to the left, the left edge of the dynamically occupied space will not exceed the left static safety boundary. Simultaneously, it ensures that a distance of not less than a preset minimum safety gap remains between the offset dynamically occupied space and the left static safety boundary. This minimum safety gap is used to accommodate changes in the robot's motion state during the offset process.

[0056] Similarly, the calculation of the maximum permissible offset on the right is based on the real-time distance between the right edge of the dynamically occupied space and the right static safety boundary. After calculation, the final maximum permissible offset on the left and the maximum permissible offset on the right are the limit distances that the robot can safely deviate from the fixed path to the left or right at its current position. The calculation results of this offset will be fed back to the robot's decision system in real time, providing key spatial constraint parameters for the selection of subsequent obstacle avoidance strategies.

[0057] Please refer to Figure 4According to some embodiments of the present invention, in step 107, the optimal behavior strategy is dynamically selected and executed from predefined behavior strategies based on the output of the decision function. This may specifically include, but is not limited to, the following: 401. Map the output of the decision function to the corresponding predefined behavior strategy; 402. Based on the output of the decision function and the maximum permissible offset, generate a set of control parameters corresponding to the mapped behavior strategy, wherein when the mapped behavior strategy is the offset driving, the set of control parameters includes at least one target lateral offset value not greater than the maximum permissible offset; the offset driving strategy is to control the robot to temporarily deviate from the fixed path by a distance less than or equal to the maximum permissible offset, and return to the fixed path after passing through the area where the potential obstacle is located; 403. Generate motion control commands based on the control parameter set; 404. Drive the robot to execute the mapped behavior strategy according to the control.

[0058] In this embodiment, after the robot completes the calculation of the decision function output, the output of the decision function is mapped to a corresponding predefined behavior strategy. Specifically, the mapping is based on a pre-calibrated correspondence between the decision function output value range and the behavior strategy. This correspondence is determined through extensive simulation experiments and real-world scenario testing. For example, when the decision function output value is in a high-confidence range with a maximum allowable offset of 0, it is directly mapped to an emergency stop strategy; when the output value is in a low-confidence range or when obstacles do not affect the original path travel, it is mapped to a path-maintaining strategy; when the output value is in a high-confidence range with a maximum allowable offset greater than 0, it is mapped to an offset travel strategy. During the mapping process, the system verifies the output value of the decision function to eliminate output value jumps caused by abnormal sensor data, ensuring that the mapped behavior strategy matches the current environmental state and task requirements.

[0059] Then, based on the output of the decision function and the maximum permissible offset, a set of control parameters corresponding to the mapped behavior strategy is generated. When the mapped behavior strategy is offset driving, the control parameter set includes at least one target lateral offset value not greater than the maximum permissible offset. The offset driving strategy involves controlling the robot to temporarily deviate from the fixed path by a distance less than or equal to the maximum permissible offset, and then returning to the fixed path after passing through the area containing potential obstacles.

[0060] Specifically, the generation logic of the control parameter set differs for different behavioral strategies. For the emergency stop strategy, the control parameter set includes parameters such as braking deceleration and braking response time. These parameters are determined based on the robot's current speed and load to ensure that the robot can stop smoothly within a safe distance. For the strategy of maintaining the original path, the control parameter set includes parameters such as target speed and path tracking accuracy threshold to ensure that the robot can travel accurately along a fixed path while maintaining continuous monitoring of the environment ahead. For the deviation strategy, the minimum lateral deviation distance required for safe obstacle avoidance is calculated first, using the maximum allowable deviation as the upper limit, combined with the position and size of the obstacle and the robot's body size. Then, the optimal target lateral deviation value is selected from the interval between the minimum lateral deviation distance and the maximum allowable deviation. The selection of this target lateral deviation value must take into account both obstacle avoidance safety and the convenience of path return.

[0061] It should be noted that the control parameter set of the offset driving strategy also includes parameters such as offset steering angle, offset driving speed, and path return point coordinates. The path return point coordinates are determined based on the rear end position of the obstacle to ensure that the robot can successfully return to the fixed path after passing through the area where the potential obstacle is located.

[0062] After obtaining the control parameter set, motion control commands are generated based on it. Specifically, the system converts various parameters in the control parameter set into a command format recognizable by the robot's motion actuators. For emergency stop strategies, motion control commands include braking system activation signals, braking pressure values, and other instructions, clearly specifying the braking trigger timing and duration to avoid accidental or incomplete braking. For maintaining the original path travel strategy, motion control commands include drive motor speed commands and steering system angle holding commands to ensure stable robot movement.

[0063] For the offset driving strategy, motion control commands are broken down into three phases: offset steering phase commands, offset constant speed driving phase commands, and path return phase commands. The offset steering phase commands include the target angle and steering speed of the steering motor, ensuring the robot can smoothly complete the steering action. The offset constant speed driving phase commands include the target speed of the drive motor, ensuring the robot travels at a constant speed along the offset path. The path return phase commands include the return angle and return speed of the steering motor, guiding the robot precisely back to the fixed path. It should be noted that during the generation of motion control commands, the system verifies the rationality of the commands. For example, it verifies whether the steering angle is within the robot's steering limits and whether the drive motor speed meets the performance requirements of the power system, avoiding robot malfunctions caused by command parameters exceeding the capabilities of the actuators.

[0064] After generating control commands, the robot executes the mapped behavior strategy according to the commands. The robot's motion control unit receives the motion control commands and transmits the command signals to the corresponding actuators. For the emergency stop strategy, the braking actuator, upon receiving the command, quickly applies braking pressure to stop the drive wheels from rotating, and simultaneously triggers the vehicle stability system to ensure the robot does not roll over or deviate during the stopping process. For the path-maintaining strategy, the drive motor and steering system maintain the current operating state according to the commands, and the robot continues to travel along the fixed path. Simultaneously, the perception system continuously collects perception data of the area ahead and updates the input parameters of the decision function in real time to adjust the strategy promptly when the environment changes. For the deviation strategy, the steering actuator adjusts the steering angle according to the deviation steering stage command, causing the robot to deviate from the fixed path according to the target lateral deviation value. The drive motor adjusts its speed according to the commands to ensure the robot's travel speed remains stable on the deviation path.

[0065] When the robot reaches the path return point, the steering actuator adjusts the steering angle according to the path return phase instructions, guiding the robot back to the fixed path. Throughout the execution of the behavior strategy, the motion control unit monitors the operating status of the actuator and the robot's position information in real time, and feeds the monitoring data back to the decision system to form a closed-loop control, ensuring that the robot can accurately and safely execute the mapped behavior strategy, complete obstacle avoidance operations, and continue to perform the task.

[0066] Please refer to Figure 5 According to some embodiments of the present invention, step 102 involves fusing the perceived data to obtain fused data, which may specifically include, but is not limited to, the following: 501. Map the perception data from different coordinate systems and at different times to the same spatiotemporal reference system centered on the robot to form registration data; 502. For each detected target in the registration data, extract feature information based on different physical principles; 503. Based on preset fusion rules, the feature information of the same detection target is associated and weighted by confidence level to generate the fused data.

[0067] In this embodiment, during the fusion processing of perception data collected by multiple sensors, perception data from different coordinate systems and at different times need to be mapped to the same spatiotemporal reference system centered on the robot to form registration data. Specifically, since different types of sensors, such as LiDAR, visual cameras, and millimeter-wave radar, are installed at different positions on the robot body, the detection coordinate systems of each sensor have inherent differences, and there is a time difference in the sampling triggering time of different sensors. If the original perception data is processed directly, problems such as target positioning deviation and feature information misalignment will occur due to the inconsistency of spatiotemporal reference. Therefore, it is necessary to complete the spatiotemporal registration operation first.

[0068] During the spatial registration process, based on the pre-calibrated extrinsic parameter matrix of each sensor relative to the robot's central coordinate system, the detection data of all sensors, including the 3D point cloud coordinate system of the LiDAR, the pixel coordinate system of the vision camera, and the polar coordinate system of the millimeter-wave radar, are uniformly transformed into the robot's central coordinate system with the robot's center of mass as the origin, the robot's forward direction as the X-axis, the horizontal axis perpendicular to the forward direction as the Y-axis, and the vertical axis upward as the Z-axis, thus eliminating spatial coordinate deviations caused by differences in sensor installation positions.

[0069] During time registration, the robot's motion control clock is used as the reference clock to synchronize and calibrate the timestamps of data collected by each sensor. For sensing data whose sampling time is earlier than the reference time, the spatial position of the data at the reference time is calculated by combining the robot's motion trajectory within the corresponding time difference and using a kinematic model. For sensing data whose sampling time is later than the reference time, it is temporarily stored in a buffer queue to be processed after the reference time is triggered. Ultimately, all sensing data are fully aligned in both the time and spatial dimensions, forming registration data with a unified spatiotemporal reference.

[0070] After obtaining the registration data, feature information based on different physical principles is extracted for each detected target in the registration data. It should be noted that the detected target refers to all suspected obstacle objects in the area in front of the path identified from the registration data. Since different sensors achieve detection based on different physical principles, the data they collect can reflect different attribute characteristics of the detected targets. Therefore, feature information corresponding to different physical principles needs to be extracted for each detected target.

[0071] Specifically, for registration data acquired by lidar, the three-dimensional geometric features of the detected target are extracted, including the target's outline dimensions, surface area, volume, centroid coordinates, and point cloud density distribution on the target surface. These features, based on the physical principle of lidar's laser reflection ranging, can accurately reflect the target's spatial morphology. For registration data acquired by visual cameras, visual features of the detected target are extracted, including the target's color histogram, texture features, edge gradient features, and target category semantic features. These features, based on the physical principles of optical imaging and image processing, can reflect the target's appearance attributes and category information. For registration data acquired by millimeter-wave radar, motion features of the detected target are extracted, including the target's relative velocity, relative acceleration, and radial distance change rate. These features, based on the physical principle of the Doppler effect of electromagnetic wave reflection, can accurately reflect the target's dynamic motion state. For registration data acquired by ultrasonic sensors, near-range features and reflection intensity features of the detected target are extracted. These features, based on the physical principle of acoustic wave reflection ranging, can supplement the near-range attribute information of the detected target. By extracting multi-dimensional and multi-physical feature information for each detection target, comprehensive and differentiated feature data can be provided for subsequent feature fusion.

[0072] After obtaining feature information from different physical principles, the feature information of the same detection target is correlated and weighted by confidence level based on a preset fusion rule to generate the fused data. It should be noted that the preset fusion rule is trained based on a large amount of sensor detection experimental data from different scenarios, and the fusion rule pre-sets the correlation conditions and confidence weight coefficients for different types of feature information.

[0073] Specifically, feature association processing is performed first. For different physical principle feature information of the same detection target, the feature matching algorithm is used to verify whether the target corresponding to each feature information is the same entity. Specifically, by comparing the overlap of key parameters such as the centroid coordinates and contour size of the target in different feature information, when the overlap is higher than a preset threshold, it is determined to be feature information of the same detection target and associated, and redundant feature information that fails to be associated is eliminated.

[0074] Then, confidence-weighted processing is performed. The fusion rules assign dynamic weight coefficients to feature information based on different physical principles, according to the detection accuracy of different sensors in different environments. For example, in direct sunlight, visual features captured by a visual camera are easily interfered with, resulting in a lower weight coefficient, while the weight coefficient of geometric features from a LiDAR is increased. In rainy or foggy weather, the weight coefficient of motion features from a millimeter-wave radar is increased. Simultaneously, combining the confidence levels of each feature, a weighted summation method is used to calculate the comprehensive feature confidence level of the detected target. Finally, the correlated multi-dimensional feature information and the comprehensive feature confidence level are integrated to form fused data that reflects the attributes of the detected target.

[0075] Please refer to Figure 6 According to some embodiments of the present invention, in step 301, the static safety boundaries on both sides of the fixed path at the current position are determined from real-time sensing data based on the robot's current position. Specifically, this may include, but is not limited to, the following: 601. Based on the robot's current position, the real-time sensing data is spatially rasterized along the lateral direction of the fixed path to construct a local environment occupancy raster map. 602. Perform static obstacle clustering and identification on the local environment occupancy grid map, distinguish between dynamic objects and static obstacles, and extract the contour information of each static obstacle in the grid map; 603. Based on the static obstacle contours of the corresponding sides of the fixed path, extend the preset static safety margin outward in a direction parallel to the fixed path to generate corresponding left static safety boundary lines and right static safety boundary lines. 604. The left static safety boundary line and the right static safety boundary line are taken as the two static safety boundaries of the fixed path at the current position.

[0076] In this embodiment of the application, during the process of the robot performing a task based on a fixed path, the real-time perception data is spatially rasterized along the lateral direction of the fixed path, with the robot's current position as the reference, in order to construct a local environmental occupancy raster map.

[0077] Specifically, the first step is to determine the spatial range of the gridded processing. This range is centered on the robot's current position, extends longitudinally along a fixed path to a preset sensing distance, and laterally to both sides along the fixed path to 1.5 times the robot's maximum allowable offset range, ensuring that the entire environmental area required for the robot to avoid obstacles is covered.

[0078] The space is then divided into uniformly sized grid cells. The size of each grid cell is determined by the sensor's detection accuracy; higher-precision sensors such as LiDAR and vision cameras use smaller grid cells to ensure accurate representation of environmental details. The environmental information from the real-time sensing data is then mapped to each grid cell. By calculating the probability that each grid cell is occupied by an obstacle, a state label of "occupied," "idle," or "unknown" is assigned to each cell. "Occupied" indicates the presence of an obstacle, "idle" indicates no obstacle, and "unknown" indicates that the environmental information within the cell has not been detected by the sensor. This process ultimately constructs an occupied grid map that reflects the real-time local environmental conditions surrounding the robot's current location.

[0079] After obtaining the local environment occupancy grid map, static obstacle clustering and identification are performed on the local environment occupancy grid map to distinguish between dynamic objects and static obstacles, and the contour information of each static obstacle in the grid map is extracted. It should be noted that since the local environment occupancy grid map contains grid data of both static obstacles and dynamic objects, and their impact on robot path planning differs, classification processing is required first. Specifically, a grid state analysis method based on temporal difference is used to compare the state changes of the same grid cell in multiple consecutive frames of the local environment occupancy grid map. If the state of a grid cell in a certain area remains stable across multiple frames without any positional movement, the obstacle in that area is determined to be a static obstacle. If the state of a grid cell in a certain area changes position across multiple frames, the object in that area is determined to be a dynamic object.

[0080] Then, a clustering algorithm is used to cluster the grid cells that are determined to be static obstacles. Adjacent grid cells that are occupied are grouped into the same class to form independent static obstacle clustering regions. For each static obstacle clustering region, the contour information in the grid map is extracted by the edge detection algorithm. The contour information includes the boundary grid coordinates, shape features and geometric center position of the clustering region.

[0081] After processing the above, based on the outlines of the static obstacles on the corresponding sides, a preset static safety margin is extended outward in a direction parallel to the fixed path on the left and right sides to generate corresponding left and right static safety boundary lines. It should be noted that this preset static safety margin is a safety distance pre-set based on the robot's body size, travel speed, and braking performance. Its value is greater than half of the robot's maximum lateral width, ensuring that the robot will not collide with static obstacles when traveling near the boundary lines.

[0082] When generating the left static safety boundary line, the outlines of all static obstacles on the left side of the fixed path are first extracted, and the closest point of each static obstacle outline to the fixed path is determined. Then, based on the closest point, the preset static safety margin is extended to the left along the horizontal direction of the fixed path to obtain a series of discrete safety boundary points. Then, an interpolation algorithm is used to fit these discrete safety boundary points along a direction parallel to the fixed path to form a continuous left static safety boundary line.

[0083] Similarly, when generating the right static safety boundary line, the outlines of all static obstacles on the right side of the fixed path are extracted, the closest point of each static obstacle outline to the fixed path is determined, and the preset static safety margin is extended to the right along the fixed path with the closest point as the reference. After obtaining discrete safety boundary points, a continuous right static safety boundary line is formed by fitting the data through an interpolation algorithm.

[0084] Finally, the left and right static safety boundary lines are used as the two static safety boundaries of the fixed path at the current position. It should be noted that these safety boundaries are updated in real time as the robot moves and the distribution of static obstacles changes. When calculating the maximum permissible deviation of the robot from the fixed path, these static safety boundaries will serve as a crucial constraint, ensuring that the calculated maximum permissible deviation does not exceed the range of the static safety boundaries. This guarantees the robot's driving safety during obstacle avoidance and also provides key parameters related to the static environment for constructing the decision function, thus connecting static environment constraints with the robot's obstacle avoidance decisions.

[0085] Please refer to Figure 7 According to some embodiments of the present invention, in step 303, based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, the maximum permissible offset by which the robot can safely deviate to the left or right of the fixed path is calculated. Specifically, this may include, but is not limited to, the following: 701. Based on the geometric projection of the dynamically occupied space onto the cross section of the fixed path, construct one or more key geometric constraint elements to characterize the robot's motion envelope; 702. For each side of the static safety boundary, calculate the real-time minimum safe distance between the key geometric constraint element and the corresponding side static safety boundary when the robot is along the fixed path; 703. Based on the relationship between the real-time minimum safe distance and the preset safety response threshold, determine the safe offset intervals for the left and right sides of the robot respectively; 704. From the safe offset intervals on the left and right, select the maximum allowable offset value that satisfies the preset offset strategy as the maximum allowable offset of the robot to the left and right respectively.

[0086] In this embodiment, during the robot's obstacle avoidance decision-making based on fused data and obstacle information, one or more key geometric constraint elements are constructed to characterize the robot's motion envelope based on the geometric projection of the dynamically occupied space onto the cross-section of the fixed path. It should be noted that the dynamically occupied space here refers to the space occupied by the robot itself and its auxiliary working parts during movement. This space changes dynamically with the robot's driving posture and the action state of the working parts. The fixed path cross-section refers to the section perpendicular to the robot's direction of travel along the fixed path.

[0087] When constructing key geometric constraint elements, the robot's dynamic occupied space is first projected onto the cross-section to obtain the corresponding projected contour. Then, based on the geometric features of the projected contour, key elements that characterize the robot's motion envelope boundary are extracted. These key geometric constraint elements can be vertices, edge segments, curvature extrema, etc., of the projected contour. For example, for a robot with a rectangular projected contour, its key geometric constraint elements may include the four vertices and four edge segments of the rectangle. For a robot with an irregular projected contour, feature points and feature segments on the contour can be extracted as key geometric constraint elements. These elements can define the spatial occupancy boundary of the robot on the fixed path cross-section during its movement.

[0088] After constructing the key geometric constraint elements, for each side of the static safety boundary, the real-time minimum safe distance between the key geometric constraint element and the corresponding side static safety boundary along the fixed path is calculated. Here, the static safety boundary refers to the pre-defined boundaries on both sides of the fixed path that the robot cannot cross during its movement. These boundaries are typically determined based on factors such as fixed obstacles and work area boundaries in the work scenario. For example, guardrails, walls, and equipment bases on both sides of the path can all serve as static safety boundaries. When calculating the real-time minimum safe distance, first, the cross-section of the fixed path corresponding to the robot's current position is determined, and the positions of the static safety boundaries on the left and right sides of this cross-section are identified. Then, for each side of the static safety boundary, all key geometric constraint elements corresponding to that side are traversed, and the straight-line distance between each key geometric constraint element and the corresponding side static safety boundary is calculated using the spatial distance calculation formula. Finally, the minimum value is selected from these calculated distance values ​​and determined as the real-time minimum safe distance between the key geometric constraint element and the static safety boundary on that side.

[0089] It should be noted that this calculation process needs to be synchronized with the robot's movement. As the robot moves along a fixed path, the projected position of its dynamically occupied space and the coordinates of key geometric constraint elements will change in real time. Therefore, the real-time minimum safe distance will also be dynamically updated to ensure that the calculation results can truly reflect the real-time spatial relationship between the robot and the static safety boundary.

[0090] Next, based on the relationship between the real-time minimum safe distance and the preset safety response threshold, safe offset intervals are determined for the left and right sides of the robot. The preset safety response threshold is a distance threshold pre-set based on factors such as the robot's braking performance, steering flexibility, and safety requirements of the work scenario. It is not specifically limited here. This threshold is an important basis for judging whether the distance between the robot and the static safety boundary is safe.

[0091] When determining the safe offset range, the left and right sides of the robot are analyzed separately. For either side, if the calculated real-time minimum safe distance is greater than the preset safety response threshold, it indicates that the distance between the robot and the static safety boundary on that side is sufficient, providing some offset space. In this case, the safe offset range for that side is determined with the real-time minimum safe distance as the upper limit and 0 as the lower limit, meaning the robot can offset within the range of 0 to the real-time minimum safe distance on that side. If the real-time minimum safe distance is equal to the preset safety response threshold, it indicates that the distance between the robot and the static safety boundary on that side is at a safety threshold. In this case, the safe offset range for that side is set to 0, meaning the robot has no offset space on that side. If the real-time minimum safe distance is less than the preset safety response threshold, it indicates that the distance between the robot and the static safety boundary on that side is insufficient, posing a collision risk. In this case, the safe offset range for that side needs to be set to an invalid range, and a warning mechanism is triggered to remind the robot to adjust its driving state in time to ensure a safe distance from the static safety boundary.

[0092] Then, from the safe offset intervals on the left and right sides, the maximum allowable offset values ​​that satisfy the preset offset strategy are selected as the maximum allowable offsets of the robot to the left and right. The preset offset strategy is an offset selection rule formulated by combining factors such as the robot's obstacle avoidance requirements, task execution requirements, and kinematic characteristics. This strategy prioritizes ensuring the robot's obstacle avoidance safety and task execution efficiency. When selecting the maximum allowable offset value, the safe offset intervals on the left and right sides are processed separately. If the safe offset interval on one side is a valid interval and the upper limit of the interval is greater than 0, then according to the preset offset strategy, the upper limit value of the interval is directly selected as the maximum allowable offset value on that side. This upper limit value is the maximum distance that the robot can safely deviate from the fixed path on that side, ensuring that when the robot is offsetting to avoid obstacles, the key geometric constraint elements of its motion envelope will not touch the static safety boundary on the corresponding side.

[0093] If the safe offset range on one side is 0, then the maximum permissible offset value on that side is set to 0, meaning the robot cannot offset on that side. If the safe offset range on one side is invalid, then the robot is prohibited from offsetting to that side. Simultaneously, combining the generated obstacle information and decision function, the driving strategy is replanned. The maximum permissible offsets to the left and right determined in the above manner provide accurate offset distance references for the robot's obstacle avoidance decisions, ensuring that the robot can safely bypass obstacles while maximizing the successful execution of the task.

[0094] Please see Figure 8 The second aspect of this application provides a control device for an unmanned navigation robot operating on a fixed route, the device comprising: The first acquisition unit 801 is used to acquire real-time sensing data of the area ahead of a fixed path; The second acquisition unit 802 is used to perform fusion processing on the perceived data to obtain fused data; The generation unit 803 is used to generate obstacle information based on the fused data, the obstacle information including information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; The calculation unit 804 is used to calculate in real time the maximum allowable deviation of the robot from the fixed path for obstacle avoidance operation based on the robot's current position and the perception data. The third acquisition unit 805 is used to acquire the real-time task priority of the task currently being performed by the robot; The construction unit 806 is used to construct a decision function, which is a function with the perception confidence, the maximum allowable offset and the real-time task priority as joint inputs; The execution unit 807 is used to dynamically select and execute the optimal behavior strategy from the predefined behavior strategies based on the output of the decision function. The behavior strategies include at least emergency stop, maintaining the original path, and deviation.

[0095] Please see Figure 9 This application also provides a control device for an unmanned navigation robot operating on a fixed route, the device comprising: Processor 901, memory 902, input / output unit 903, bus 904; The processor 901 is connected to the memory 902, the input / output unit 903, and the bus 904; The memory 902 stores a program, and the processor 901 calls the program to execute any of the methods described above.

[0096] This application also relates to a computer-readable storage medium on which a program is stored, which, when run on a computer, causes the computer to perform any of the methods described above.

[0097] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0099] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0100] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A control method for an unmanned navigation robot operating on a fixed route, characterized in that, The method includes: Real-time acquisition of sensing data of the area ahead along a fixed path; The sensed data is fused to obtain fused data; Obstacle information is generated based on the fused data, and the obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; Based on the robot's current position and the perception data, the maximum permissible deviation of the robot from the fixed path that it can safely deviate from in order to perform obstacle avoidance operations is calculated in real time. Obtain the real-time task priority of the task currently being performed by the robot; Construct a decision function, which is a function with the perceived confidence, the maximum allowable offset, and the real-time task priority as joint inputs; Based on the output of the decision function, the optimal behavior strategy is dynamically selected and executed from the predefined behavior strategies. The behavior strategies include at least emergency stop, maintaining the original path, and deviation.

2. The unmanned navigation robot control method under a fixed route according to claim 1, characterized in that, Obstacle information is generated based on the fused data. This obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to each obstacle, including: Based on the fused data, scene representation information of the area in front of the fixed path is generated, and the scene representation information includes at least an occupied grid map, a set of object detection boxes, or a semantic segmentation map. The scene representation information is analyzed to identify at least one candidate obstacle. The perceptual confidence level of each candidate obstacle is calculated based on the matching between the fused data and the scene representation information. The individual candidate obstacle and its corresponding perception confidence are encapsulated as obstacle information.

3. The unmanned navigation robot control method under a fixed route according to claim 1, characterized in that, Based on the robot's current position and the perceived data, the maximum permissible deviation of the robot from the fixed path for obstacle avoidance operations is calculated in real time, including: Based on the robot's current position, the static safety boundaries on both sides of the fixed path at the current position are determined from real-time perception data. The static safety boundaries are determined by the positions of static obstacles on both sides of the path and a preset static safety margin. Calculate the dynamic space occupied by the robot body based on the robot's body geometry parameters and current motion state; Based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, the maximum permissible offset by which the robot can safely deviate to the left or right of the fixed path is calculated.

4. The unmanned navigation robot control method under a fixed route according to claim 1, characterized in that, Based on the output of the decision function, the optimal behavior strategy is dynamically selected and executed from predefined behavior strategies, including: The output of the decision function is mapped to the corresponding predefined behavior strategy; Based on the output of the decision function and the maximum permissible offset, a set of control parameters corresponding to the mapped behavior strategy is generated. When the mapped behavior strategy is the offset driving, the set of control parameters includes at least one target lateral offset value that is not greater than the maximum permissible offset. The offset driving strategy is to control the robot to temporarily deviate from the fixed path by a distance less than or equal to the maximum permissible offset, and return to the fixed path after passing through the area where the potential obstacle is located. Motion control commands are generated based on the set of control parameters; The robot is driven to execute the mapped behavioral strategy according to the control.

5. The unmanned navigation robot control method under a fixed route according to claim 1, characterized in that, The sensed data is fused to obtain fused data, including: Perception data from different coordinate systems and at different times are mapped to the same spatiotemporal reference system centered on the robot to form registration data; For each detected target in the registration data, feature information based on different physical principles is extracted; Based on preset fusion rules, the feature information of the same detection target is associated and weighted by confidence level to generate the fused data.

6. The unmanned navigation robot control method under a fixed route according to claim 3, characterized in that, Based on the robot's current position, determine the static safety boundaries on both sides of the fixed path at the current position from real-time perception data, including: Based on the robot's current position, the real-time sensing data is spatially rasterized along the lateral direction of the fixed path to construct a local environmental occupancy raster map. Static obstacle clustering and identification are performed on the local environment occupancy grid map to distinguish between dynamic objects and static obstacles, and the contour information of each static obstacle in the grid map is extracted. Based on the static obstacle contours on the corresponding sides of the fixed path, the preset static safety margin is extended outward in a direction parallel to the fixed path to generate corresponding left and right static safety boundary lines. The left and right static safety boundary lines are used as the two static safety boundaries of the fixed path at the current position.

7. The unmanned navigation robot control method under fixed route as described in claim 3, characterized in that, Based on the real-time relative positional relationship between the dynamically occupied space and the static safety boundaries on both sides, the maximum permissible offset by which the robot can safely deviate to the left or right of the fixed path is calculated, including: Based on the geometric projection of the dynamically occupied space onto the cross section of the fixed path, one or more key geometric constraint elements are constructed to characterize the robot's motion envelope. For each side of the static safety boundary, calculate the real-time minimum safe distance between the key geometric constraint element and the corresponding side static safety boundary when the robot is along the fixed path; Based on the relationship between the real-time minimum safe distance and the preset safety response threshold, safe offset intervals are determined for the left and right sides of the robot, respectively. From the safe offset intervals on the left and right, the maximum allowable offset value that satisfies the preset offset strategy is selected as the maximum allowable offset amount for the robot to move left and right.

8. A control device for an unmanned navigation robot operating on a fixed route, characterized in that, The device includes: The first acquisition unit is used to acquire real-time sensing data of the area ahead of a fixed path; The second acquisition unit is used to perform fusion processing on the perceived data to obtain fused data; The generation unit is used to generate obstacle information based on the fused data, wherein the obstacle information includes information about the existence of potential obstacles and the perception confidence level corresponding to the obstacles; The computing unit is used to calculate in real time the maximum allowable deviation of the robot from the fixed path for obstacle avoidance operations, based on the robot's current position and the perception data. The third acquisition unit is used to acquire the real-time task priority of the task currently being performed by the robot; A construction unit is used to construct a decision function, which is a function with the perception confidence, the maximum allowable offset, and the real-time task priority as joint inputs; The execution unit is used to dynamically select and execute the optimal behavior strategy from the predefined behavior strategies based on the output of the decision function. The behavior strategies include at least emergency stop, maintaining the original path, and deviation.

9. A control device for an unmanned navigation robot operating on a fixed route, characterized in that, The device includes: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor invokes to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the method as described in any one of claims 1 to 7.

Images