A multimodal transport path planning method based on a large and medium-sized sorting robot and a special sorting robot

Abstract

This invention provides a multimodal transport path planning method and a dedicated sorting robot based on a medium-to-large cargo sorting robot. The method includes: identifying the center of gravity offset and rotational inertia tensor of the target medium-to-large cargo based on detected data to construct a three-dimensional overturning risk field with the robot's center of mass as the origin; mapping the three-dimensional overturning risk field to the interior of a three-dimensional multimodal transport spatiotemporal network, and assigning corresponding safety margin values ​​to each spatiotemporal node by combining the equipment parameters, communication time windows, and dynamic load states of each transport node; using a spatiotemporal A* algorithm for path search, and retaining only spatiotemporal nodes with safety margin values ​​greater than a preset margin threshold as path nodes during the search process; and performing three-dimensional integration processing on each path node to generate a multimodal transport path adapted to the target medium-to-large cargo. This invention can significantly improve path planning efficiency.

Description

Technical Field

[0001] This invention relates to the field of logistics technology, and in particular to a multimodal transport path planning method based on a large and medium-sized sorting robot and a dedicated sorting robot. Background Technology

[0002] With the rapid development of e-commerce logistics and industrial logistics, the demand for sorting medium and large-sized goods (standard pallets, standard carriers, and standard cages) has exploded, making sorting automation an inevitable trend in the industry. Path planning, as the core technology of sorting robot systems, directly determines sorting efficiency and operational safety. Most existing path planning technologies originate from small-item sorting scenarios. In small-item sorting, due to the light weight and small size of the goods, their impact on the robot's motion characteristics is negligible, and path planning only needs to be designed for the robot itself to meet the requirements. However, there is a fundamental difference between medium and large-sized sorting and small-item sorting. Medium and large-sized goods are oversized, heavy, and irregularly shaped; the physical characteristics of the goods themselves directly determine the feasibility of the path. This core difference has not yet received sufficient attention from existing technologies.

[0003] Currently, the multimodal transport path planning methods for most medium and large-item sorting robots still follow the technical approach of small-item (e-commerce) sorting, generally exhibiting the defect of "heavy on robots, light on goods." Statistics show that over 80% of existing related invention patents treat goods merely as a static "rectangular bounding box," considering only its outer dimensions and nominal weight, completely ignoring the dynamic changes of goods during transportation. This simplified approach cannot accurately assess the impact of factors such as the shift in the center of gravity, inertial sway, and flexible deformation on the robot's stability, forcing the system to adopt excessively large safety margins, limiting the robot's speed and turning radius, and even, in extreme cases, causing safety accidents such as robot tipping over and goods falling and breaking, severely restricting the overall efficiency of medium and large-item sorting systems.

[0004] General path planning techniques cannot fundamentally solve the above problems because their technical system is based on the premise that "the robot itself is the only decision-making object," and lacks specific modeling and constraint mechanisms for the dynamic physical characteristics of medium and large-sized goods. Existing technologies neither achieve real-time perception and quantitative evaluation of the dynamic characteristics of goods, nor deeply integrate these characteristics with robot motion control and multi-device collaborative scheduling. Therefore, there is an urgent need for a multimodal transport path planning method and a dedicated sorting robot that can fully consider the dynamic physical characteristics of medium and large-sized goods to fill the gaps in existing technologies and improve the operational efficiency and safety of medium and large-sized sorting systems. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide a multimodal transport path planning method and a dedicated sorting robot based on a medium-to-large item sorting robot, so as to solve the problem that the existing technology for multimodal transport path planning of medium-to-large item sorting robots mostly follows the technical approach of small item sorting, without establishing a special modeling and constraint mechanism for the dynamic physical characteristics of medium-to-large items, which leads to a reduction in the safety and efficiency of path planning.

[0006] The first aspect of the present invention proposes: A multimodal transport path planning method based on a medium-to-large item sorting robot, wherein the method includes: After the sorting robot grabs the target large item, it controls the sorting robot to perform preset multi-directional excitation actions to detect the torque of each joint, the torque of the end effector, and the body posture data of the sorting robot. Based on the detected data, the center of gravity offset and rotational inertia tensor of the target large item are identified to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. A three-dimensional multimodal transport spatiotemporal network is constructed, which includes spatial coordinates, time dimension and intermodal transport node attributes. The three-dimensional overturning risk field is mapped to the interior of the three-dimensional multimodal transport spatiotemporal network. Based on the equipment parameters, communication time windows and dynamic load status of each intermodal transport node, a corresponding safety margin value is assigned to each spatiotemporal node. Starting with the initial spatiotemporal node as the initial state and ending with the target spatiotemporal node as the target state, the spatiotemporal A* algorithm is used for path search, and only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes during the search process. The path nodes are integrated in three dimensions to generate a multimodal transport path that is compatible with the large cargo in the target.

[0007] The beneficial effects of this invention are as follows: This technical solution, by performing multi-directional excitation actions after grasping medium and large-sized goods, accurately detects the robot's joint torque, end effector torque, and body posture data, thereby identifying the cargo's center of gravity offset and rotational inertia tensor and constructing a three-dimensional overturning risk field. This effectively compensates for the deficiency of existing technologies that do not specifically model the dynamic physical characteristics of medium and large-sized goods, and achieves a quantitative assessment of cargo overturning risk. By constructing a three-dimensional multimodal transport spatiotemporal network that integrates spatial coordinates, time dimensions, and intermodal node attributes, and mapping the overturning risk field into it, and combining the safety margins allocated by each node's equipment parameters, communication time windows, and dynamic load status, a deep coupling between risk factors and the spatiotemporal characteristics of multimodal transport is achieved. The spatiotemporal A* algorithm is used for path search, and only nodes that meet the safety margin are retained, ultimately generating a multimodal transport path adapted to the target cargo. This significantly improves the safety of multimodal transport path planning for medium and large-sized cargo sorting while taking into account the efficiency and rationality of path planning, avoiding overturning accidents and efficiency losses caused by differences in cargo physical characteristics.

[0008] Furthermore, the step of identifying the center of gravity offset and moment of inertia tensor of the large cargo in the target based on the detected data, in order to construct a three-dimensional overturning risk field with the robot's center of mass as the origin, includes: Based on the joint torques, the end effector torques, and the body posture data, the actual ground loads of the robot's four wheels are obtained by inverting the static equilibrium equations. The load data of each wheel under different excitation actions are collected, and the wheel load data are processed by a preset linear algorithm to calculate the center of gravity offset and the moment of inertia tensor. A corresponding attitude space is constructed based on the linear acceleration, angular velocity, and heading angle of the sorting robot. An adaptive sampling algorithm is used to calculate the load transfer rate of the four wheels for each sampled attitude point in the attitude space, and the critical attitude point of the load transfer rate is determined accordingly. The moving least squares method is used to fit each of the critical attitude points to generate a continuous three-dimensional critical overturning surface, and the three-dimensional overturning risk field is constructed according to the attitude space and the three-dimensional critical overturning curve.

[0009] Furthermore, the step of constructing the three-dimensional overturning risk field based on the attitude space and the three-dimensional critical overturning curve includes: According to the preset time window, the load transfer rate time series of the four wheels of the sorting robot are detected, and based on Miner's linear cumulative damage theory, the cumulative damage factor of each wheel is calculated according to the load transfer rate time series. Based on the calculated cumulative damage factor of each wheel, the equivalent stiffness coefficient of the sorting robot is updated accordingly. Based on the updated equivalent stiffness coefficient, the three-dimensional critical overturning curve is dynamically corrected to obtain the intermediate three-dimensional critical overturning curve. The attitude space is divided into voxel grids of a preset resolution. The Euclidean distance between the attitude point corresponding to each voxel grid and the intermediate three-dimensional critical overturning curve is calculated. The Euclidean distance is multiplied by the cumulative damage factor of the corresponding attitude point to obtain the risk field strength value of each voxel grid. Each risk field strength value is then inserted into the interior of the intermediate three-dimensional critical overturning curve to generate the three-dimensional overturning risk field.

[0010] Furthermore, the step of mapping the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network, and allocating corresponding safety margin values ​​to each spatiotemporal node by combining the equipment parameters, communication time windows, and dynamic load status of each intermodal transport node, includes: The independent component analysis algorithm is used to perform modal decomposition on the three-dimensional overturning risk field to obtain a feature risk field containing several basic motion modes, and the modal feature parameters of each feature risk field are detected accordingly to construct a corresponding motion mode-risk feature correspondence library. The three-dimensional multimodal transport spatiotemporal network is divided into several modal segments, and the target feature risk field corresponding to each modal segment is matched in the motion modality-risk feature correspondence library. Several target spatiotemporal nodes contained within the target feature risk field are detected, and the safety margin value of each target spatiotemporal node is calculated according to the device parameters, the communication time window, and the dynamic load status.

[0011] Furthermore, the step of calculating the safety margin value of each target spatiotemporal node based on the device parameters, the communication time window, and the dynamic load state includes: The device parameters, the communication time window, and the dynamic load state are converted into corresponding constraints, and based on the constraints, the state sequence of each constraint within the future preset time window is predicted by the Kalman filter algorithm. Based on the historical operation data of the sorting robot, the propagation coefficient matrix of each constraint condition between different working modes is calculated, and the total influence coefficient of each constraint condition on the target spatiotemporal node is calculated according to the propagation coefficient matrix. The state sequence is converted into a corresponding risk value sequence, and the corresponding risk entropy is calculated based on the risk value sequence. The risk entropy of each constraint is weighted and summed based on the total influence coefficient to calculate the safety margin value of each target spatiotemporal node.

[0012] Furthermore, the step of using the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state to perform path search using the spatiotemporal A* algorithm includes: During the path search process, the safety margin value of the parent node and the safety margin value of the child node are weighted and averaged to obtain the cumulative safety margin value of each child node. When the cumulative safety margin value of a child node is higher than the preset margin threshold, and the search reaches the termination spatiotemporal node, an initial path is generated by backtracking from the termination spatiotemporal node, and each initial node in the initial path is identified. The actual arrival time of each initial node is detected to be within its communication time window. For initial nodes that are outside the communication time window, a local re-search is performed within a preset time and space range with the current initial node as the center to generate an alternative path segment that meets the communication time window. The alternative path segment is then merged with the initial path to find the optimal path.

[0013] Furthermore, the step of performing three-dimensional integration processing on each of the path nodes to generate a multimodal transport route adapted to the large cargo in the target includes: Discrete path nodes are used as anchor points of the spatiotemporal corridor, and an initial spatiotemporal corridor is generated by expanding in the spatial and temporal dimensions with each anchor point as the center. The initial spatiotemporal corridor is then checked for safety margin in order to generate the corresponding target spatiotemporal corridor. The spatiotemporal corridor information of other sorting robots in the current multimodal transport scenario is detected, so as to determine whether the target spatiotemporal corridor overlaps with the spatiotemporal corridors of other sorting robots. If so, for each overlapping area, the time interval and spatial range of the conflict are calculated to adjust the target spatiotemporal corridor accordingly, and the adjusted target spatiotemporal corridor is converted into the multimodal transport path.

[0014] The second aspect of the present invention proposes: A multimodal transport path planning system based on a medium-to-large item sorting robot, wherein the system includes: The detection module is used to control the sorting robot to perform preset multi-directional excitation actions after the sorting robot grabs the target large item, so as to detect the torque of each joint, the torque of the end effector and the body posture data of the sorting robot, and identify the center of gravity offset and rotational inertia tensor of the target large item based on the detected data, so as to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. The construction module is used to construct a three-dimensional multimodal transport spatiotemporal network that includes spatial coordinates, time dimension and intermodal transport node attributes, and to map the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network. Combining the equipment parameters, communication time windows and dynamic load status of each intermodal transport node, the module allocates corresponding safety margin values ​​to each spatiotemporal node. The search module is used to perform path search using the spatiotemporal A* algorithm with the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state. During the search process, only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes. The integration module is used to perform three-dimensional integration processing on each of the path nodes to generate a multimodal transport route that is compatible with the large cargo in the target.

[0015] Furthermore, the detection module is specifically used for: Based on the joint torques, the end effector torques, and the body posture data, the actual ground loads of the robot's four wheels are obtained by inverting the static equilibrium equations. The load data of each wheel under different excitation actions are collected, and the wheel load data are processed by a preset linear algorithm to calculate the center of gravity offset and the moment of inertia tensor. A corresponding attitude space is constructed based on the linear acceleration, angular velocity, and heading angle of the sorting robot. An adaptive sampling algorithm is used to calculate the load transfer rate of the four wheels for each sampled attitude point in the attitude space, and the critical attitude point of the load transfer rate is determined accordingly. The moving least squares method is used to fit each of the critical attitude points to generate a continuous three-dimensional critical overturning surface, and the three-dimensional overturning risk field is constructed according to the attitude space and the three-dimensional critical overturning curve.

[0016] Furthermore, the detection module is specifically used for: According to the preset time window, the load transfer rate time series of the four wheels of the sorting robot are detected, and based on Miner's linear cumulative damage theory, the cumulative damage factor of each wheel is calculated according to the load transfer rate time series. Based on the calculated cumulative damage factor of each wheel, the equivalent stiffness coefficient of the sorting robot is updated accordingly. Based on the updated equivalent stiffness coefficient, the three-dimensional critical overturning curve is dynamically corrected to obtain the intermediate three-dimensional critical overturning curve. The attitude space is divided into voxel grids of a preset resolution. The Euclidean distance between the attitude point corresponding to each voxel grid and the intermediate three-dimensional critical overturning curve is calculated. The Euclidean distance is multiplied by the cumulative damage factor of the corresponding attitude point to obtain the risk field strength value of each voxel grid. Each risk field strength value is then inserted into the interior of the intermediate three-dimensional critical overturning curve to generate the three-dimensional overturning risk field.

[0017] Furthermore, the building module is specifically used for: The independent component analysis algorithm is used to perform modal decomposition on the three-dimensional overturning risk field to obtain a feature risk field containing several basic motion modes, and the modal feature parameters of each feature risk field are detected accordingly to construct a corresponding motion mode-risk feature correspondence library. The three-dimensional multimodal transport spatiotemporal network is divided into several modal segments, and the target feature risk field corresponding to each modal segment is matched in the motion modality-risk feature correspondence library. Several target spatiotemporal nodes contained within the target feature risk field are detected, and the safety margin value of each target spatiotemporal node is calculated according to the device parameters, the communication time window, and the dynamic load status.

[0018] Furthermore, the building module is specifically used for: The device parameters, the communication time window, and the dynamic load state are converted into corresponding constraints, and based on the constraints, the state sequence of each constraint within the future preset time window is predicted by the Kalman filter algorithm. Based on the historical operation data of the sorting robot, the propagation coefficient matrix of each constraint condition between different working modes is calculated, and the total influence coefficient of each constraint condition on the target spatiotemporal node is calculated according to the propagation coefficient matrix. The state sequence is converted into a corresponding risk value sequence, and the corresponding risk entropy is calculated based on the risk value sequence. The risk entropy of each constraint is weighted and summed based on the total influence coefficient to calculate the safety margin value of each target spatiotemporal node.

[0019] Furthermore, the search module is specifically used for: During the path search process, the safety margin value of the parent node and the safety margin value of the child node are weighted and averaged to obtain the cumulative safety margin value of each child node. When the cumulative safety margin value of a child node is higher than the preset margin threshold, and the search reaches the termination spatiotemporal node, an initial path is generated by backtracking from the termination spatiotemporal node, and each initial node in the initial path is identified. The actual arrival time of each initial node is detected to be within its communication time window. For initial nodes that are outside the communication time window, a local re-search is performed within a preset time and space range with the current initial node as the center to generate an alternative path segment that meets the communication time window. The alternative path segment is then merged with the initial path to find the optimal path.

[0020] Furthermore, the integration module is specifically used for: Discrete path nodes are used as anchor points of the spatiotemporal corridor, and an initial spatiotemporal corridor is generated by expanding in the spatial and temporal dimensions with each anchor point as the center. The initial spatiotemporal corridor is then checked for safety margin in order to generate the corresponding target spatiotemporal corridor. The spatiotemporal corridor information of other sorting robots in the current multimodal transport scenario is detected, so as to determine whether the target spatiotemporal corridor overlaps with the spatiotemporal corridors of other sorting robots. If so, for each overlapping area, the time interval and spatial range of the conflict are calculated to adjust the target spatiotemporal corridor accordingly, and the adjusted target spatiotemporal corridor is converted into the multimodal transport path.

[0021] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the multimodal transport path planning method based on a medium-to-large item sorting robot as described above.

[0022] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the multimodal transport path planning method based on a medium-to-large item sorting robot as described above.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] Figure 1A flowchart of a multimodal transport path planning method based on a large and medium-sized sorting robot provided in the first embodiment of the present invention; Figure 2 This is a structural block diagram of a multimodal transport path planning system based on a large and medium-sized sorting robot, provided in the third embodiment of the present invention.

[0025] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0026] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0027] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] Please see Figure 1 The image shows a multimodal transport path planning method based on a medium-to-large-sized sorting robot provided in the first embodiment of the present invention. This method can construct a three-dimensional multimodal transport spatiotemporal network that integrates spatial coordinates, time dimension, and intermodal node attributes, and map the overturning risk field into it. By combining the equipment parameters of each node, communication time window, and dynamic load status to allocate safety margin, it achieves deep coupling between risk factors and the spatiotemporal characteristics of multimodal transport. The spatiotemporal A* algorithm is used for path search and only nodes with sufficient safety margin are retained. Finally, a multimodal transport path adapted to the target goods is generated. This significantly improves the safety of multimodal transport path planning for medium-to-large-sized sorting while taking into account the efficiency and rationality of path planning, and avoids overturning accidents and efficiency losses caused by differences in the physical characteristics of goods.

[0030] Specifically, this embodiment provides: A multimodal transport path planning method based on a medium-to-large item sorting robot, wherein the method includes: Step S10: After the sorting robot grabs the large item in the target, the sorting robot is controlled to perform a preset multi-directional excitation action to detect the torque of each joint, the torque of the end effector and the body posture data of the sorting robot. Based on the detected data, the center of gravity offset and rotational inertia tensor of the large item in the target are identified to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. It's important to note that the irregular packaging and uneven distribution of internal contents of medium and large-sized goods result in significant uncertainties in their center of gravity and moment of inertia. These two parameters are core factors determining the robot's tipping risk. Traditional methods rely on manual input of cargo parameters or the use of fixed safety factors, which cannot adapt to the characteristics of different goods. This can easily lead to inefficiency due to excessively high safety factors or tipping due to insufficient safety factors. This step controls the robot to execute preset multi-directional excitation actions (such as slow acceleration and deceleration forward and backward, small-angle turns left and right, and rotation in place). Utilizing the robot's built-in torque sensors and inertial measurement unit (IMU), it collects real-time data on joint torques, end effector torques, and body posture. Through static and dynamic inversion algorithms, it accurately calculates the cargo's three-dimensional center of gravity offset and moment of inertia tensor. Based on these parameters, it constructs a three-dimensional tipping risk field, quantifying the robot's tipping risk level in any posture. This online identification method accurately assesses the tipping risk without prior knowledge of any cargo parameters, providing a safe basis for subsequent path planning.

[0031] Step S20: Construct a three-dimensional multimodal transport spatiotemporal network that includes spatial coordinates, time dimension and intermodal node attributes, and map the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network. Combine the equipment parameters, communication time windows and dynamic load status of each intermodal node, and assign corresponding safety margin values ​​to each spatiotemporal node. It's important to note that traditional path planning only considers two-dimensional spatial coordinates, neglecting the constraints of the time dimension and node attributes. However, in multimodal transport scenarios, each transport node (such as conveyor belt connections, elevator entrances, and rack locations) has strict communication time windows (receiving goods only within specific time periods) and dynamic load limitations (the node's processing capacity changes over time). Furthermore, different path segments correspond to different robot motion postures, resulting in varying tipping risks. This step constructs a three-dimensional multimodal transport spatiotemporal network incorporating spatial coordinates (x, y, z), the time dimension (t), and transport node attributes, abstracting each possible operational location and time point as a spatiotemporal node. The previously constructed three-dimensional tipping risk field is mapped into the spatiotemporal network, assigning each spatiotemporal node a tipping risk value. Simultaneously, by combining the node's equipment parameters (such as load-bearing capacity, aisle width, and turning radius), communication time windows, and dynamic load states, a comprehensive safety margin value is assigned to each spatiotemporal node. A higher safety margin value indicates a lower operational risk for that spatiotemporal node, making it more suitable as a path node.

[0032] Step S30: Using the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state, the spatiotemporal A* algorithm is used to search for the path, and during the search process, only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes. It's important to note that the traditional A* algorithm searches for the shortest path only in two-dimensional space, failing to handle time and safety constraints. This step extends the traditional A* algorithm by incorporating a time dimension and safety constraints, creating a spatiotemporal A* algorithm. During the search, the algorithm not only calculates the path length and time cost but also evaluates the safety margin of each candidate spatiotemporal node in real time. Only nodes with a safety margin greater than a preset threshold are retained as path nodes, fundamentally eliminating paths with overturning risks or node constraint conflicts. This safety-first search strategy ensures that the generated path meets safety requirements at all times, preventing overturning accidents and node conflicts.

[0033] Step S40: Perform three-dimensional integration processing on each of the path nodes to generate a multimodal transport path that is compatible with the large cargo in the target.

[0034] It's important to note that discrete path nodes only contain key spatiotemporal location information and cannot be directly executed by the robot. This step integrates the searched path nodes in three dimensions according to time sequence and spatial location, generating a complete multimodal transport path that includes robot motion trajectory, speed planning, node operation time, and equipment switching instructions. Simultaneously, considering the robot's kinematic constraints, the path is smoothed to ensure the robot can execute the path smoothly and continuously. The final generated path fully adapts to the tipping risk characteristics of large goods and the spatiotemporal constraints of multimodal transport, enabling safe and efficient completion of cargo sorting and transfer tasks.

[0035] Second Embodiment Furthermore, the step of identifying the center of gravity offset and moment of inertia tensor of the large cargo in the target based on the detected data, in order to construct a three-dimensional overturning risk field with the robot's center of mass as the origin, includes: Based on the joint torques, the end effector torques, and the body posture data, the actual ground loads of the robot's four wheels are obtained by inverting the static equilibrium equations. The load data of each wheel under different excitation actions are collected, and the wheel load data are processed by a preset linear algorithm to calculate the center of gravity offset and the moment of inertia tensor. A corresponding attitude space is constructed based on the linear acceleration, angular velocity, and heading angle of the sorting robot. An adaptive sampling algorithm is used to calculate the load transfer rate of the four wheels for each sampled attitude point in the attitude space, and the critical attitude point of the load transfer rate is determined accordingly. The moving least squares method is used to fit each of the critical attitude points to generate a continuous three-dimensional critical overturning surface, and the three-dimensional overturning risk field is constructed according to the attitude space and the three-dimensional critical overturning curve.

[0036] It's important to note that after the robot grasps the goods, the center of gravity of the entire robot-cargo system shifts, causing a redistribution of the ground load on the four wheels. This step, based on static equilibrium equations, uses collected joint torques, end moments, and body posture data to deduce the actual ground load on each wheel. Then, the robot is controlled to perform multiple excitation actions in different directions, collecting wheel load data under different postures. By establishing a system of linear equations between wheel loads, center of gravity position, and moment of inertia, the three-dimensional center of gravity offset of the goods (offset relative to the robot's center of mass in the x, y, and z directions) and the moment of inertia tensor (containing three moment of inertia components and three inertia products) are solved to obtain the goods' three-dimensional center of gravity offset (offset in the x, y, and z directions relative to the robot's center of mass). This identification method requires no additional sensors, utilizing only the robot's built-in torque and posture sensors. It is low-cost, highly accurate, and can complete identification within seconds of goods grasping, without affecting sorting efficiency.

[0037] Load transfer ratio (LTR) is a classic indicator for assessing the tipping risk of wheeled robots. It is defined as the ratio of the load difference between the left and right wheels to the total load (or the front and rear wheels), and its value ranges from -1 to 1. When the absolute value of the load transfer ratio is 1, it indicates that the ground load of one wheel is 0, the wheel is off the ground, and the robot is about to tip over. The corresponding posture at this time is the critical tipping posture. This step constructs a three-dimensional posture space based on the robot's motion parameters (linear acceleration, angular velocity, and heading angle). Each point in the posture space corresponds to a motion posture of the robot. An adaptive sampling algorithm is used to sample the posture space, increasing the sampling density in high-risk posture areas (such as large acceleration and large angular velocity) and decreasing the sampling density in low-risk areas, thus reducing the amount of computation while ensuring accuracy. The load transfer ratio of the four wheels corresponding to each sampled posture point is calculated. When the absolute value of the load transfer ratio of a certain posture point reaches 1, it is marked as a critical posture point.

[0038] Discrete critical attitude points can only assess the rollover risk of a specific attitude and cannot cover all possible motion attitudes. Moving least squares (LMS) is a high-precision surface fitting algorithm that can optimally fit discrete points within a local region, generating a continuous and smooth surface. This step uses LMS to fit all critical attitude points, generating a three-dimensional critical rollover surface. This surface divides the attitude space into two regions: the inner side of the surface is the safe region, and the outer side is the rollover region. The distance from any attitude point to the critical rollover surface represents the safety margin of that attitude; the closer the distance, the higher the risk. Based on the critical rollover surface and the attitude space, a three-dimensional rollover risk field is finally constructed. Each voxel in the risk field corresponds to a risk value, which can accurately quantify the rollover risk of the robot under any motion attitude.

[0039] Furthermore, the step of constructing the three-dimensional overturning risk field based on the attitude space and the three-dimensional critical overturning curve includes: According to the preset time window, the load transfer rate time series of the four wheels of the sorting robot are detected, and based on Miner's linear cumulative damage theory, the cumulative damage factor of each wheel is calculated according to the load transfer rate time series. Based on the calculated cumulative damage factor of each wheel, the equivalent stiffness coefficient of the sorting robot is updated accordingly. Based on the updated equivalent stiffness coefficient, the three-dimensional critical overturning curve is dynamically corrected to obtain the intermediate three-dimensional critical overturning curve. The attitude space is divided into voxel grids of a preset resolution. The Euclidean distance between the attitude point corresponding to each voxel grid and the intermediate three-dimensional critical overturning curve is calculated. The Euclidean distance is multiplied by the cumulative damage factor of the corresponding attitude point to obtain the risk field strength value of each voxel grid. Each risk field strength value is then inserted into the interior of the intermediate three-dimensional critical overturning curve to generate the three-dimensional overturning risk field.

[0040] It is important to note that during long-term operation, the repeated acceleration, deceleration, turning, and load transfer of the sorting robot can cause fatigue damage to the wheels, suspension, and frame, leading to a decrease in the equivalent stiffness of the structure. This decrease in equivalent stiffness results in a larger tilt angle for the robot under the same load and posture, altering the critical overturning posture. If the initial critical overturning surface is still used for risk assessment, the safety margin may be overestimated. Miner's linear cumulative damage theory is a classic theory for fatigue damage assessment, assuming that fatigue damage to materials is linearly cumulative; fatigue failure occurs when the cumulative damage reaches 1. This step collects the load transfer rate time series of four wheels according to a preset time window (e.g., daily); based on Miner's theory, the damage amount of each wheel under different load transfer rates is calculated and accumulated to obtain the cumulative damage factor for each wheel. The larger the cumulative damage factor, the more severe the structural fatigue damage of the wheel, and the greater the decrease in equivalent stiffness.

[0041] This step establishes a mapping relationship between the cumulative damage factor and the equivalent stiffness coefficient beforehand. This mapping relationship is obtained through structural mechanics simulation and durability testing of the robot. Based on the calculated cumulative damage factor, the equivalent stiffness coefficients of the four wheels and suspension are updated. The updated equivalent stiffness coefficients are then substituted into the static equilibrium equation to recalculate the critical attitude point. Finally, the moving least squares method is used to fit and obtain the corrected intermediate three-dimensional critical overturning curve. The corrected critical overturning surface can accurately reflect the overturning critical value under the current structural state of the robot, avoiding risk assessment deviations caused by structural wear.

[0042] This step divides the 3D attitude space into a voxel mesh of preset resolution, with each voxel corresponding to a tiny attitude region. The Euclidean distance from the center attitude point of each voxel to the central 3D critical overturning surface is calculated; a smaller distance indicates a higher risk. Then, the Euclidean distance is multiplied by the cumulative damage factor corresponding to that attitude point to obtain the risk field strength value for that voxel. The cumulative damage factor, used as a weight, amplifies the risk value in areas of severe structural damage, ensuring that the risk field considers not only the dynamic characteristics of the current cargo but also the structural health of the robot itself. The resulting 3D overturning risk field is a dynamically updated risk field that adjusts in real time according to changes in the robot's structural damage and cargo characteristics, always maintaining the highest risk assessment accuracy.

[0043] Furthermore, the step of mapping the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network, and allocating corresponding safety margin values ​​to each spatiotemporal node by combining the equipment parameters, communication time windows, and dynamic load status of each intermodal transport node, includes: The independent component analysis algorithm is used to perform modal decomposition on the three-dimensional overturning risk field to obtain a feature risk field containing several basic motion modes, and the modal feature parameters of each feature risk field are detected accordingly to construct a corresponding motion mode-risk feature correspondence library. The three-dimensional multimodal transport spatiotemporal network is divided into several modal segments, and the target feature risk field corresponding to each modal segment is matched in the motion modality-risk feature correspondence library. Several target spatiotemporal nodes contained within the target feature risk field are detected, and the safety margin value of each target spatiotemporal node is calculated according to the device parameters, the communication time window, and the dynamic load status.

[0044] It should be noted that the robot's motion in multimodal transport scenarios consists of a series of basic motion modes, such as linear uniform motion, linear acceleration / deceleration motion, fixed-radius turning motion, variable-radius turning motion, and uphill / downhill motion. Different motion modes correspond to different overturning risk characteristics: linear acceleration / deceleration motion is mainly affected by the forward / backward shift of the center of gravity, while turning motion is mainly affected by the left / right shift of the center of gravity and rotational inertia. Independent Component Analysis (ICA) is a blind source separation algorithm that can separate independent source signals from complex mixed signals. This step uses the ICA algorithm to perform modal decomposition on the three-dimensional overturning risk field, obtaining several characteristic risk fields corresponding to several independent basic motion modes; each characteristic risk field corresponds to a basic motion mode, containing the risk distribution characteristics under that mode; modal feature parameters (such as risk peak position, risk distribution range, and risk attenuation coefficient) are extracted for each characteristic risk field, constructing a motion mode-risk feature correspondence library to provide a basis for subsequent spatiotemporal network matching.

[0045] Multimodal transport paths consist of different motion segments, such as straight sections from the rack to the conveyor belt, turning sections at the conveyor belt junctions, and uphill and downhill sections of the elevators. Each segment corresponds to a basic motion mode. This step divides the spatiotemporal network into several modal segments based on its spatial geometric characteristics and operational requirements, with each segment corresponding to a basic motion mode. Then, the target feature risk field corresponding to that modal segment is matched against the motion mode-risk feature mapping library. This segmented matching method decomposes the complex global risk field mapping into multiple simple local risk field mappings, significantly reducing computational complexity while improving the accuracy of risk mapping.

[0046] This step first extracts the overturning risk value of all spatiotemporal nodes within each modal segment from the matched target feature risk field. Then, it combines the equipment parameters of each intermodal node within the segment (such as whether the channel width meets the cargo size and whether the node load-bearing capacity meets the cargo weight), communication time window (whether the node is within the operable time), and dynamic load status (the current queue length and processing capacity of the node) to convert these factors into corresponding constraint risk values. Finally, the overturning risk value and the constraint risk value are weighted and summed to obtain the comprehensive safety margin value of each spatiotemporal node. The safety margin value ranges from [0,1], with a larger value indicating a lower comprehensive risk and making the node more suitable as a path node. In this way, the safety margin value of each spatiotemporal node comprehensively considers cargo overturning risk, equipment constraints, time constraints, and load constraints, and can fully reflect the operational safety of the node.

[0047] Furthermore, the step of calculating the safety margin value of each target spatiotemporal node based on the device parameters, the communication time window, and the dynamic load state includes: The device parameters, the communication time window, and the dynamic load state are converted into corresponding constraints, and based on the constraints, the state sequence of each constraint within the future preset time window is predicted by the Kalman filter algorithm. Based on the historical operation data of the sorting robot, the propagation coefficient matrix of each constraint condition between different working modes is calculated, and the total influence coefficient of each constraint condition on the target spatiotemporal node is calculated according to the propagation coefficient matrix. The state sequence is converted into a corresponding risk value sequence, and the corresponding risk entropy is calculated based on the risk value sequence. The risk entropy of each constraint is weighted and summed based on the total influence coefficient to calculate the safety margin value of each target spatiotemporal node.

[0048] It's important to note that constraints in multimodal transport scenarios are dynamically changing: node loads change with the arrival of upstream goods, communication time windows adjust due to equipment failures or operational delays, and equipment parameters change due to wear and tear. Traditional methods calculate safety margins using the current constraint state, failing to account for future constraint changes. This can easily lead to path failure after path planning is completed, as node constraints have already changed. The Kalman filter algorithm is an efficient linear system state prediction algorithm capable of predicting future system states based on historical observation data. This step converts equipment parameters, communication time windows, and dynamic load states into three linear constraints. Using historical observation data for each constraint, the Kalman filter algorithm predicts the state sequence of each constraint within a future preset time window (e.g., the next 5 minutes), obtaining the predicted value and prediction error for each constraint at each future moment.

[0049] The constraints are not independent but interconnected. For example, a communication time window delay increases the dynamic load on a node, leading to increased equipment wear and tear, which in turn affects equipment parameters. This step uses a large amount of historical operational data from the sorting robot and employs correlation analysis to calculate the propagation coefficient matrix between the three constraints. Each element of the propagation coefficient matrix represents the change in one constraint caused by a unit change in another. Then, combining the location of the target spatiotemporal node with the job type, the total influence coefficient of each constraint on that node is calculated. A larger total influence coefficient indicates a greater impact of the change in that constraint on the node's operational safety.

[0050] Risk entropy is an indicator used in information theory to measure uncertainty. A higher entropy value indicates greater uncertainty and risk. This step first converts the predicted state sequence for each constraint into a risk value sequence; a higher risk value indicates the constraint is closer to the critical value. Then, the risk entropy of each constraint is calculated based on the risk value sequence to quantify its uncertainty. Finally, the risk entropy of each constraint is multiplied by the corresponding total influence coefficient, and a weighted sum is performed to obtain the comprehensive risk value for that spatiotemporal node. Subtracting the comprehensive risk value from 1 yields the safety margin value for that node. This weighted fusion method based on risk entropy considers not only the predicted values ​​of constraints but also their uncertainty, enabling a more comprehensive and accurate assessment of operational safety at spatiotemporal nodes.

[0051] Furthermore, the step of using the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state to perform path search using the spatiotemporal A* algorithm includes: During the path search process, the safety margin value of the parent node and the safety margin value of the child node are weighted and averaged to obtain the cumulative safety margin value of each child node. When the cumulative safety margin value of a child node is higher than the preset margin threshold, and the search reaches the termination spatiotemporal node, an initial path is generated by backtracking from the termination spatiotemporal node, and each initial node in the initial path is identified. The actual arrival time of each initial node is detected to be within its communication time window. For initial nodes that are outside the communication time window, a local re-search is performed within a preset time and space range with the current initial node as the center to generate an alternative path segment that meets the communication time window. The alternative path segment is then merged with the initial path to find the optimal path.

[0052] It's important to note that traditional methods only evaluate the safety margin of individual nodes, retaining a node as long as it meets a threshold requirement. However, this can lead to a low average safety margin for the entire path and a high overall risk. This step introduces the concept of cumulative safety margin during the spatiotemporal A* algorithm search process: for each candidate child node, a weighted average is calculated between the cumulative safety margin of its parent node and the child node's safety margin. The parent node's weight is dynamically adjusted based on the path length; the farther away from the starting node, the smaller the parent node's weight. This yields the cumulative safety margin value for the child node. The cumulative safety margin value reflects the average safety level of the entire path from the starting node to the child node. Only when the cumulative safety margin value exceeds a preset threshold is the child node retained as a path node. This evaluation method ensures that not only are each node in the generated path safe, but the overall safety level of the entire path also meets the requirements.

[0053] When the spacetime A* algorithm reaches the terminating spacetime node and the cumulative safety margin of that node meets the requirements, a feasible path has been found. This step starts from the terminating spacetime node and backtracks along the parent node pointers of each node until the starting spacetime node is reached. All nodes traversed during the backtracking process are arranged in chronological order to generate the initial path. The initial path is a feasible path that satisfies both safety and spacetime constraints, but it may have local time window conflicts, requiring further optimization.

[0054] Because Kalman filtering predictions have inherent errors, the actual arrival times of some nodes in the initial path may exceed their communication time windows, making it impossible to perform operations at those nodes. This step first checks whether the actual arrival time of each node in the initial path is within its communication time window. For nodes exceeding the time window, a local re-search is performed within a preset spatiotemporal range (e.g., 1 minute before and after, and 5 meters around the node) centered on that node to search for alternative path segments that satisfy the time window constraints. Then, the alternative path segments are merged with the rest of the initial path to generate the final optimal path. This local re-search method only re-searches conflict areas, preserving most of the content of the initial path and significantly improving the efficiency of path optimization.

[0055] Furthermore, the step of performing three-dimensional integration processing on each of the path nodes to generate a multimodal transport route adapted to the large cargo in the target includes: Discrete path nodes are used as anchor points of the spatiotemporal corridor, and an initial spatiotemporal corridor is generated by expanding in the spatial and temporal dimensions with each anchor point as the center. The initial spatiotemporal corridor is then checked for safety margin in order to generate the corresponding target spatiotemporal corridor. The spatiotemporal corridor information of other sorting robots in the current multimodal transport scenario is detected, so as to determine whether the target spatiotemporal corridor overlaps with the spatiotemporal corridors of other sorting robots. If so, for each overlapping area, the time interval and spatial range of the conflict are calculated to adjust the target spatiotemporal corridor accordingly, and the adjusted target spatiotemporal corridor is converted into the multimodal transport path.

[0056] It's important to note that while discrete path nodes are precise spatiotemporal points, the robot's actual motion contains inherent errors (such as positioning and speed control errors). Strictly adhering to discrete points can easily lead to deviations. A spatiotemporal corridor is a continuous range in both spatial and temporal dimensions. If the robot traverses the spatial range within the temporal range, it is considered to have met the path requirements. This step uses the obtained discrete path nodes as anchor points for the spatiotemporal corridor. Centered on each anchor point, a certain spatial range is expanded based on the robot's size and positioning error, and a certain temporal range is expanded based on the robot's speed control error, generating an initial spatiotemporal corridor. Then, a safety margin check is performed on the initial spatiotemporal corridor to ensure that the safety margin value of all spatiotemporal points within the corridor is greater than a preset threshold. For areas with insufficient safety margin, the corridor range is reduced or the anchor point positions are adjusted, ultimately generating a target spatiotemporal corridor that meets the safety requirements. The spatiotemporal corridor provides a certain tolerance for the robot's motion, effectively addressing various errors during movement and improving the path's robustness.

[0057] In scenarios involving multi-robot collaborative operations, multiple robots moving simultaneously within the same area are prone to collisions. Traditional conflict detection methods, based on discrete pathpoints, are susceptible to missed or false detections. This step utilizes the workshop's central scheduling system to acquire real-time spatiotemporal corridor information for all sorting robots. It then performs dual spatial and temporal overlap detection on the target spatiotemporal corridor and the corridors of other robots: if two spatiotemporal corridors overlap spatially and their temporal intervals also intersect, it indicates that the two robots will collide in that area. This spatiotemporal corridor-based conflict detection method can accurately and comprehensively detect all possible conflicts, providing a basis for subsequent conflict avoidance.

[0058] When a conflict is detected, this step first calculates the specific time interval and spatial range of the conflict. Then, it adjusts the target spatiotemporal corridor using either a time-priority or space-priority strategy: the time-priority strategy adjusts the robot's passage time through the conflict area to avoid the passage time of other robots; the space-priority strategy adjusts the robot's path to bypass the conflict area. After adjustment, the spatiotemporal corridor undergoes a safety margin check to ensure that the adjusted corridor still meets safety requirements. Finally, the adjusted target spatiotemporal corridor is converted into a multimodal transport path executable by the robot, including detailed motion trajectories, speed planning, node operation times, and equipment switching instructions. This spatiotemporal corridor-based conflict avoidance method effectively resolves collisions between multiple robots without affecting overall operational efficiency, enabling collaborative operation among multiple robots.

[0059] Please see Figure 2 The third embodiment of the present invention provides: A multimodal transport path planning system based on a medium-to-large item sorting robot, wherein the system includes: The detection module is used to control the sorting robot to perform preset multi-directional excitation actions after the sorting robot grabs the target large item, so as to detect the torque of each joint, the torque of the end effector and the body posture data of the sorting robot, and identify the center of gravity offset and rotational inertia tensor of the target large item based on the detected data, so as to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. The construction module is used to construct a three-dimensional multimodal transport spatiotemporal network that includes spatial coordinates, time dimension and intermodal transport node attributes, and to map the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network. Combining the equipment parameters, communication time windows and dynamic load status of each intermodal transport node, the module allocates corresponding safety margin values ​​to each spatiotemporal node. The search module is used to perform path search using the spatiotemporal A* algorithm with the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state. During the search process, only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes. The integration module is used to perform three-dimensional integration processing on each of the path nodes to generate a multimodal transport route that is compatible with the large cargo in the target.

[0060] Furthermore, the detection module is specifically used for: Based on the joint torques, the end effector torques, and the body posture data, the actual ground loads of the robot's four wheels are obtained by inverting the static equilibrium equations. The load data of each wheel under different excitation actions are collected, and the wheel load data are processed by a preset linear algorithm to calculate the center of gravity offset and the moment of inertia tensor. A corresponding attitude space is constructed based on the linear acceleration, angular velocity, and heading angle of the sorting robot. An adaptive sampling algorithm is used to calculate the load transfer rate of the four wheels for each sampled attitude point in the attitude space, and the critical attitude point of the load transfer rate is determined accordingly. The moving least squares method is used to fit each of the critical attitude points to generate a continuous three-dimensional critical overturning surface, and the three-dimensional overturning risk field is constructed according to the attitude space and the three-dimensional critical overturning curve.

[0061] Furthermore, the detection module is specifically used for: According to the preset time window, the load transfer rate time series of the four wheels of the sorting robot are detected, and based on Miner's linear cumulative damage theory, the cumulative damage factor of each wheel is calculated according to the load transfer rate time series. Based on the calculated cumulative damage factor of each wheel, the equivalent stiffness coefficient of the sorting robot is updated accordingly. Based on the updated equivalent stiffness coefficient, the three-dimensional critical overturning curve is dynamically corrected to obtain the intermediate three-dimensional critical overturning curve. The attitude space is divided into voxel grids of a preset resolution. The Euclidean distance between the attitude point corresponding to each voxel grid and the intermediate three-dimensional critical overturning curve is calculated. The Euclidean distance is multiplied by the cumulative damage factor of the corresponding attitude point to obtain the risk field strength value of each voxel grid. Each risk field strength value is then inserted into the interior of the intermediate three-dimensional critical overturning curve to generate the three-dimensional overturning risk field.

[0062] Furthermore, the building module is specifically used for: The independent component analysis algorithm is used to perform modal decomposition on the three-dimensional overturning risk field to obtain a feature risk field containing several basic motion modes, and the modal feature parameters of each feature risk field are detected accordingly to construct a corresponding motion mode-risk feature correspondence library. The three-dimensional multimodal transport spatiotemporal network is divided into several modal segments, and the target feature risk field corresponding to each modal segment is matched in the motion modality-risk feature correspondence library. Several target spatiotemporal nodes contained within the target feature risk field are detected, and the safety margin value of each target spatiotemporal node is calculated according to the device parameters, the communication time window, and the dynamic load status.

[0063] Furthermore, the building module is specifically used for: The device parameters, the communication time window, and the dynamic load state are converted into corresponding constraints, and based on the constraints, the state sequence of each constraint within the future preset time window is predicted by the Kalman filter algorithm. Based on the historical operation data of the sorting robot, the propagation coefficient matrix of each constraint condition between different working modes is calculated, and the total influence coefficient of each constraint condition on the target spatiotemporal node is calculated according to the propagation coefficient matrix. The state sequence is converted into a corresponding risk value sequence, and the corresponding risk entropy is calculated based on the risk value sequence. The risk entropy of each constraint is weighted and summed based on the total influence coefficient to calculate the safety margin value of each target spatiotemporal node.

[0064] Furthermore, the search module is specifically used for: During the path search process, the safety margin value of the parent node and the safety margin value of the child node are weighted and averaged to obtain the cumulative safety margin value of each child node. When the cumulative safety margin value of a child node is higher than the preset margin threshold, and the search reaches the termination spatiotemporal node, an initial path is generated by backtracking from the termination spatiotemporal node, and each initial node in the initial path is identified. The actual arrival time of each initial node is detected to be within its communication time window. For initial nodes that are outside the communication time window, a local re-search is performed within a preset time and space range with the current initial node as the center to generate an alternative path segment that meets the communication time window. The alternative path segment is then merged with the initial path to find the optimal path.

[0065] Furthermore, the integration module is specifically used for: Discrete path nodes are used as anchor points of the spatiotemporal corridor, and an initial spatiotemporal corridor is generated by expanding in the spatial and temporal dimensions with each anchor point as the center. The initial spatiotemporal corridor is then checked for safety margin in order to generate the corresponding target spatiotemporal corridor. The spatiotemporal corridor information of other sorting robots in the current multimodal transport scenario is detected, so as to determine whether the target spatiotemporal corridor overlaps with the spatiotemporal corridors of other sorting robots. If so, for each overlapping area, the time interval and spatial range of the conflict are calculated to adjust the target spatiotemporal corridor accordingly, and the adjusted target spatiotemporal corridor is converted into the multimodal transport path.

[0066] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the multimodal transport path planning method based on a medium and large-sized sorting robot as described above.

[0067] The fifth embodiment of the present invention provides a readable storage medium storing a computer program thereon, wherein the program, when executed by a processor, implements the multimodal transport path planning method based on a medium and large-sized sorting robot as described above.

[0068] In summary, the multimodal transport path planning method and system based on medium and large-sized sorting robots provided in the above embodiments of the present invention can achieve deep coupling between risk factors and the spatiotemporal characteristics of multimodal transport by constructing a three-dimensional multimodal transport spatiotemporal network that integrates spatial coordinates, time dimension, and intermodal transport node attributes, and mapping the overturning risk field into it. By combining the equipment parameters of each node, communication time window, and dynamic load status to allocate safety margin, the method and system can achieve deep coupling between risk factors and the spatiotemporal characteristics of multimodal transport. The method and system use the spatiotemporal A* algorithm to search for paths and retain only nodes that meet the safety margin requirements, and finally generate multimodal transport paths that are suitable for the target goods. This significantly improves the safety of multimodal transport path planning for medium and large-sized sorting while taking into account the efficiency and rationality of path planning.

[0069] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0070] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0071] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0072] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0073] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A multimodal transport path planning method based on a medium-to-large item sorting robot, characterized in that, The method includes: After the sorting robot grabs the target large item, it controls the sorting robot to perform preset multi-directional excitation actions to detect the torque of each joint, the torque of the end effector, and the body posture data of the sorting robot. Based on the detected data, the center of gravity offset and rotational inertia tensor of the target large item are identified to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. A three-dimensional multimodal transport spatiotemporal network is constructed, which includes spatial coordinates, time dimension and intermodal transport node attributes. The three-dimensional overturning risk field is mapped to the interior of the three-dimensional multimodal transport spatiotemporal network. Based on the equipment parameters, communication time windows and dynamic load status of each intermodal transport node, a corresponding safety margin value is assigned to each spatiotemporal node. Starting with the initial spatiotemporal node as the initial state and ending with the target spatiotemporal node as the target state, the spatiotemporal A* algorithm is used for path search, and only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes during the search process. The path nodes are integrated in three dimensions to generate a multimodal transport path that is compatible with the large cargo in the target.

2. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 1, characterized in that, The step of identifying the center of gravity offset and rotational inertia tensor of the large cargo in the target based on the detected data, in order to construct a three-dimensional overturning risk field with the robot's center of mass as the origin, includes: Based on the joint torques, the end effector torques, and the body posture data, the actual ground loads of the robot's four wheels are obtained by inverting the static equilibrium equations. The load data of each wheel under different excitation actions are collected, and the wheel load data are processed by a preset linear algorithm to calculate the center of gravity offset and the moment of inertia tensor. A corresponding attitude space is constructed based on the linear acceleration, angular velocity and heading angle of the sorting robot. An adaptive sampling algorithm is used to calculate the load transfer rate of the four wheels for each sampled attitude point in the attitude space, and the critical attitude point of the load transfer rate is determined accordingly. The moving least squares method is used to fit each of the critical attitude points to generate a continuous three-dimensional critical overturning surface, and the three-dimensional overturning risk field is constructed according to the attitude space and the three-dimensional critical overturning curve.

3. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 2, characterized in that, The step of constructing the three-dimensional overturning risk field based on the attitude space and the three-dimensional critical overturning curve includes: According to the preset time window, the load transfer rate time series of the four wheels of the sorting robot are detected, and based on Miner's linear cumulative damage theory, the cumulative damage factor of each wheel is calculated according to the load transfer rate time series. Based on the calculated cumulative damage factor of each wheel, the equivalent stiffness coefficient of the sorting robot is updated accordingly. Based on the updated equivalent stiffness coefficient, the three-dimensional critical overturning curve is dynamically corrected to obtain the intermediate three-dimensional critical overturning curve. The attitude space is divided into voxel grids of a preset resolution. The Euclidean distance between the attitude point corresponding to each voxel grid and the intermediate three-dimensional critical overturning curve is calculated. The Euclidean distance is multiplied by the cumulative damage factor of the corresponding attitude point to obtain the risk field strength value of each voxel grid. Each risk field strength value is then inserted into the interior of the intermediate three-dimensional critical overturning curve to generate the three-dimensional overturning risk field.

4. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 1, characterized in that, The step of mapping the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network, and allocating corresponding safety margin values ​​to each spatiotemporal node by combining the equipment parameters, communication time windows, and dynamic load status of each intermodal transport node, includes: The independent component analysis algorithm is used to perform modal decomposition on the three-dimensional overturning risk field to obtain a feature risk field containing several basic motion modes, and the modal feature parameters of each feature risk field are detected accordingly to construct a corresponding motion mode-risk feature correspondence library. The three-dimensional multimodal transport spatiotemporal network is divided into several modal segments, and the target feature risk field corresponding to each modal segment is matched in the motion modality-risk feature correspondence library. Several target spatiotemporal nodes contained within the target feature risk field are detected, and the safety margin value of each target spatiotemporal node is calculated according to the device parameters, the communication time window, and the dynamic load status.

5. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 4, characterized in that, The step of calculating the safety margin value of each target spatiotemporal node based on the device parameters, the communication time window, and the dynamic load status includes: The device parameters, the communication time window, and the dynamic load state are converted into corresponding constraints, and based on the constraints, the state sequence of each constraint within the future preset time window is predicted by the Kalman filter algorithm. Based on the historical operation data of the sorting robot, the propagation coefficient matrix of each constraint condition between different working modes is calculated, and the total influence coefficient of each constraint condition on the target spatiotemporal node is calculated according to the propagation coefficient matrix. The state sequence is converted into a corresponding risk value sequence, and the corresponding risk entropy is calculated based on the risk value sequence. The risk entropy of each constraint is weighted and summed based on the total influence coefficient to calculate the safety margin value of each target spatiotemporal node.

6. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 1, characterized in that, The steps of using the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state to perform path search using the spatiotemporal A* algorithm include: During the path search process, the safety margin value of the parent node and the safety margin value of the child node are weighted and averaged to obtain the cumulative safety margin value of each child node. When the cumulative safety margin value of a child node is higher than the preset margin threshold, and the search reaches the termination spatiotemporal node, an initial path is generated by backtracking from the termination spatiotemporal node, and each initial node in the initial path is identified. The actual arrival time of each initial node is detected to be within its communication time window. For initial nodes that are outside the communication time window, a local re-search is performed within a preset time and space range with the current initial node as the center to generate an alternative path segment that meets the communication time window. The alternative path segment is then merged with the initial path to find the optimal path.

7. The multimodal transport path planning method based on a medium-to-large item sorting robot according to claim 1, characterized in that, The step of performing three-dimensional integration processing on each of the path nodes to generate a multimodal transport route adapted to the target oversized cargo includes: Discrete path nodes are used as anchor points of the spatiotemporal corridor, and an initial spatiotemporal corridor is generated by expanding in the spatial and temporal dimensions with each anchor point as the center. The initial spatiotemporal corridor is then checked for safety margin in order to generate the corresponding target spatiotemporal corridor. The spatiotemporal corridor information of other sorting robots in the current multimodal transport scenario is detected, so as to determine whether the target spatiotemporal corridor overlaps with the spatiotemporal corridors of other sorting robots. If so, for each overlapping area, the time interval and spatial range of the conflict are calculated to adjust the target spatiotemporal corridor accordingly, and the adjusted target spatiotemporal corridor is converted into the multimodal transport path.

8. A multimodal transport path planning system based on a medium-to-large item sorting robot, characterized in that, The system includes: The detection module is used to control the sorting robot to perform preset multi-directional excitation actions after the sorting robot grabs the target large item, so as to detect the torque of each joint, the torque of the end effector and the body posture data of the sorting robot, and identify the center of gravity offset and rotational inertia tensor of the target large item based on the detected data, so as to construct a three-dimensional overturning risk field with the robot's center of mass as the origin. The construction module is used to construct a three-dimensional multimodal transport spatiotemporal network that includes spatial coordinates, time dimension and intermodal transport node attributes, and to map the three-dimensional overturning risk field to the interior of the three-dimensional multimodal transport spatiotemporal network. Combining the equipment parameters, communication time windows and dynamic load status of each intermodal transport node, the module allocates corresponding safety margin values ​​to each spatiotemporal node. The search module is used to perform path search using the spatiotemporal A* algorithm with the starting spatiotemporal node as the initial state and the ending spatiotemporal node as the target state. During the search process, only spatiotemporal nodes with a safety margin value greater than a preset margin threshold are retained as path nodes. The integration module is used to perform three-dimensional integration processing on each of the path nodes to generate a multimodal transport route that is compatible with the large cargo in the target.

9. A dedicated sorting robot, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multimodal transport path planning method based on a medium-to-large item sorting robot as described in any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the multimodal transport path planning method based on a large and medium-sized sorting robot as described in any one of claims 1 to 7.

Images