Autonomous transfer method, apparatus and storage medium for a robot

Abstract

The application relates to the technical field of intelligent robots, and provides an autonomous transfer method and device of a robot and a storage medium, the method comprising the following steps: a cloud platform issuing a transfer task set with priorities to multiple robots; after receiving the task, the robot generates an initial path based on reinforcement learning and feeds back a state to the cloud platform; the robot fuses laser radar and multispectral camera data to perform three-dimensional semantic modeling, and processes multiple-source information to construct a dynamic traffic flow field model; after navigating to a task starting point, the robot verifies the target identity through positioning and image recognition; after verification, the robot performs adaptive loading with force control feedback; after loading, the robot generates an energy consumption optimal path according to a real-time traffic flow model to perform transportation; after arriving at a destination, the robot identifies unloading area features through three-dimensional scene reconstruction and autonomously performs unloading; a return path is calculated, resources are released, and whole-process data is uploaded to the cloud platform to generate a tamper-proof task certificate. The application can realize safe, efficient and fully autonomous transfer of the robot in a dynamic environment.

Description

Technical Field

[0001] This application relates to the field of intelligent robot technology, and in particular to a method, device and storage medium for autonomous transfer of a robot. Background Technology

[0002] With the rapid development of industries such as intelligent manufacturing and smart logistics, mobile robots are increasingly being used for automated transport in environments such as warehouses and workshops. Traditional automated guided vehicles (AGVs) mostly rely on pre-set magnetic strips, QR codes, or laser reflectors to operate in structured environments, which limits their flexibility and environmental adaptability. To cope with more complex and dynamic unstructured factory environments, the new generation of autonomous mobile robots is beginning to use LiDAR, visual sensors, and other technologies for autonomous navigation and obstacle avoidance, achieving a higher degree of flexible task execution.

[0003] In existing robotic transport solutions, a central control system typically assigns tasks to multiple robots. This system sends target point instructions to the robot swarm via a wireless network, and the robots plan their paths and move based on built-in maps. Another common approach is to deploy fixed identification and guidance devices (such as RFID readers and visual beacons) at the transport start and end points to assist the robots in positioning and alignment, thereby enabling them to perform loading or unloading operations.

[0004] However, the aforementioned existing technical solutions still suffer from significant drawbacks, including a lack of predictability in path planning, low efficiency and safety risks, weak coupling between task scheduling and robot autonomous decision-making, and a lack of proactive and precise verification. These shortcomings prevent existing technologies from fully exploiting the potential value of unlabeled data and limit the robustness of models in complex and open environments. Summary of the Invention

[0005] This application provides a method, device, and storage medium for autonomous transfer of a robot, which enables safe, efficient, and fully autonomous transfer of the robot in a dynamic environment through multi-source perception and collaborative decision-making.

[0006] On the one hand, this application provides an autonomous transfer method for robots, the method comprising:

[0007] The cloud platform synchronously distributes priority-based transfer task sets to multiple distributed robot transfer bases.

[0008] The robot transfer base receives the transfer task set, generates an initial navigation path based on a reinforcement learning algorithm, and feeds back the resource scheduling status to the cloud platform.

[0009] The robot transfer base moves along the initial navigation path. During the movement, it performs three-dimensional semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. At the same time, it constructs a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds and broadcast information.

[0010] The robot transfer base navigates to the task starting point based on the dynamic traffic flow field model;

[0011] At the starting point of the task, the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device;

[0012] If the identity verification is successful, the robot transfer base performs an adaptive loading operation based on force control feedback. After loading is completed, it dynamically generates an energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model.

[0013] Upon arrival at the destination, the robot transfer base identifies the unloading area features of the destination through 3D scene reconstruction and autonomously performs the unloading operation.

[0014] After unloading is completed, the robot transfer base calculates the return path and sends a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

[0015] On the other hand, this application provides an autonomous transfer device for robots, the device comprising:

[0016] The distribution module is used to synchronously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through the cloud platform;

[0017] The first processing module is used to receive the transfer task set from the robot transfer base, generate an initial navigation path based on a reinforcement learning algorithm, and report the resource scheduling status to the cloud platform.

[0018] The model building module is used for the robot transfer base to move along the initial navigation path. During the movement, it performs three-dimensional semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. At the same time, it constructs a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds and broadcast information.

[0019] The navigation module is used to navigate the robot transfer base to the task starting point based on the dynamic traffic flow field model;

[0020] The verification module is used at the starting point of the task to verify the identity of the target robot through its positioning module and image acquisition device.

[0021] The second processing module is used to perform an adaptive loading operation based on force control feedback on the robot transfer base if the identity verification is successful. After the loading is completed, the module dynamically generates the energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model.

[0022] The operation module is used to enable the robot transfer base to autonomously perform unloading operations after reaching the destination by recognizing the unloading area features of the destination through three-dimensional scene reconstruction.

[0023] The feedback module is used to calculate the return path of the robot transfer base after unloading and send a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

[0024] Thirdly, this application provides an electronic device, the device 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 steps of the technical solution of the autonomous transfer method of the robot described above.

[0025] Fourthly, this application provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described autonomous transfer method for robots.

[0026] As can be seen from the technical solution provided in this application, on the one hand, during the movement process, by fusing perception data from LiDAR and multispectral cameras to perform three-dimensional semantic modeling of unstructured roads, and simultaneously by processing visual data, LiDAR point clouds, and broadcast information to construct a dynamic traffic flow field model centered on itself, the robot can achieve deep semantic understanding of the unstructured environment and global traffic situation awareness, providing a data foundation for subsequent high-level decision-making and significantly enhancing the robot's autonomous operation capability in dynamic and uncertain environments. On the other hand, by distributing a set of transfer tasks with priority identifiers through the cloud platform and receiving resource status feedback from the robots, a global collaborative framework is formed, while empowering each robot with [missing information - likely related to traffic situation awareness]. The system leverages reinforcement learning to generate initial paths and autonomous decision-making capabilities to dynamically generate energy-optimized paths based on real-time traffic flow. This ensures system-level task coordination and resource optimization while fully utilizing the environmental adaptability of individual robots, thereby improving the overall transfer efficiency and flexibility of the multi-robot system. Thirdly, by having the robot actively initiate and complete dual authentication based on localization and image recognition at the task's starting point, along with adaptive loading operations based on force control feedback, the system eliminates reliance on fixed auxiliary facilities. This not only enhances the robot's operational autonomy but also ensures the safety of the interaction process through force / position hybrid control, significantly reducing the risk of equipment damage or task failure due to positioning or docking deviations. In summary, the technical solution of this application achieves safe, efficient, and fully autonomous transfer of robots in dynamic environments through multi-source perception and collaborative decision-making. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0028] Figure 1 This is a flowchart of the autonomous transfer method for robots provided in the embodiments of this application;

[0029] Figure 2 This is a schematic diagram of the structure of the autonomous transfer device for the robot provided in the embodiments of this application;

[0030] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] In this specification, adjectives such as "first" and "second" are used only to distinguish one element or action from another, without necessarily requiring or implying any actual such relationship or order. Where circumstances permit, reference to an element or component or step (etc.) should not be construed as being limited to only one of the elements, components, or steps, but may be one or more of the elements, components, or steps, etc.

[0033] For ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn to actual scale.

[0034] Traditional automated guided vehicles (AGVs) mostly rely on pre-set magnetic strips, QR codes, or laser reflectors to operate in structured environments, which limits their flexibility and environmental adaptability. To cope with the more complex and dynamic unstructured factory environments, the next generation of autonomous mobile robots is beginning to use LiDAR, visual sensors, and other technologies for autonomous navigation and obstacle avoidance, achieving a higher degree of flexible task execution. In existing robot transfer solutions, a central control system typically assigns tasks to multiple robots, sending target point instructions to the robot swarm via a wireless network. The robots then plan their paths and move based on built-in maps. Another common approach is to deploy fixed identification and guidance devices (such as RFID readers and visual beacons) at the transfer start and end points to assist the robots in positioning and alignment, thereby performing loading or unloading operations. However, the above-mentioned existing technologies still have significant drawbacks: First, the environmental perception dimensions of existing solutions are limited, relying heavily on static maps and local obstacle avoidance. They struggle to effectively model semantic information such as variable lane lines and temporary traffic signs in unstructured roads, as well as global dynamic traffic flow, resulting in unpredictable path planning, low efficiency, and safety risks. Secondly, the coupling between task scheduling and robot autonomous decision-making is weak, making it difficult for the central system to perform dynamic and optimal collaborative scheduling based on real-time changes in robot status and environmental information. Finally, in critical authentication and interaction stages (e.g., loading), existing solutions largely rely on fixed site facilities, and the robot itself lacks proactive, precise authentication and compliant control capabilities, limiting its true autonomy and deployment flexibility throughout the entire process. These shortcomings prevent existing technologies from fully exploiting the potential value of unlabeled data and limit the robustness of models in complex open environments.

[0035] To address the problems of the prior art, this application provides an autonomous transfer method for robots, the flowchart of which is attached. Figure 1 As shown, the main steps include S101 to S108, which are detailed below:

[0036] Step S101: Simultaneously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through the cloud platform.

[0037] It should be noted that in this embodiment, the robotic transfer base, or Automatic Mobile Robot (AMR), is similar to a forklift—a mobile, proactive transportation tool responsible for performing the entire process of finding goods at one location, picking them up, transporting them to another, and then putting them down. When the robotic transfer base starts, the following initialization steps are performed: a lightweight neural network model is downloaded via a cloud platform; during subsequent operation, locally collected perception data is used for online incremental learning of the lightweight neural network model, and the updated model parameters are uploaded to the cloud platform for federated learning aggregation to continuously optimize the perception capabilities of all robots; the lightweight neural network model is a convolutional neural network used to recognize temporary traffic signs, and the online incremental learning specifically involves: after each successful recognition and cross-validation using Vehicle-to-Everything (V2X) information, the image data and validation results used in that recognition are used as new training samples to fine-tune the local model.

[0038] In existing technologies, on the one hand, if robots communicate and negotiate tasks peer-to-peer through self-organizing networks (e.g., ad-hoc), the negotiation process is complex, communication overhead is huge, and it is difficult to form an optimal scheduling strategy at the global level, easily getting trapped in local optima, resulting in low overall system efficiency. On the other hand, if the transfer tasks have no priority identifiers and simple scheduling rules such as first-in-first-out (FIFO) are used, it is unable to cope with dynamic situations such as emergency task queue jumping, uneven robot battery levels, and traffic congestion, resulting in low scheduling intelligence and poor system flexibility. Therefore, in order to achieve efficient and orderly collaborative management of multiple robots without stifling the autonomy of individual robots and to avoid system inefficiency caused by task conflicts and resource competition, this application can synchronously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through a cloud platform. This set of transfer tasks includes dynamic path point sequences, time constraints, and environmental adaptation parameters, etc. In the above embodiments, the cloud platform, as the command center, can assign coordinated task instructions to all robots based on global information, while the priority identifier can ensure that system resources can be given priority to the most critical transfer tasks, thereby improving the overall operational efficiency and responsiveness at the system level.

[0039] Step S102: The robot transfer base receives the transfer task set, generates an initial navigation path based on the reinforcement learning algorithm, and feeds back the resource scheduling status to the cloud platform.

[0040] In the field of robotic transport, there is an inherent contradiction between global path optimization and real-time changes in the local environment. That is, while the cloud platform may issue macroscopic path points, the specific decision on how to safely and efficiently arrive requires the robot to make its own understanding of the local environment. Existing technologies, where the cloud platform uniformly calculates and issues detailed end-to-end paths for all robots, have significant drawbacks, including enormous computational pressure, high communication bandwidth requirements, difficulty in real-time response to dynamic obstacles encountered by each robot, high delays in replanning if the path is interrupted, and poor robot autonomy. To achieve distributed autonomous decision-making and resolve the contradiction between global path optimization and real-time changes in the local environment, the robot transport base receives the transport task set and generates an initial navigation path based on a reinforcement learning algorithm, while also feeding back the resource scheduling status to the cloud platform. Because this solution empowers the robot with the ability to autonomously generate initial paths based on reinforcement learning algorithms, it possesses a certain degree of thinking and decision-making ability, enabling it to quickly respond to task initiation needs. The status feedback to the cloud platform provides a data foundation for subsequent dynamic collaborative scheduling.

[0041] In the above embodiments, feeding back the resource scheduling status to the cloud platform can specifically involve uploading the robot's calculated initial navigation path, current battery level, and estimated arrival time at the task starting point as status information to the cloud platform via a 5G-V2X communication module. Upon receiving this status information, the cloud platform dynamically adjusts the priority identifiers of the transfer task set based on the status information fed back by all robot transfer bases, using a multi-agent collaborative negotiation scheduling algorithm. Specifically, this adjustment involves: generating a scheduling preference value for each robot corresponding to its pending task based on the received status information; the rule for the scheduling preference value is: the shorter the estimated arrival time at the task starting point and the higher the current battery level, the higher the generated scheduling preference value; using the scheduling preference values ​​of all robots as input, running the multi-agent collaborative negotiation algorithm to simulate multiple rounds of negotiation to find an equilibrium point, where the equilibrium point corresponds to an optimal task priority allocation scheme; dynamically adjusting the priority identifiers of each task in the transfer task set according to the obtained optimal task priority allocation scheme; and synchronously sending the adjusted transfer task set or its priority identifiers to the corresponding robot transfer bases.

[0042] Step S103: The robot transfer base moves along the initial navigation path. During the movement, it performs three-dimensional semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. At the same time, it constructs a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds and broadcast information.

[0043] In the field of intelligent navigation, intelligent agents such as robots in complex and unstructured environments cannot obtain sufficient semantic information (e.g., the meaning of traffic signs) if relying solely on a single sensor (e.g., pure LiDAR). This means that if only LiDAR is used for SLAM mapping and obstacle avoidance, the robot cannot recognize semantic information such as lane lines and traffic signs, nor understand traffic rules, and its behavior may violate regulations or fail to integrate into traffic flow. If only a visual camera is used, the robot is greatly affected by lighting and weather, its ranging accuracy is inferior to LiDAR, and its reliability is poor in harsh environments. Therefore, in order to achieve highly robust environmental perception and deep understanding, and to resolve the contradiction between the limitations of a single perception modality and the need for comprehensive environmental understanding, the technical means adopted in this application can be that the robot's transport base moves along the initial navigation path. During the movement, the robot performs 3D semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. At the same time, by processing visual data, LiDAR point clouds, and broadcast information, a dynamic traffic flow field model centered on the robot is constructed. In the above embodiments, the 3D semantic modeling of unstructured roads by fusing perception data from LiDAR and multispectral cameras can specifically be as follows: high-precision point cloud data is acquired through LiDAR to reconstruct the 3D geometric structure of the road; visible light and infrared band images are acquired through multispectral cameras to extract semantic information such as lane lines and traffic signs; a spatiotemporal registration algorithm is used to align the point cloud data with the multispectral images to eliminate temporal and spatial biases between sensors; the fused data is segmented and classified based on a deep learning model (e.g., a convolutional neural network) to generate a 3D voxel map containing semantic labels such as road surface, obstacles, and traffic signs; and this semantic map is used as the basic input for a dynamic traffic flow field model.

[0044] In the above embodiments, the fusion of multiple sensors (LiDAR, multispectral cameras, and broadcast terminals, etc.) compensates for the deficiencies of a single sensor, enabling the robot not only to perceive the existence of obstacles, but also to understand their semantics ("what is it") and dynamic intentions ("how it will move"). Therefore, it can provide richer information input than traditional obstacle avoidance for subsequent high-order decisions.

[0045] Specifically, as an embodiment of this application, a dynamic traffic flow field model centered on itself can be constructed by processing visual data, LiDAR point clouds, and broadcast information through steps S1031 to S1033, as detailed below:

[0046] Step S1031: The lane lines and traffic signs identified by visual data, the obstacle outlines and positions clustered by LiDAR point clouds, and the trajectory prediction results of other traffic participants in V2X broadcast information are spatiotemporally aligned and fused to generate a dynamic grid map containing vector traffic direction, velocity field and obstacle probability distribution.

[0047] The aforementioned dynamic grid map, which includes vector traffic direction, velocity field, and obstacle probability distribution, is a key component of the dynamic traffic flow field model, and the robot base can use it for navigation.

[0048] Step S1032: Cross-validate the temporary traffic signs identified in the dynamic traffic flow field model with the traffic control information received through the 5G-V2X communication module.

[0049] Here, the temporary traffic signs identified in the dynamic traffic flow model are cross-validated with the traffic control information received through the 5G-V2X communication module. Specifically, the robot transfer base performs a semantic comparison between the temporary traffic signs identified in the dynamic traffic flow model by its own perception system and the traffic control information received through the 5G-V2X communication module. If the comparison results show a semantic conflict, it is determined that the temporary traffic sign and the traffic control information conflict. Here, a semantic conflict means that the self-perceived information (i.e., the temporary traffic signs identified by the robot transfer base's own perception system in the dynamic traffic flow model) indicates that passage is possible, while the traffic control information received through the 5G-V2X communication module indicates that passage is not possible.

[0050] Step S1033: If a conflict is found between temporary traffic signs and traffic control information, the traffic control information shall be used first to update the dynamic traffic flow model, and local routes shall be replanned to avoid entering the controlled area.

[0051] Specifically, once it is determined that there is a conflict between temporary traffic signs and traffic control information, the traffic control information received through the 5G-V2X communication module is adopted as the basis for decision-making. Based on the adopted traffic control information, the dynamic traffic flow field model is updated, and new control areas are marked in the model or the traffic status of relevant areas is updated. Based on the updated dynamic traffic flow field model, a local path replanning algorithm is triggered to calculate a new local path that can avoid the control area. The robot transfer base is controlled to travel along the newly planned local path.

[0052] Step S104: The robot transfer base navigates to the task starting point based on the dynamic traffic flow field model.

[0053] Because the dynamic traffic flow field model contains a dynamic grid map of vector traffic direction, velocity field, and obstacle probability distribution, the robot transfer base can navigate to the task starting point based on the dynamic traffic flow field model.

[0054] Step S105: At the starting point of the task, the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device.

[0055] As mentioned earlier, in this embodiment, the robot transfer base, or Automatic Mobile Robot (AMR), is similar to a forklift—a mobile, proactive transportation tool responsible for locating goods, picking them up, transporting them to another location, and then putting them down. The target robots are those with limited mobility or in a dormant / waiting state that need to be moved from one workstation, charging station, or storage area to another by the robot transfer base. This includes the base of a fixed-position robotic arm, a warehouse shelf robot, a sweeping robot that needs to return to its charging station, or an AGV (Automated Guided Vehicle) on a production line, etc. The positioning module mounted on the robot transfer base can be an Ultra Wide Band (UWB) positioning module. However, while UWB positioning offers high accuracy, it lacks visual confirmation, posing a risk of misalignment with similarly sized objects. Conversely, image recognition alone is susceptible to lighting and occlusion, and simple recognition cannot provide accurate relative pose and distance information, hindering subsequent precise physical interaction. Therefore, in order to achieve highly reliable interactive object recognition and positioning, and to resolve the contradiction between accuracy and reliability in object recognition in complex environments, this application can adopt a positioning scheme that combines UWB positioning and image recognition. That is, at the starting point of the task, the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device; here, the positioning module can be a UWB positioning module; and the image acquisition device can be a multispectral camera.

[0056] Specifically, as one embodiment of this application, the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device via steps S1051 to S1053, as detailed below:

[0057] Step S1051: The robot transfer base obtains the distance and angle information of the target robot through its onboard UWB positioning module to roughly locate the approximate area of ​​the target robot.

[0058] Step S1052: The robot transfer base activates the multispectral camera to scan the approximate area where the target robot is located, and identifies specific visual identifiers on the target robot through a deep learning model.

[0059] Specifically, step S1052 can be implemented as follows: based on the coarse area provided by UWB positioning, control a multispectral camera to perform panoramic scanning and acquire visible light and infrared image sequences; then, preprocess the images (e.g., denoising, enhancement) and input them into a pre-trained deep learning model (e.g., YOLO or Faster R-CNN) for object detection and locate the bounding boxes of visual identifiers; parse the texture and shape features of the identifiers through a feature extraction network (e.g., ResNet) and perform similarity matching with pre-stored identifier templates; if the matching degree exceeds a threshold, output the recognition result.

[0060] Step S1053: The robot transfer base matches the visual recognition results with the UWB positioning data and the identity authentication information pre-stored in the transfer task set. If the match is successful, the identity is confirmed.

[0061] Specifically, step S1053 can be implemented as follows: compare the identifier content obtained by visual recognition with the target identity code pre-stored in the task set to check for consistency; align the approximate coordinates of the target robot provided by UWB positioning with the spatial position of the identifier obtained by visual recognition to verify whether the two coincide within the tolerance range; if the visual identifier is successfully matched and the spatial position is consistent, then the identity verification is determined to be successful; otherwise, trigger a retry or report an exception.

[0062] Step S106: If the identity verification is successful, the robot transfer base performs an adaptive loading operation based on force control feedback. After loading is completed, the optimal energy consumption path to the destination is dynamically generated based on the real-time updated dynamic traffic flow field model.

[0063] On the one hand, pure trajectory position control cannot handle minute deviations during the docking process, easily leading to mechanical collisions, jamming, or damage to the goods. This means that for the loading operation of the robot transfer base, if a pure position control mode is used, the positioning accuracy requirements for the robot and the target object are extremely high, with very poor fault tolerance. Any slight deviation may lead to task failure or equipment damage, lacking intelligence and safety. Therefore, for safe and compliant physical interaction, if authentication is successful, the robot transfer base performs an adaptive loading operation based on force control feedback. On the other hand, after loading, the robot's weight, inertia, and other states change, and the energy consumption model also changes accordingly, while the environment is also constantly changing. This means that after loading is complete, if the robotic transfer platform continues to use the initial path to the task starting point or a fixed path for transportation, it will be unable to adapt to the energy consumption characteristics after the load changes. It may choose a seemingly shorter but actually more energy-consuming path (e.g., a steep slope path), and it will also be unable to cope with newly emerging congestion during transportation, resulting in low efficiency. Therefore, in order to achieve state adaptation and high-precision control, after loading is completed, the robotic transfer platform dynamically generates the energy-optimal path to the destination based on a real-time updated dynamic traffic flow field model. As can be seen from the above embodiments, by introducing force control feedback, the robotic transfer platform can sense the contact force and flexibly adjust its movements like a human to compensate for deviations, thereby achieving successful and damage-free loading under non-ideal conditions, greatly improving the reliability and intelligence level of the system. The dynamic generation of the energy-optimal path reflects the robotic transfer platform's dual adaptability to its own state and the external environment.

[0064] As one embodiment of this application, the robot transfer base performing an adaptive loading operation based on force control feedback can be as follows: during the process of the fork inserting into the target robot pallet along the planned three-dimensional trajectory, the data of the six-dimensional force (or torque) sensor at the end of the fork is monitored in real time; if the force vector is detected to deviate from the expected insertion direction, the compliant attitude control algorithm based on force feedback dynamically adjusts the fork attitude to compensate for the docking deviation until the insertion is successful; wherein, the specific implementation of the compliant attitude control algorithm based on force feedback dynamically adjusting the fork attitude to compensate for the docking deviation until the insertion is successful can include steps S1 to S6, as described below: Step S1: During the process of the fork inserting into the target pallet along the preset three-dimensional trajectory, the data of the six-dimensional force (or torque) sensor at the end of the fork is read and parsed in real time to obtain the actual contact force vector; Step S2: The actual contact force... Step S3: Compare the force (or torque) deviation value with the ideal contact force vector in the expected insertion direction to calculate the force (or torque) deviation value; Step S4: Input the force (or torque) deviation value into the preset impedance control model to map the force (or torque) deviation into the pose adjustment amount that the fork end needs to make. The impedance control model is based on preset impedance parameters (including stiffness matrix and damping matrix); Step S5: Calculate the pose adjustment amount through robot inverse kinematics and convert it into control commands for each joint motor; Step S6: Drive the joint motors to execute the control commands, so that the fork end produces a compliant posture adjustment action to compensate for the docking deviation; Step S7: Repeat steps S1 to S5 until the force (or torque) deviation value is continuously within an acceptable threshold range near zero and is maintained for more than a preset time threshold, then the insertion is considered successful.

[0065] As can be seen from the above embodiments, the robot transfer base's adaptive loading operation based on force control feedback solves the problem of how to safely and accurately insert and remove pallets. However, successful insertion does not guarantee a safe transportation process. This is because the load's center of gravity may not be at the center of the forks (due to manufacturing tolerances, misalignment, or minor deviations during insertion), resulting in a tipping moment during lifting. If this is not addressed, it can lead to minor issues such as goods swaying and slipping, or even serious issues such as the robot becoming unstable and tipping over due to a shift in its center of gravity, or damage to the mechanical structure. Therefore, in the specific implementation of the robot transfer base's adaptive loading operation based on force control feedback, after the forks successfully insert the pallet into the target robot along the planned three-dimensional trajectory, the system further includes: monitoring the load's center of gravity shift using a six-dimensional force (or torque) sensor, and dynamically adjusting the lifting speed of the forks on both sides or activating a leveling mechanism to ensure that the load remains balanced during lifting. Specifically, monitoring the load's center of gravity shift using a six-dimensional force (or torque) sensor and dynamically adjusting the lifting speed of the forks on both sides or activating a leveling mechanism to ensure the load remains balanced during lifting can be achieved by: real-time reading of sensor data, calculating the torque distribution of the load on the fork plane, and determining center of gravity shift if the detected torque deviation exceeds a threshold; dynamically adjusting the lifting speed of the forks on both sides (e.g., decelerating on the offset side and accelerating on the opposite side) or activating an electric leveling mechanism (e.g., hydraulic outriggers) to compensate for tilt; ensuring the load remains balanced during lifting and preventing tipping.

[0066] The aforementioned force (or torque) sensor can not only sense contact, but also continuously sense the balance of the load during the lifting process, and actively intervene before tilting occurs, fundamentally avoiding the risk of tipping over. The dynamic adjustment or activation of the leveling mechanism is an intelligent and adaptive compensation method that ensures that the load is lifted smoothly and horizontally from a static state, avoiding the sudden stress or shaking caused by the hard pulling that may occur in traditional methods, thus protecting the goods and the robot itself.

[0067] To ensure the accurate execution of the energy-optimal path to the destination and guarantee the efficiency, stability, and safety of the transportation process, after the robot transfer base dynamically generates the energy-optimal path based on a real-time updated dynamic traffic flow model, centimeter-level trajectory tracking is achieved through fusion positioning during transportation. Furthermore, considering that no single sensor (e.g., IMU, vision, or wheeled odometer, etc.) can simultaneously meet the requirements of high frequency, high precision, and no cumulative error, effectively fusing them to leverage their strengths and compensate for their weaknesses is the core challenge for achieving accurate tracking. Therefore, to achieve centimeter-level trajectory tracking through a specific, efficient, and reliable technical approach, the aforementioned centimeter-level trajectory tracking through fusion positioning during transportation can be specifically achieved as follows: using the high-frequency pose changes output by the Inertial Measurement Unit (IMU) sensor on the robot transfer base as the prediction basis, and the low-frequency absolute pose output by the visual odometer and wheeled odometer on the robot transfer base as the observation values, a fusion method based on prediction and observation is used to generate a high-frequency, high-precision positioning estimate. This estimate is then used to control the drive motor in a closed loop to achieve tracking of the energy-optimal path. Specifically, using the high-frequency pose changes output by the IMU sensor mounted on the robot transport base as the prediction basis, and the low-frequency absolute pose output by the visual odometry and wheeled odometry mounted on the robot transport base as the observation value, a high-frequency, high-precision positioning estimate is generated by fusing the prediction and observations using an optimal estimation method. This can be achieved through steps S1 to S6, namely: Step S1: In each control cycle, receive the high-frequency angular velocity and acceleration data output by the IMU sensor, and calculate the predicted value of the robot transport base pose and its corresponding prediction uncertainty based on the robot kinematic model; Step S2: At the output update time of the visual odometry or wheeled odometry, obtain the low-frequency absolute pose data provided by it as the observation value, and obtain the correlation between the observation value and the prediction uncertainty. The corresponding observation uncertainty; Step S3: Calculate the optimal weighting factor based on the prediction uncertainty and observation uncertainty, whereby the optimal weighting factor is used to weigh the degree of confidence in the predicted value and the observed value; Step S4: Using the optimal weighting factor, fuse the pose prediction value with the absolute pose observation value to calculate the optimal estimate of the robot transfer base pose at the current moment; Step S5: Based on the optimal weighting factor, synchronously update the uncertainty assessment of the current optimal estimate; Step S6: Output the optimal estimate as the high-frequency and high-precision positioning result at the current moment for closed-loop control of the drive motor, and at the same time, use the updated optimal estimate and its uncertainty as the initial value predicted in step S1 at the next moment, and repeat the above steps S1 to S6.

[0068] Step S107: Upon arrival at the destination, the robot transfer base identifies the unloading area features of the destination through 3D scene reconstruction and autonomously performs the unloading operation.

[0069] Since the environment at the destination may vary each time (e.g., cargo stacking, other obstacles, etc.), assuming the unloading area is always ideal and open, and using a fixed unloading trajectory, as in existing technologies, collisions are highly likely to occur in real dynamic environments, posing a significant risk. To resolve the contradiction between the uncertainty of the unloading environment and operational safety, and to achieve adaptive operation capabilities in unknown environments, in this embodiment, upon arrival at the destination, the robot transfer base identifies the unloading area features through 3D scene reconstruction and autonomously performs the unloading operation. This proactive 3D scene reconstruction for identifying and evaluating the unloading environment enables the robot transfer base to autonomously decide how to safely complete the unloading based on the actual situation, further enhancing its autonomy and safety throughout the entire process. Specifically, as an embodiment of this application, the robot transfer base identifies the unloading area features at the destination through 3D scene reconstruction and autonomously performs the unloading operation through steps S1071 to S1074, as detailed below:

[0070] Step S1071: Within the predetermined range of the destination, control the lidar to perform a circular scan, and use a point cloud data iterative alignment algorithm to stitch together the point cloud to generate a three-dimensional dense map of the unloading area.

[0071] Specifically, the implementation process of step S1071 may include steps S1 to S6, detailed as follows: Step S1: Control the robot's transfer base to slowly rotate in place within the predetermined range of the destination, and simultaneously trigger its onboard LiDAR to perform a 360° circular scan to acquire a series of continuous multi-frame LiDAR point cloud data; Step S2: Perform filtering, noise reduction, and outlier removal processing on each acquired frame of point cloud data to improve data quality; Step S3: Select two adjacent frames of point cloud data as the point clouds to be registered, and initially assign them a coarse pose transformation estimate; Step S4: For the current For the two point clouds to be registered, for each point in one frame, find its nearest neighbor in the other frame, and calculate an optimal rigid body transformation matrix that minimizes the sum of the distances between corresponding points based on all point pairs; Step S5: Apply the calculated transformation matrix to the point cloud to be registered, and repeat step S4 until the transformation matrix converges to a stable value or the number of iterations reaches the upper limit; Step S6: Convert all the multi-frame point cloud data registered by steps S3 to S5 to the same global coordinate system, and stitch them together to generate a complete 3D dense map describing the unloading area environment.

[0072] Step S1072: Extract ground flatness, available space size and typical landmark features from the 3D dense map.

[0073] To determine whether the unloading area is suitable for subsequent unloading operations, whether the space is large enough to accommodate the goods to be unloaded, whether there is sufficient safety margin for the robotic arm to operate, and to achieve precise positioning, ground flatness, available space size, and typical landmark features can be extracted from the three-dimensional dense map generated in step S1071. These features are the environmental characteristics of the unloading area.

[0074] Step S1073: Calculate the safe descent height of the forks and the unloading path based on the 3D dense map.

[0075] Specifically, step S1073 can be implemented as follows: reading and parsing the environmental features of the unloading area extracted from the 3D dense map, namely the ground flatness, the 3D size information of the available space, and typical landmark features for precise positioning of the current unloading area; accurately calculating the current position and attitude of the robot transfer base in the global map by matching the real-time sensing data with the typical landmark features; calculating a 3D spatial bounding box for safe unloading operation by deducting the safety margin based on the 3D size information of the available space; calculating a collision-free vertical descent trajectory with the current fork height as the starting point and the bottom boundary of the 3D spatial bounding box for safe unloading operation (while also considering the ground flatness) as the ending point, the ending height of this trajectory is the safe descent height; within the 3D spatial bounding box for safe unloading operation, calculating a collision-free and smooth horizontal translation trajectory from the current fork position to the target unloading point based on the motion planning algorithm, this trajectory is the unloading path; and sending the calculated safe descent height and unloading path sequence to the underlying motion controller to prepare for unloading.

[0076] Step S1074: During the descent, if an unknown obstacle is detected to have intruded into the unloading area via real-time point cloud, the descent is immediately paused and the unloading path is replanned.

[0077] Step S108: After unloading is completed, the robot transfer base calculates the return path and sends a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

[0078] If no feedback is provided after the task is completed, or only a simple "complete" signal is given, the cloud platform cannot accurately grasp the real-time status of the robot transfer base for the next scheduling. Furthermore, the lack of reliable assessment and traceability of task execution quality hinders system optimization and accountability. To achieve a fully digital closed-loop and reliable traceability, in this embodiment, after unloading, the robot transfer base calculates the return path and sends a resource release signal to the cloud platform. Simultaneously, it uploads all operational data to the cloud platform to generate an immutable task execution certificate. This resource release signal mechanism makes system resource management more precise, while uploading all data and generating an immutable certificate provides a data foundation for performance analysis and system optimization, as well as reliable technical assurance for commercial settlement and accountability, thereby improving the overall system reliability and commercial value.

[0079] As one embodiment of this application, in the above embodiment, the robot transfer base calculates the return path as follows: based on the current power level, task set priority, and resource status issued by the cloud platform, with the goal of minimizing idle energy consumption, using the A algorithm or Dijkstra's algorithm to plan the path from the destination back to the charging station or waiting area; querying the real-time dynamic traffic flow model to avoid congested areas. As for uploading the entire operation data to the cloud platform to generate an immutable task execution certificate, its specific implementation includes: the cloud platform compares the received entire operation data with the requirements of the task set to generate a comprehensive score including task completion, time efficiency, and energy efficiency; writing this comprehensive score along with the hash value of key perception data (e.g., images when verification is successful, point cloud snapshots when unloading is complete, etc.) into a blockchain smart contract, which generates the task execution certificate; when the robot transfer base receives the next transfer task set, it queries the blockchain for its own historical task execution certificates and reports the comprehensive score in the certificate as a credit indicator to the cloud platform to participate in the decision-making of new task priorities.

[0080] From the above appendix Figure 1The example of the robot's autonomous transport method demonstrates that, on the one hand, during movement, the robot performs 3D semantic modeling of unstructured roads by fusing perception data from LiDAR and multispectral cameras. Simultaneously, by processing visual data, LiDAR point clouds, and broadcast information, a dynamic traffic flow field model centered on the robot is constructed. This enables the robot to achieve deep semantic understanding of the unstructured environment and global traffic situation awareness, providing a data foundation for subsequent high-level decision-making and significantly enhancing its autonomous operation capabilities in dynamic and uncertain environments. On the other hand, by distributing transport task sets with priority identifiers through a cloud platform and receiving resource status feedback from the robots, a global collaborative framework is formed, while simultaneously empowering each robot with [specific capabilities / resources]. The system leverages reinforcement learning to generate initial paths and autonomous decision-making capabilities to dynamically generate energy-optimized paths based on real-time traffic flow. This ensures system-level task coordination and resource optimization while fully utilizing the environmental adaptability of individual robots, thereby improving the overall transfer efficiency and flexibility of the multi-robot system. Thirdly, by having the robot actively initiate and complete dual authentication based on localization and image recognition at the task's starting point, along with adaptive loading operations based on force control feedback, the system eliminates reliance on fixed auxiliary facilities. This not only enhances the robot's operational autonomy but also ensures the safety of the interaction process through force / position hybrid control, significantly reducing the risk of equipment damage or task failure due to positioning or docking deviations. In summary, the technical solution of this application achieves safe, efficient, and fully autonomous transfer of robots in dynamic environments through multi-source perception and collaborative decision-making.

[0081] Please see the appendix Figure 2 This application provides an autonomous transfer device for a robot. The device may include a distribution module 201, a first processing module 202, a model building module 203, a navigation module 204, a verification module 205, a second processing module 206, an operation module 207, and a feedback module 208, as detailed below:

[0082] The distribution module 201 is used to synchronously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through the cloud platform;

[0083] The first processing module 202 is used to receive the transfer task set on the robot transfer base, generate an initial navigation path based on the reinforcement learning algorithm, and report the resource scheduling status to the cloud platform.

[0084] The model building module 203 is used for the robot transfer base to move along the initial navigation path. During the movement, it performs three-dimensional semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. At the same time, it constructs a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds and broadcast information.

[0085] Navigation module 204 is used for navigating the robot transfer base to the task starting point based on a dynamic traffic flow field model;

[0086] The verification module 205 is used at the starting point of the task, where the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device.

[0087] The second processing module 206 is used to perform an adaptive loading operation based on force control feedback on the robot transfer base if the identity verification is successful. After the loading is completed, the optimal energy consumption path to the destination is dynamically generated according to the real-time updated dynamic traffic flow field model.

[0088] Operation module 207 is used to identify the unloading area features of the destination through 3D scene reconstruction after the robot transfer base arrives at the destination and autonomously perform the unloading operation.

[0089] The feedback module 208 is used to calculate the return path of the robot transfer base after unloading and send a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

[0090] From the above appendix Figure 2 As can be seen from the example of the robot's autonomous transport device, on the one hand, during movement, by fusing perception data from LiDAR and multispectral cameras to perform 3D semantic modeling of unstructured roads, and simultaneously by processing visual data, LiDAR point clouds, and broadcast information to construct a dynamic traffic flow field model centered on itself, the robot can achieve deep semantic understanding of the unstructured environment and global traffic situation awareness, providing a data foundation for subsequent high-level decision-making and significantly enhancing the robot's autonomous operation capability in dynamic and uncertain environments. On the other hand, by distributing transport task sets with priority identifiers through the cloud platform and receiving resource status feedback from the robots, a global collaborative framework is formed, while empowering each robot with [missing information - likely related to resource allocation]. The system leverages reinforcement learning to generate initial paths and autonomous decision-making capabilities to dynamically generate energy-optimized paths based on real-time traffic flow. This ensures system-level task coordination and resource optimization while fully utilizing the environmental adaptability of individual robots, thereby improving the overall transfer efficiency and flexibility of the multi-robot system. Thirdly, by having the robot actively initiate and complete dual authentication based on localization and image recognition at the task's starting point, along with adaptive loading operations based on force control feedback, the system eliminates reliance on fixed auxiliary facilities. This not only enhances the robot's operational autonomy but also ensures the safety of the interaction process through force / position hybrid control, significantly reducing the risk of equipment damage or task failure due to positioning or docking deviations. In summary, the technical solution of this application achieves safe, efficient, and fully autonomous transfer of robots in dynamic environments through multi-source perception and collaborative decision-making.

[0091] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example... Figure 3 As shown, the electronic device 3 in this embodiment mainly includes: a processor 30, a memory 31, and a computer program 32 stored in the memory 31 and executable on the processor 30, such as a program for an autonomous transfer method for a robot. When the processor 30 executes the computer program 32, it implements the steps in the above-described embodiment of the autonomous transfer method for a robot, for example... Figure 1 The steps S101 to S108 are shown. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 2 The functions of the distribution module 201, the first processing module 202, the model building module 203, the navigation module 204, the verification module 205, the second processing module 206, the operation module 207, and the feedback module 208 are shown.

[0092] For example, the computer program 32 of the robot's autonomous transfer method mainly includes: synchronously distributing a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through a cloud platform; the robot transfer base receiving the transfer task set and generating an initial navigation path based on a reinforcement learning algorithm and feeding back the resource scheduling status to the cloud platform; the robot transfer base moving along the initial navigation path, and during the movement, performing three-dimensional semantic modeling of unstructured roads by fusing perception data from LiDAR and multispectral cameras, and simultaneously constructing a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds, and broadcast information; and the robot transfer base based on the dynamic traffic flow field model... The robot navigates to the task starting point. At the starting point, the robot transfer base verifies the identity of the target robot using its onboard positioning module and image acquisition device. If the identity verification is successful, the robot transfer base performs an adaptive loading operation based on force control feedback. After loading is completed, it dynamically generates an energy-optimal path to the destination based on a real-time updated dynamic traffic flow model. Upon reaching the destination, the robot transfer base identifies the unloading area features of the destination through 3D scene reconstruction and autonomously performs the unloading operation. After unloading is completed, the robot transfer base calculates the return path and sends a resource release signal to the cloud platform, while simultaneously uploading the entire operation data to the cloud platform to generate an unalterable task execution certificate. The computer program 32 can be divided into one or more modules / units, one or more of which are stored in the memory 31 and executed by the processor 30 to complete this application. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 32 in the electronic device 3.For example, computer program 32 can be divided into the functions of a distribution module 201, a first processing module 202, a model building module 203, a navigation module 204, a verification module 205, a second processing module 206, an operation module 207, and a feedback module 208 (a module in the virtual device). The specific functions of each module are as follows: Distribution module 201 is used to synchronously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through the cloud platform; First processing module 202 is used for the robot transfer base to receive the transfer task set, generate an initial navigation path based on a reinforcement learning algorithm, and provide feedback on the resource scheduling status to the cloud platform; Model building module 203 is used for the robot transfer base to move along the initial navigation path. During the movement, it performs 3D semantic modeling of unstructured roads by fusing perception data from LiDAR and multispectral cameras, and simultaneously constructs a model based on visual data, LiDAR point clouds, and broadcast information. The system comprises: a self-centered dynamic traffic flow field model; a navigation module 204 for navigating the robot transfer base to the task starting point based on the dynamic traffic flow field model; a verification module 205 for verifying the identity of the target robot at the task starting point using its onboard positioning module and image acquisition device; a second processing module 206 for performing an adaptive loading operation based on force control feedback if the identity verification is successful, and dynamically generating an energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model after loading is completed; an operation module 207 for autonomously performing unloading operations after reaching the destination by recognizing the unloading area features of the destination through 3D scene reconstruction; and a feedback module 208 for calculating the return path and sending a resource release signal to the cloud platform after unloading is completed, while simultaneously uploading the entire operation data to the cloud platform to generate an unalterable task execution certificate.

[0093] Electronic device 3 may include, but is not limited to, processor 30 and memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic devices may also include input / output devices, network access devices, buses, etc.

[0094] The processor 30 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0095] The memory 31 can be an internal storage unit of the electronic device 3, such as a hard disk or RAM. The memory 31 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 31 can include both internal and external storage units of the electronic device 3. The memory 31 is used to store computer programs and other programs and data required by the electronic device. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0096] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed. That is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above-described device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0097] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0098] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0099] In the embodiments provided in this application, it should be understood that the disclosed apparatus / device and method can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules or 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 device, 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; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0100] 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.

[0101] 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.

[0102] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program for the autonomous transfer method of the robot can be stored in a storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments, that is, synchronously distributing a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through a cloud platform; the robot transfer base receiving the transfer task set and generating an initial navigation path based on a reinforcement learning algorithm and feeding back the resource scheduling status to the cloud platform; the robot transfer base moving along the initial navigation path, and during the movement, performing three-dimensional semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras, and simultaneously processing visual data and LiDAR data. Point cloud data and broadcast information are used to construct a dynamic traffic flow field model centered on the robot. The robot's transfer platform navigates to the task starting point based on this model. At the starting point, the platform verifies the target robot's identity using its onboard positioning module and image acquisition device. If the identity verification is successful, the platform performs an adaptive loading operation based on force control feedback. After loading, it dynamically generates an energy-optimal path to the destination based on the real-time updated traffic flow field model. Upon reaching the destination, the platform identifies the unloading area features through 3D scene reconstruction and autonomously performs the unloading operation. After unloading, the platform calculates the return path and sends a resource release signal to the cloud platform, simultaneously uploading all operational data to generate an unalterable task execution certificate. The computer program includes computer program code, which can be in source code form, object code form, executable files, or some intermediate form. Storage media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the contents of storage media can be appropriately added to or removed according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, storage media may not include electrical carrier signals and telecommunication signals.

[0103] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application. The specific embodiments described above further illustrate the purpose, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the protection scope of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this invention.

Claims

1. A method for autonomous transfer of a robot, characterized in that, The method includes: The cloud platform synchronously distributes priority-based transfer task sets to multiple distributed robot transfer bases. The robot transfer base receives the transfer task set, generates an initial navigation path based on a reinforcement learning algorithm, and feeds back the resource scheduling status to the cloud platform. The robot's transport base moves along the initial navigation path. During the movement, it performs 3D semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. Simultaneously, it constructs a dynamic traffic flow model centered on itself by processing visual data, LiDAR point clouds, and broadcast information. This construction of the dynamic traffic flow model includes: spatiotemporally aligning and fusing lane lines and traffic signs identified by visual data, obstacle outlines and positions clustered from LiDAR point clouds, and trajectory prediction results of other traffic participants from V2X broadcast information to generate a dynamic grid map containing vector travel direction, velocity field, and obstacle probability distribution; cross-validating temporary traffic signs identified in the dynamic traffic flow model with traffic control information received through the 5G-V2X communication module; if a conflict is found between the temporary traffic signs and the traffic control information, the traffic control information is prioritized to update the dynamic traffic flow model, and a local path is replanned to avoid entering the controlled area. The robot transfer base navigates to the task starting point based on the dynamic traffic flow field model; At the starting point of the task, the robot transfer base verifies the identity of the target robot through its onboard positioning module and image acquisition device; If the identity verification is successful, the robot transfer base performs an adaptive loading operation based on force control feedback. After loading is completed, it dynamically generates an energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model. Upon arrival at the destination, the robot transfer base identifies the unloading area features of the destination through 3D scene reconstruction and autonomously performs the unloading operation. After unloading is completed, the robot transfer base calculates the return path and sends a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

2. The autonomous transfer method for a robot according to claim 1, characterized in that, The image acquisition device is a multispectral camera, and the robot transfer base verifies the identity of the target robot through its onboard positioning module and the image acquisition device, including: The robot transfer base obtains the distance and angle information of the target robot through its onboard UWB positioning module to roughly locate the area where the target robot is located. The robot transfer base activates the multispectral camera to scan the area where the target robot is located, and identifies specific visual identifiers on the target robot through a deep learning model; The robot transfer base matches the visual recognition results with UWB positioning data and the identity authentication information pre-stored in the transfer task set. Once the match is successful, the identity is confirmed.

3. The autonomous transfer method for a robot according to claim 1, characterized in that, The step of dynamically generating the energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model includes: using the velocity field and slope information in the dynamic traffic flow field model as input, and the current load weight and power as constraints, and using an energy prediction algorithm based on convex optimization to calculate a path with the lowest total energy consumption.

4. The autonomous transfer method for a robot according to claim 1, characterized in that, The robotic transfer base identifies the unloading area features of the destination through 3D scene reconstruction and autonomously performs unloading operations, including: Within the predetermined range of the destination, the lidar is controlled to perform a circular scan, and a three-dimensional dense map of the unloading area is generated by stitching together point clouds using the ICP algorithm. Extract ground flatness, available space size, and typical landmark features from the three-dimensional dense map; Based on the aforementioned three-dimensional dense map, the safe descent height of the forks and the unloading path are calculated. If an unknown obstacle is detected encroaching on the unloading area during the descent, the descent will be immediately paused and the unloading path will be replanned.

5. The autonomous transfer method for a robot according to claim 1, characterized in that, The process of uploading all runtime data to the cloud platform to generate an unalterable task execution certificate includes: The cloud platform will compare the received full-process operation data with the requirements of the task set and generate a comprehensive score that includes task completion, time efficiency, and energy efficiency. The comprehensive score and the hash value of the key perception data are written together into a blockchain smart contract, and the smart contract generates the task execution certificate. When the robot transfer base receives the next transfer task set, it queries the blockchain for its own historical task execution certificates and reports the comprehensive score in the certificate as a credit indicator to the cloud platform to participate in the decision-making of new task priorities.

6. The autonomous transfer method for a robot according to claim 1, characterized in that, When the robot transfer base is started, the method also performs the following initialization steps: Download a lightweight neural network model through the cloud platform; In subsequent operation, the collected local perception data will be used for online incremental learning of the lightweight neural network model, and the updated model parameters will be uploaded to the cloud platform for federated learning aggregation to continuously optimize the perception capabilities of all robots.

7. An autonomous transfer device for a robot, characterized in that, The device includes: The distribution module is used to synchronously distribute a set of transfer tasks with priority identifiers to multiple robot transfer bases deployed in a distributed manner through the cloud platform; The first processing module is used to receive the transfer task set from the robot transfer base, generate an initial navigation path based on a reinforcement learning algorithm, and report the resource scheduling status to the cloud platform. The model building module is used for the robot's transfer base to move along the initial navigation path. During the movement, it performs 3D semantic modeling of the unstructured road by fusing perception data from LiDAR and multispectral cameras. Simultaneously, it constructs a dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds, and broadcast information. The construction of the dynamic traffic flow field model centered on itself by processing visual data, LiDAR point clouds, and broadcast information includes: spatiotemporally aligning and fusing lane lines and traffic signs identified by visual data, obstacle outlines and positions clustered by LiDAR point clouds, and trajectory prediction results of other traffic participants in V2X broadcast information to generate a dynamic grid map containing vector travel direction, velocity field, and obstacle probability distribution; cross-validating temporary traffic signs identified in the dynamic traffic flow field model with traffic control information received through the 5G-V2X communication module; if a conflict is found between the temporary traffic signs and the traffic control information, the traffic control information is prioritized to update the dynamic traffic flow field model, and a local path is replanned to avoid entering the controlled area. The navigation module is used to navigate the robot transfer base to the task starting point based on the dynamic traffic flow field model; The verification module is used at the starting point of the task to verify the identity of the target robot through its positioning module and image acquisition device. The second processing module is used to perform an adaptive loading operation based on force control feedback on the robot transfer base if the identity verification is successful. After the loading is completed, the module dynamically generates the energy-optimal path to the destination based on the real-time updated dynamic traffic flow field model. The operation module is used to enable the robot transfer base to autonomously perform unloading operations after reaching the destination by recognizing the unloading area features of the destination through three-dimensional scene reconstruction. The feedback module is used to calculate the return path of the robot transfer base after unloading and send a resource release signal to the cloud platform. At the same time, it uploads the entire operation data to the cloud platform to generate an unalterable task execution certificate.

8. An electronic device, the device 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 steps of the method as described in any one of claims 1 to 6.

9. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 6.

Images