Fork type mobile robot system based on multi-sensor fusion SLAM (Simultaneous Localization and Mapping) technology

Abstract

The invention relates to the technical field of fork type mobile robots, in particular to a fork type mobile robot system based on multi-sensor fusion SLAM (Simultaneous Localization and Mapping) technology, which comprises a fork type mobile robot body, an autonomous navigation module, a pedestrian detection and obstacle avoidance module, an intelligent scheduling optimization module and a self-adaptive sensor data processing module. High-precision positioning and map building are achieved in a complex environment, autonomous navigation is achieved, the fork type mobile robot body can accurately recognize pedestrians and make avoidance in time in the complex environment through the pedestrian detection and obstacle avoidance module, and through arrangement of the intelligent scheduling optimization module, the safety of the robot is improved. According to the invention, the system enables the fork-type mobile robot body to plan an optimal path in a real-time changing environment and avoid dynamic obstacles, and employs the above modules to enable the fork-type mobile robot to accurately recognize pedestrians and dynamic obstacles in a complex environment, and to avoid the pedestrians and dynamic obstacles.

Description

Technical Field

[0001] The present invention relates to the technical field of fork-type mobile robots, and in particular to a fork-type mobile robot system using multi-sensor fusion SLAM technology. Background Art

[0002] Pallets are widely used in modern intelligent manufacturing or logistics warehousing to carry goods, and forklifts lift and move pallets to carry goods. Traditional forklifts are manually operated, and operators are prone to affect work efficiency and quality due to subjective and objective factors such as emotions or physiology due to long-term repetitive and high-intensity work.

[0003] To solve the above problems, the existing technology uses fork-type mobile robots to transport goods. Fork-type mobile robots integrate forklift technology and AGV technology. Compared with ordinary AGVs, in addition to completing point-to-point material handling, they can also achieve logistics and transportation tasks that connect multiple production links. They are not only good at three major scenarios: high-rise warehouses, off-site receiving areas, and production line transfers, but also have an irreplaceable role in heavy loads, special handling and other scenarios. Through the application of fork-type mobile robots, problems such as large logistics volume and high labor intensity of manual handling in industrial production and warehousing logistics operations can be solved.

[0004] However, existing fork-type mobile robots are unable to accurately identify pedestrians and dynamic obstacles in complex environments and avoid them, which has great limitations. Summary of the invention

[0005] The purpose of the present invention is to provide a fork-type mobile robot system with multi-sensor fusion SLAM technology to solve the problem that the existing fork-type mobile robots are unable to accurately identify pedestrians and dynamic obstacles in complex environments and avoid pedestrians and dynamic obstacles, which has great limitations.

[0006] To achieve the above-mentioned object, the present invention provides a fork-type mobile robot system with multi-sensor fusion SLAM technology, wherein the fork-type mobile robot system with multi-sensor fusion SLAM technology comprises a fork-type mobile robot body, an autonomous navigation module, a pedestrian detection and obstacle avoidance module, an intelligent scheduling optimization module and an adaptive sensor data processing module;

[0007] The autonomous navigation module is used to achieve high-precision positioning and map construction in complex environments through SLAM technology, thereby achieving autonomous navigation of the fork-type mobile robot body;

[0008] The pedestrian detection and obstacle avoidance module is used to accurately identify pedestrians and make timely avoidance in complex environments;

[0009] The intelligent scheduling optimization module is used to plan the optimal path for the movement of the fork-type mobile robot body in a real-time changing environment and avoid dynamic obstacles through an adaptive genetic algorithm;

[0010] The adaptive sensor data processing module is used to improve the stability and accuracy of sensor data processing in complex environments.

[0011] Wherein, the autonomous navigation module includes an algorithm optimization submodule, a visual processing submodule and a multimodal data fusion submodule;

[0012] The algorithm optimization submodule is used to solve the positioning and mapping problems of SLAM technology in complex outdoor environments through an adaptive filtering algorithm based on particle filtering;

[0013] The visual processing submodule is used to improve the processing speed and accuracy of SLAM technology in complex environments by fusing lightweight neural network ECMobileNet and lidar data;

[0014] The multimodal data fusion submodule is used to realize SLAM technology through the fusion of multimodal data in cooperation with IMU and GPS data, so that it can continue to provide reliable location information and map updates when a single sensor data is lost.

[0015] The pedestrian detection and obstacle avoidance module includes a pedestrian trajectory prediction submodule, a pedestrian positioning submodule and a pedestrian area center calculation submodule;

[0016] The pedestrian trajectory prediction submodule is used to improve the prediction ability of the fork-type mobile robot body on pedestrian behavior in a complex environment through a trajectory prediction algorithm based on Social-STGCNN;

[0017] The pedestrian positioning submodule is used to realize dynamic tracking and obstacle avoidance of pedestrian positions through binocular vision technology;

[0018] The pedestrian area center calculation submodule is used to calculate the pedestrian area center through the Yolov8 pedestrian detection algorithm improved based on GhostNet and combined with the target detection and segmentation method to improve the accuracy and real-time performance of dynamic obstacle avoidance.

[0019] The intelligent scheduling optimization module includes an evolutionary decay factor optimization submodule, an adaptive crossover and mutation strategy submodule, an evolutionary reversal submodule, a multi-level search submodule, a multi-objective optimization submodule, an intelligent scheduling submodule and a path optimization submodule;

[0020] The evolutionary decay factor optimization submodule is used to introduce a dynamic evolutionary decay factor to improve the performance of the genetic algorithm in large-scale and complex scenarios;

[0021] The adaptive crossover and mutation strategy submodule is used in the genetic algorithm to dynamically adjust the crossover and mutation probabilities according to the population fitness distribution, thereby improving the adaptability of the algorithm at different stages;

[0022] The evolution reversal submodule is used to activate the evolution reversal mechanism when the genetic algorithm detects that the fitness difference of the population decreases and falls into a local optimal solution;

[0023] The multi-level search submodule is used to divide the search process into a low-level coarse-grained search and a high-level refined search;

[0024] The multi-objective optimization submodule is used to optimize path planning and task scheduling through a Pareto optimal solution model based on a multi-objective genetic algorithm;

[0025] The intelligent scheduling submodule is used to combine the reinforcement learning technology with the adaptive genetic algorithm, so that the fork-type mobile robot body can quickly adapt to changes in the external environment;

[0026] The path optimization submodule is used to improve the efficiency and accuracy of path planning through a dynamic weight adjustment mechanism of turning costs based on the A* algorithm.

[0027] Wherein, the multi-objective optimization submodule includes a load balancing unit, an energy consumption optimization unit and a dynamic obstacle avoidance unit;

[0028] The load balancing unit is used to dynamically adjust the task priority of each forklift according to the current load condition of each forklift;

[0029] The energy consumption optimization unit is used to comprehensively calculate the energy consumption model of each feasible path according to the task path length, turning frequency, speed and current state of the forklift;

[0030] The dynamic obstacle avoidance unit is used to detect conflicts on the paths of each forklift in real time and re-plan the paths when dynamic obstacles appear in the environment.

[0031] Wherein, the adaptive sensor data processing module includes an environment perception submodule, a dynamic data fusion submodule, and a multi-level environment modeling submodule;

[0032] The environmental perception submodule dynamically adjusts the filtering parameters according to changes in environmental conditions through an adaptive filtering algorithm based on deep learning;

[0033] The dynamic data fusion submodule is used to monitor the quality of sensor data in real time and dynamically adjust the weight of each sensor according to the reliability of the data;

[0034] The multi-level environment modeling submodule is used to divide the environment into three categories: static background, dynamic objects and short-term changes, and adopt different processing methods for different types of data.

[0035] A fork-type mobile robot system of the present invention that integrates multi-sensor fusion SLAM technology includes a fork-type mobile robot body, an autonomous navigation module, a pedestrian detection and obstacle avoidance module, an intelligent scheduling optimization module and an adaptive sensor data processing module. Through the setting of the autonomous navigation module, high-precision positioning and map construction are achieved in complex environments based on SLAM technology, thereby achieving autonomous navigation. Through the pedestrian detection and obstacle avoidance module, the fork-type mobile robot body can accurately identify pedestrians and make avoidance in a complex environment. Through the setting of the intelligent scheduling optimization module, the fork-type mobile robot body can plan the optimal path for the movement of the fork-type mobile robot body in a real-time changing environment based on an adaptive genetic algorithm, and avoid dynamic obstacles. The above modules are used to enable the fork-type mobile robot to accurately identify pedestrians and dynamic obstacles in a complex environment, and avoid them. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 It is a principle block diagram of the fork-type mobile robot system of the multi-sensor fusion SLAM technology provided by the present invention.

[0038] Figure 2 It is a specific principle block diagram of the intelligent scheduling optimization module provided by the present invention.

[0039] Figure 3 It is a structural diagram of the Social-STGCNN model in the pedestrian trajectory prediction submodule provided by the present invention.

[0040] Figure 4 It is the technical roadmap of the pedestrian positioning submodule provided by the present invention.

[0041] Figure 5 This is a diagram showing the difference between the feature map generated by the Ghost module in the pedestrian area center calculation submodule provided by the present invention and the feature map generated by the ordinary convolutional network.

[0042] Figure 6 It is a pedestrian feature map when the pedestrian area center calculation submodule provided by the present invention is actually applied.

[0043] Figure 7 It is a flow chart of the improved A* algorithm that introduces turning cost value into the path optimization submodule provided by the present invention.

[0044] 101-fork-type mobile robot body, 102-autonomous navigation module, 103-pedestrian detection and obstacle avoidance module, 104-intelligent scheduling optimization module, 105-adaptive sensor data processing module, 106-algorithm optimization submodule, 107-visual processing submodule, 108-multimodal data fusion submodule, 109-pedestrian trajectory prediction submodule, 110-pedestrian positioning submodule, 111-pedestrian area center calculation submodule, 112-evolutionary decay factor optimization submodule, 113-adaptive crossover and mutation strategy submodule, 114-evolutionary reversal submodule, 115-multi-level search submodule, 116-multi-objective optimization submodule, 117-intelligent scheduling submodule, 118-path optimization submodule, 119-load balancing unit, 120-energy consumption optimization unit, 121-dynamic obstacle avoidance unit, 122-environmental perception submodule, 123-dynamic data fusion submodule, 124-multi-level environmental modeling submodule. DETAILED DESCRIPTION

[0045] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0046] See also Figures 1 to 7 The present invention provides a fork-type mobile robot system with multi-sensor fusion SLAM technology, the fork-type mobile robot system with multi-sensor fusion SLAM technology comprises a fork-type mobile robot body 101, an autonomous navigation module 102, a pedestrian detection and obstacle avoidance module 103, an intelligent scheduling optimization module 104 and an adaptive sensor data processing module 105;

[0047] The autonomous navigation module 102 is used to achieve high-precision positioning and map construction in complex environments through SLAM technology, thereby achieving autonomous navigation of the fork-type mobile robot body 101;

[0048] The pedestrian detection and obstacle avoidance module 103 is used to accurately identify pedestrians and make timely avoidance in complex environments;

[0049] The intelligent scheduling optimization module 104 is used to plan the optimal path for the movement of the fork-type mobile robot body 101 in a real-time changing environment and avoid dynamic obstacles through an adaptive genetic algorithm;

[0050] The adaptive sensor data processing module 105 is used to improve the stability and accuracy of sensor data processing in complex environments.

[0051] In this embodiment, through the setting of the autonomous navigation module 102, high-precision positioning and map construction are achieved in complex environments based on SLAM technology, and then autonomous navigation is achieved. Through the pedestrian detection and obstacle avoidance module 103, the fork-type mobile robot body 101 can accurately identify pedestrians in complex environments and make avoidances in time. Through the setting of the intelligent scheduling optimization module 104, the fork-type mobile robot body 101 can plan the optimal path for the movement of the fork-type mobile robot body 101 in a real-time changing environment based on an adaptive genetic algorithm, and avoid dynamic obstacles. The above modules enable the fork-type mobile robot to accurately identify pedestrians and dynamic obstacles in a complex environment, and avoid pedestrians and dynamic obstacles.

[0052] Furthermore, the autonomous navigation module 102 includes an algorithm optimization submodule 106, a visual processing submodule 107 and a multimodal data fusion submodule 108;

[0053] The algorithm optimization submodule 106 is used to solve the positioning and mapping problems of SLAM technology in complex outdoor environments through an adaptive filtering algorithm based on particle filtering;

[0054] The visual processing submodule 107 is used to improve the processing speed and accuracy of SLAM technology in complex environments by fusing the lightweight neural network ECMobileNet and laser radar data;

[0055] The multimodal data fusion submodule 108 is used to realize the SLAM technology through the fusion of multimodal data in cooperation with IMU and GPS data, so that it can continue to provide reliable location information and map updates when a single sensor data is lost.

[0056] In this embodiment, the algorithm optimization submodule 106 introduces and optimizes an adaptive filtering algorithm based on particle filtering to solve the positioning and mapping problems of the SLAM system in complex outdoor environments. The visual processing submodule 107 introduces a lightweight neural network ECMobileNet and lidar data fusion to improve the processing speed and accuracy of SLAM technology in complex environments. The lightweight neural network ECMobileNet has a small number of parameters and low computing requirements, and is suitable for running on resource-constrained hardware platforms to ensure the balance between real-time and energy efficiency of the system. The multimodal data fusion submodule 108 combines IMU (inertial measurement unit) and GPS data in addition to traditional vision and lidar sensors. Through the fusion of multimodal data, the system can continue to provide reliable location information and map updates when a single sensor data is lost. This multimodal fusion technology performs well in harsh environments and ensures the reliability of unmanned forklifts under extreme conditions.

[0057] Furthermore, the pedestrian detection and obstacle avoidance module 103 includes a pedestrian trajectory prediction submodule 109, a pedestrian positioning submodule 110 and a pedestrian area center calculation submodule 111;

[0058] The pedestrian trajectory prediction submodule 109 is used to improve the prediction ability of the fork-type mobile robot body 101 on pedestrian behavior in a complex environment through a trajectory prediction algorithm based on Social-STGCNN;

[0059] The pedestrian positioning submodule 110 is used to realize dynamic tracking and obstacle avoidance of pedestrian positions through binocular vision technology;

[0060] The pedestrian region center calculation submodule 111 is used to calculate the pedestrian region center by using the Yolov8 pedestrian detection algorithm improved based on GhostNet and combining the target detection and segmentation method to improve the accuracy and real-time performance of dynamic obstacle avoidance.

[0061] In this embodiment, the pedestrian trajectory prediction submodule 109 is based on the trajectory prediction algorithm of Social-STGCNN, and the pedestrian trajectory is modeled by the spatiotemporal graph convolutional neural network. First, the model uses the convolutional neural network to extract the spatiotemporal characteristics of pedestrians, and expands the two-dimensional pedestrian interaction graph to a three-dimensional space with a time dimension through the spatiotemporal graph convolution. Then, the time extrapolator is used to further capture the dynamic behavior of pedestrians at different times, and generate the future trajectory prediction of pedestrians, so that the trajectory prediction algorithm based on Social-STGCNN can significantly improve the unmanned forklift's ability to predict pedestrian behavior in complex environments, so as to perform obstacle avoidance operations more accurately.

[0062] The pedestrian positioning submodule 110 uses two cameras to capture images of the same scene at different angles, and uses parallax calculation to estimate the depth of objects in the scene. Specifically, first, a reference coordinate system is set and a coordinate system conversion is performed to ensure that the data of the binocular camera is consistent with the coordinate system of the forklift. Then, the precise position of the pedestrian relative to the forklift is calculated by using a binocular vision ranging algorithm combined with the calibration parameters of the binocular camera. Finally, based on the motion model of the forklift, a forklift travel coordinate system is established to achieve dynamic tracking of the pedestrian position and obstacle avoidance operations.

[0063] The pedestrian area center calculation submodule 111 is based on the Yolov8 pedestrian detection algorithm improved by GhostNet, and combines the target detection and segmentation method to calculate the pedestrian area center, wherein GhostNet is a lightweight neural network structure, which reduces the computational complexity of the model by introducing lightweight convolution operations while maintaining the detection accuracy. GhostNet is integrated into the Yolov8 model, and the model is used to perform pedestrian detection and segmentation. Subsequently, the detected pedestrian area is accurately segmented, the center position of the pedestrian is calculated, and the accuracy and real-time performance of dynamic obstacle avoidance are improved, so that the system can better understand the position and behavior of pedestrians in the environment, thereby making more reasonable obstacle avoidance decisions.

[0064] Further, the intelligent scheduling optimization module 104 includes an evolutionary decay factor optimization submodule 112, an adaptive crossover and mutation strategy submodule 113, an evolutionary reversal submodule 114, a multi-level search submodule 115, a multi-objective optimization submodule 116, an intelligent scheduling submodule 117 and a path optimization submodule 118;

[0065] The evolutionary decay factor optimization submodule 112 is used to introduce a dynamic evolutionary decay factor to improve the performance of the genetic algorithm in large-scale and complex scenarios;

[0066] The adaptive crossover and mutation strategy submodule 113 is used in the genetic algorithm to dynamically adjust the crossover and mutation probabilities according to the population fitness distribution, thereby improving the adaptability of the algorithm at different stages;

[0067] The evolution reversal submodule 114 is used to activate the evolution reversal mechanism when the genetic algorithm detects that the fitness difference of the population decreases and falls into a local optimal solution;

[0068] The multi-level search submodule 115 is used to divide the search process into a low-level coarse-grained search and a high-level refined search;

[0069] The multi-objective optimization submodule 116 is used to optimize path planning and task scheduling through a Pareto optimal solution model based on a multi-objective genetic algorithm;

[0070] The intelligent scheduling submodule 117 is used to combine the reinforcement learning technology with the adaptive genetic algorithm, so that the fork-type mobile robot body 101 can quickly adapt to changes in the external environment;

[0071] The path optimization submodule 118 is used to improve the efficiency and accuracy of path planning through a dynamic weight adjustment mechanism of turning costs based on the A* algorithm.

[0072] In this embodiment, the evolutionary decay factor optimization submodule 112 introduces a dynamic evolutionary decay factor, which dynamically adjusts the crossover rate and mutation rate in real time according to the evolutionary stage and fitness distribution of the population. This adaptive adjustment based on the evolutionary process not only significantly improves the performance of the genetic algorithm in large-scale and complex scenarios, but also avoids the problem of falling into a local optimal solution too early, thereby ensuring that the algorithm can find the global optimal solution;

[0073] The adaptive crossover and mutation strategy submodule 113 sets a high crossover rate (0.7 to 0.9) in the early stage of the algorithm because of the high population diversity and the need for a large-scale global search. As the evolution proceeds, the fitness of the individuals gradually becomes similar, and the probability of the crossover operation decreases (0.4 to 0.6) to retain the genetic information of excellent individuals and reduce the interference caused by randomness. In order to improve the search capability in the local area, the individuals with higher fitness are locally disturbed to produce small mutations. Through fine search for a better solution, it is helpful to jump out of the local optimal area and further improve the possibility of the global optimal solution.

[0074] The evolution reversal submodule 114 is used to activate the evolution reversal mechanism when the algorithm detects that the fitness difference of the population decreases and falls into a local optimal solution. The mechanism will randomly select some individuals for gene reversal operation, that is, re-randomize some chromosomes of some individuals. By introducing new gene diversity, the convergence state of the population is broken and the possibility of global search is reopened;

[0075] The multi-level search submodule 115 divides the search process into low-level coarse-grained search and high-level refined search. In the early stage, a coarse-grained search with a larger step size is used to quickly lock the area where the global optimal solution is located; in the later stage, a refined search with a smaller step size is used to deeply explore the local optimal solution in the area, thereby ensuring the global search efficiency while further optimizing the quality of the local solution.

[0076] The multi-objective optimization submodule 116 includes a load balancing unit 119, an energy consumption optimization unit 120 and a dynamic obstacle avoidance unit 121. The load balancing unit 119 is used to dynamically adjust the task priority of each forklift according to the current load condition of each forklift. The energy consumption optimization unit 120 is used to comprehensively calculate the energy consumption model of each feasible path according to the task path length, turning frequency, speed and the current state of the forklift. The dynamic obstacle avoidance unit 121 is used to detect conflicts on the paths of each forklift in real time and replan the paths when dynamic obstacles appear in the environment. By introducing the multi-objective optimization algorithm, the system can achieve efficient compromises among multiple objectives such as task allocation, resource scheduling and path planning.

[0077] The intelligent scheduling submodule 117 combines reinforcement learning technology with an adaptive genetic algorithm. Through the feedback mechanism of reinforcement learning, the system can learn environmental changes in real time and adjust the parameters of the genetic algorithm. The reinforcement learning model continuously obtains feedback from the environment, identifies emergencies such as obstacles and path blockages, and dynamically adjusts the crossover rate and mutation of the genetic algorithm according to actual conditions.

[0078] The path optimization submodule 118 introduces turning cost into the cost function of the A* algorithm to address the time cost caused by frequent turning of forklifts in complex environments, dynamically adjusts weights to reduce unnecessary turns, thereby optimizing path length and time. Combined with the global optimization capability of the genetic algorithm, the system can plan a smoother path in complex scenarios. Experimental results show that the improved algorithm reduces the path running time by about 10% and significantly improves the real-time performance of the system.

[0079] Further, the adaptive sensor data processing module 105 includes an environment perception submodule 122, a dynamic data fusion submodule 123, and a multi-level environment modeling submodule 124;

[0080] The environment perception submodule 122 dynamically adjusts the filtering parameters according to the changes in environmental conditions through an adaptive filtering algorithm based on deep learning;

[0081] The dynamic data fusion submodule 123 is used to monitor the quality of sensor data in real time and dynamically adjust the weight of each sensor according to the reliability of the data;

[0082] The multi-level environment modeling submodule 124 is used to divide the environment into three categories: static background, dynamic objects and short-term changes, and adopt different processing methods for different types of data.

[0083] In this embodiment, the environment perception submodule 122 can dynamically adjust the filtering parameters according to the changes in environmental conditions through an adaptive filtering algorithm based on deep learning. This algorithm can effectively filter out noise data when processing lidar data, ensuring data stability and high accuracy in complex outdoor environments. In particular, when the light and weather change greatly, the adaptive filtering algorithm can automatically adjust the parameters, thereby improving the robustness and accuracy of the SLAM system. This adaptive filtering technology monitors the changes in the external environment in real time and dynamically adjusts the characteristics of the filter by introducing an environment perception module, so that the system can maintain stable performance under different environmental conditions.

[0084] The dynamic data fusion submodule 123 monitors the quality of sensor data in real time and dynamically adjusts the weight of each sensor according to the reliability of the data, thereby ensuring that the SLAM system can always provide high-precision location information and environmental perception. This strategy not only improves the accuracy of the system, but also significantly improves the robustness of SLAM in complex scenes. In addition, by introducing a multi-sensor data fusion model based on Bayesian theory, the credibility of each sensor data is calculated, and the fusion weight is dynamically adjusted to ensure the reliability and stability of the system in a changing environment.

[0085] The multi-level environment modeling submodule 124 divides the environment into three categories: static background, dynamic objects, and short-term changes, and adopts different processing methods for different types of data. The static background uses a long-term memory method to build a stable map, the dynamic objects are processed using a short-term update strategy, and the short-term changes are processed through fast filtering. This multi-level processing method enables the system to respond quickly to environmental changes while maintaining high-precision environmental modeling and data processing capabilities.

[0086] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A fork-type mobile robot system with multi-sensor fusion SLAM technology, characterized in that: It includes a fork-type mobile robot body, an autonomous navigation module, a pedestrian detection and obstacle avoidance module, an intelligent scheduling optimization module, and an adaptive sensor data processing module; The autonomous navigation module is used to achieve high-precision positioning and map construction in complex environments through SLAM technology, thereby achieving autonomous navigation of the fork-type mobile robot body; The pedestrian detection and obstacle avoidance module is used to accurately identify pedestrians and make timely avoidance in complex environments; The intelligent scheduling optimization module is used to plan the optimal path for the movement of the fork-type mobile robot body in a real-time changing environment and avoid dynamic obstacles through an adaptive genetic algorithm; The adaptive sensor data processing module is used to improve the stability and accuracy of sensor data processing in complex environments.

2. The fork-type mobile robot system of multi-sensor fusion SLAM technology as claimed in claim 1, characterized in that: The autonomous navigation module includes an algorithm optimization submodule, a visual processing submodule and a multimodal data fusion submodule; The algorithm optimization submodule is used to solve the positioning and mapping problems of SLAM technology in complex outdoor environments through an adaptive filtering algorithm based on particle filtering; The visual processing submodule is used to improve the processing speed and accuracy of SLAM technology in complex environments by fusing lightweight neural network ECMobileNet and lidar data; The multimodal data fusion submodule is used to realize SLAM technology through the fusion of multimodal data in cooperation with IMU and GPS data, so that it can continue to provide reliable location information and map updates when a single sensor data is lost.

3. The fork-type mobile robot system of multi-sensor fusion SLAM technology as claimed in claim 2, characterized in that: The pedestrian detection and obstacle avoidance module includes a pedestrian trajectory prediction submodule, a pedestrian positioning submodule and a pedestrian area center calculation submodule; The pedestrian trajectory prediction submodule is used to improve the prediction ability of the fork-type mobile robot body on pedestrian behavior in a complex environment through a trajectory prediction algorithm based on Social-STGCNN; The pedestrian positioning submodule is used to realize dynamic tracking and obstacle avoidance of pedestrian positions through binocular vision technology; The pedestrian area center calculation submodule is used to calculate the pedestrian area center through the Yolov8 pedestrian detection algorithm improved based on GhostNet and combined with the target detection and segmentation method to improve the accuracy and real-time performance of dynamic obstacle avoidance.

4. The fork-type mobile robot system of multi-sensor fusion SLAM technology as claimed in claim 3 is characterized in that: The intelligent scheduling optimization module includes an evolutionary decay factor optimization submodule, an adaptive crossover and mutation strategy submodule, an evolutionary reversal submodule, a multi-level search submodule, a multi-objective optimization submodule, an intelligent scheduling submodule and a path optimization submodule; The evolutionary decay factor optimization submodule is used to introduce a dynamic evolutionary decay factor to improve the performance of the genetic algorithm in large-scale and complex scenarios; The adaptive crossover and mutation strategy submodule is used in the genetic algorithm to dynamically adjust the crossover and mutation probabilities according to the population fitness distribution, thereby improving the adaptability of the algorithm at different stages; The evolution reversal submodule is used to activate the evolution reversal mechanism when the genetic algorithm detects that the fitness difference of the population decreases and falls into a local optimal solution; The multi-level search submodule is used to divide the search process into a low-level coarse-grained search and a high-level refined search; The multi-objective optimization submodule is used to optimize path planning and task scheduling through a Pareto optimal solution model based on a multi-objective genetic algorithm; The intelligent scheduling submodule is used to combine the reinforcement learning technology with the adaptive genetic algorithm, so that the fork-type mobile robot body can quickly adapt to changes in the external environment; The path optimization submodule is used to improve the efficiency and accuracy of path planning through a dynamic weight adjustment mechanism of turning costs based on the A* algorithm.

5. The fork-type mobile robot system of multi-sensor fusion SLAM technology as claimed in claim 4, characterized in that: The multi-objective optimization submodule includes a load balancing unit, an energy consumption optimization unit and a dynamic obstacle avoidance unit; The load balancing unit is used to dynamically adjust the task priority of each forklift according to the current load condition of each forklift; The energy consumption optimization unit is used to comprehensively calculate the energy consumption model of each feasible path according to the task path length, turning frequency, speed and current state of the forklift; The dynamic obstacle avoidance unit is used to detect conflicts on the paths of each forklift in real time and re-plan the paths when dynamic obstacles appear in the environment.

6. The fork-type mobile robot system of multi-sensor fusion SLAM technology as claimed in claim 5, characterized in that: The adaptive sensor data processing module includes an environment perception submodule, a dynamic data fusion submodule, and a multi-level environment modeling submodule; The environmental perception submodule dynamically adjusts the filtering parameters according to changes in environmental conditions through an adaptive filtering algorithm based on deep learning; The dynamic data fusion submodule is used to monitor the quality of sensor data in real time and dynamically adjust the weight of each sensor according to the reliability of the data; The multi-level environment modeling submodule is used to divide the environment into three categories: static background, dynamic objects and short-term changes, and adopt different processing methods for different types of data.

Images