Modular and highly customizable robot control system and method

Abstract

The invention discloses a modularized and highly-customizable robot control system and method, and relates to the technical field of robot control, and the system comprises a hardware layer, a communication layer, a software layer and an application layer. Modular design is adopted, flexible combination and expansion can be conducted according to actual requirements, high customizability is achieved, a user can select different hardware modules and software modules to conduct combination and configuration according to the actual requirements so as to meet different application requirements, meanwhile, the user can customize and develop the software modules according to requirements, and the user experience is improved. According to the robot control system, a specific function is achieved, intelligent control over the robot is achieved in combination with the artificial intelligence technology, by introducing the machine learning technology, the deep learning technology and other technologies, the system can autonomously learn and optimize a control strategy, the adaptability and the intelligent level of the robot are improved, and intelligent control over the robot is achieved in combination with the artificial intelligence technology.

Description

Technical Field

[0001] The present invention relates to the field of robot control technology, in particular to a modular and highly customizable robot control system and system. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, robots are increasingly used in various fields. However, existing robot control systems often have the following problems: fixed system architecture, difficult to customize according to different application scenarios; limited decision-making ability, unable to effectively cope with complex and changing environments; inefficient path planning, prone to collisions; insufficient system reliability, high maintenance costs. These problems limit the application effect and scope of robots in complex environments. Summary of the invention

[0003] In view of the problems existing in the existing robot control system, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is how to design a modular, highly customizable robot control system and method, which can realize flexible combination of functional modules, adaptive optimization of intelligent decision-making, real-time planning of dynamic paths, and automatic diagnosis and repair of system faults.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In the first aspect, an embodiment of the present invention provides a modular, highly customizable robot control system, which includes a hardware layer, including a main control module, a multimodal perception module and an intelligent drive module, wherein the main control module is connected to the multimodal perception module and the intelligent drive module through a bus; a software layer, including a distributed control system, a deep learning decision module, a dynamic path planning module and a task scheduling module, wherein the output end of the distributed control system is respectively connected to the input end of the deep learning decision module, the dynamic path planning module and the task scheduling module; a communication layer, including an industrial Ethernet, an adaptive protocol module and a data encryption module, and the industrial Ethernet is used to connect the hardware layer, the software layer and the application layer to realize data transmission; the application layer, including a fault diagnosis module and a self-repair module, which are connected to the distributed control system for monitoring the system status and execution Maintenance operation; wherein, the multimodal perception module collects first environmental data; the distributed control system performs preliminary processing on the first environmental data to generate second environmental data; the deep learning decision module updates the learning parameters according to the second environmental data, and generates a first decision strategy based on the updated learning parameters; the dynamic path planning module constructs an environmental model according to the second environmental data, and calculates the first path based on the environmental model and the first decision strategy; the task scheduling module generates a control instruction sequence according to the first decision strategy and the first path; the intelligent drive module executes the control instruction sequence; the fault diagnosis module monitors the operating status of the main control module, the multimodal perception module, the intelligent drive module, the distributed control system, the deep learning decision module, the dynamic path planning module and the task scheduling module in real time, and generates a fault diagnosis result; the self-repair module receives the fault diagnosis result, and automatically starts the corresponding self-repair program when the fault diagnosis result indicates an abnormality.

[0007] As a preferred solution of the modular and highly customizable robot control system described in the present invention, the workflow of the distributed control system is as follows: receiving the first environmental data collected by the multimodal perception module, and parsing the first environmental data according to a predefined format; time synchronization of the first environmental data based on the timestamp information, and spatial registration of the first environmental data with spatial information; preprocessing the first environmental data after time synchronization and spatial registration; extracting key features from the preprocessed first environmental data, and performing preliminary multimodal data fusion on the extracted key features; constructing and optimizing the environmental representation model based on the fused key features, and compressing and encoding the optimized environmental representation model; evaluating the quality of the compressed and encoded environmental representation model, screening out the first environmental data that exceeds the quality threshold, and using it as the second environmental data; outputting the second environmental data to the deep learning decision module, the dynamic path planning module and the task scheduling module.

[0008] As a preferred solution of the modular, highly customizable robot control system described in the present invention, the workflow of the deep learning decision module is as follows: receiving the second environment data generated by the distributed control system, and preprocessing the received second environment data; extracting features from the preprocessed second environment data, fusing features of different modalities, and generating a comprehensive feature representation; using the current deep learning model parameters, forward propagating the comprehensive feature representation to generate a preliminary decision output; calculating the loss value of the preliminary decision output; based on the loss value of the preliminary decision output, using the back propagation algorithm to update the deep learning model parameters; using the updated deep learning model parameters to infer the current environment state, and generating a first decision strategy based on the inference results; outputting the first decision strategy to the dynamic path planning module and the task scheduling module.

[0009] As a preferred solution of the modular and highly customizable robot control system described in the present invention, the workflow of the dynamic path planning module is as follows: receiving the second environment data transmitted by the distributed control system and the first decision strategy output by the deep learning decision module; constructing an environment model according to the second environment data; predicting the trajectory of the dynamic obstacles identified by the environment model and estimating the position within the first time; comprehensively considering the trajectory prediction results of the static obstacles, terrain passability information and dynamic obstacles identified by the environment model, and generating a cost map reflecting the difficulty of environmental passage; using the first decision strategy to set a framework for path search, the framework includes selecting a path search algorithm; applying the selected path search algorithm to search on the cost map to generate an initial path; using a smoothing algorithm to optimize the initial path; during the path execution process, continuously monitoring environmental changes, if any condition in the dynamic path adjustment trigger set is met, then performing local path replanning and dynamic path adjustment; evaluating the adjusted initial path to verify whether the initial path meets the requirements of the first decision strategy; if it meets the requirements of the first decision strategy, the optimized and evaluated initial path is output as the first path to the task scheduling module and the motion control module.

[0010] As a preferred solution of the modular, highly customizable robot control system described in the present invention, the construction of the environment model includes the following steps: converting the second environment data into a 2D or 3D grid map; performing semantic segmentation on the 2D or 3D grid map, and adding occupancy probability and semantic information to each grid unit; using a computer vision algorithm to identify static obstacles and dynamic obstacles in the environment based on the semantic segmentation results, and classify and label them; based on the grid map and obstacle information, analyzing terrain features, generating terrain passability information, and providing terrain constraints for path planning; based on the grid map, constructing a topological map representation of the environment.

[0011] As a preferred solution of the modular and highly customizable robot control system described in the present invention, the workflow of the task scheduling module is as follows: receiving the first decision strategy output by the deep learning decision module and the first path output by the dynamic path planning module; decomposing the overall task into a set of subtasks according to the first decision strategy; evaluating the available hardware and software resources of the system; dividing the entire task cycle into multiple time windows according to the estimated execution time of the first path; preliminarily sorting the subtasks according to the priority, dependency and resource requirements of the subtasks; dividing the first path into multiple path segments corresponding to the subtasks, matching the sorted subtasks with the path segments, and allocating the required hardware and software resources to each subtask; identifying subtasks that can be executed in parallel, detecting and eliminating potential conflicts between subtasks, and formulating corresponding emergency plans for potential abnormal situations; estimating the execution time of subtasks and generating an execution schedule; verifying the feasibility of the execution schedule through simulation or rapid deduction, and adjusting and optimizing the execution plan according to the verification results; converting the task execution plan into a control instruction sequence, and outputting the control instruction sequence to the intelligent drive module.

[0012] As a preferred solution of the modular, highly customizable robot control system of the present invention, the specific formula of the cost map is as follows:

[0013]

[0014] Among them, C norm (x, y, t) is the normalized cost, σ is the buffer range parameter, k b is the buffer strength coefficient, d(x,y) is the Euclidean distance between the grid cell (x,y) and the nearest obstacle, buffer_value is the buffer contribution value, C buffer (x, y, t) is the final cost value after adding the safety buffer.

[0015] In the second aspect, an embodiment of the present invention provides a modular, highly customizable robot control method, which includes a multimodal perception module collecting first environmental data; a distributed control system performing preliminary processing on the first environmental data to generate second environmental data; a deep learning decision module updating learning parameters according to the second environmental data, and generating a first decision strategy based on the updated learning parameters; a dynamic path planning module constructing an environmental model according to the second environmental data, and calculating a first path based on the environmental model and the first decision strategy; a task scheduling module generating a control instruction sequence according to the first decision strategy and the first path; an intelligent drive module executing the control instruction sequence; a fault diagnosis module monitoring the operating status of a main control module, a multimodal perception module, an intelligent drive module, a distributed control system, a deep learning decision module, a dynamic path planning module and a task scheduling module in real time, and generating a fault diagnosis result; a self-repair module receiving the fault diagnosis result, and automatically starting a corresponding self-repair program when the fault diagnosis result indicates an abnormality.

[0016] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: the computer program instructions, when executed by the processor, implement the steps of the modular, highly customizable robot control system as described in the first aspect of the present invention.

[0017] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: the computer program instructions, when executed by a processor, implement the steps of the modular, highly customizable robot control system as described in the first aspect of the present invention.

[0018] The beneficial effects of the present invention are as follows: the present invention adopts a modular design, can be flexibly combined and expanded according to actual needs, and is highly customizable. Users can select different hardware modules and software modules for combination and configuration according to actual needs to meet different application requirements. At the same time, users can also customize and develop software modules as needed to achieve specific functions. Combined with artificial intelligence technology, intelligent control of robots is achieved. By introducing machine learning, deep learning and other technologies, the system can autonomously learn and optimize control strategies, improve the adaptability and intelligence level of the robot, and combine with artificial intelligence technology to achieve intelligent control of robots. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0020] Figure 1 Module connection diagram for a modular, highly customizable robot control system.

[0021] Figure 2 Software layer structure diagram of a modular and highly customizable robot control system.

[0022] Figure 3 Flowchart of a method for a modular, highly customizable robot control system.

[0023] Figure 4 Computer equipment diagram for a modular, highly customizable robot control system. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] Example 1

[0028] Reference Figure 1 to Figure 4 , is the first embodiment of the present invention, which provides a modular, highly customizable robot control system, including,

[0029] The hardware layer includes a main control module, a multimodal perception module and an intelligent drive module, wherein the main control module is connected to the multimodal perception module and the intelligent drive module through a bus.

[0030] The software layer includes a distributed control system, a deep learning decision module, a dynamic path planning module and a task scheduling module, wherein the output end of the distributed control system is connected to the input end of the deep learning decision module, the dynamic path planning module and the task scheduling module respectively.

[0031] Communication layer, including industrial Ethernet, adaptive protocol module and data encryption module.

[0032] The application layer includes a fault diagnosis module and a self-repair module, which are connected to the distributed control system to monitor the system status and perform maintenance operations.

[0033] Among them, the multimodal perception module collects the first environmental data; the distributed control system performs preliminary processing on the first environmental data to generate the second environmental data; the deep learning decision module updates the learning parameters according to the second environmental data, and generates the first decision strategy based on the updated learning parameters; the dynamic path planning module constructs the environmental model according to the second environmental data, and calculates the first path based on the environmental model and the first decision strategy; the task scheduling module generates a control instruction sequence according to the first decision strategy and the first path; the intelligent drive module executes the control instruction sequence; the fault diagnosis module monitors the operating status of the main control module, the multimodal perception module, the intelligent drive module, the distributed control system, the deep learning decision module, the dynamic path planning module and the task scheduling module in real time, and generates a fault diagnosis result; the self-repair module receives the fault diagnosis result, and when the fault diagnosis result indicates an abnormality, automatically starts the corresponding self-repair program.

[0034] Specifically, the workflow of the distributed control system is as follows: receiving the first environmental data collected by the multimodal perception module, and parsing the first environmental data according to a predefined format; performing time synchronization on the first environmental data based on the timestamp information, and performing spatial registration on the first environmental data with spatial information to ensure the spatiotemporal consistency of the first environmental data; preprocessing the first environmental data after time synchronization and spatial registration, including noise removal and error correction, and the specific formula is as follows:

[0035]

[0036] X c (t i )=f(X n (t i ))=aX n (t i )+b+ε(t i )

[0037] Where N is the window size of the filter, X(t i ) is the i-th timestamp t i The first environmental data collected, X n (t i ) is the first environmental data after noise removal, X c (t i ) is the first environmental data after error correction, ε(t i ) is the correction error, g(·) is the filter function used for noise removal, and f(·) is the error correction function.

[0038] Furthermore, key features are extracted from the preprocessed first environmental data, and preliminary multimodal data fusion is performed on the extracted key features; an environmental representation model is constructed and optimized based on the fused key features, and the environmental representation model includes but is not limited to environmental maps and scene understanding; the optimized environmental representation model is compressed and encoded to reduce storage and transmission burdens; the quality of the compressed and encoded environmental representation model is evaluated, and the first environmental data that exceeds the quality threshold is screened out and used as the second environmental data; the second environmental data is output to the deep learning decision module, the dynamic path planning module, and the task scheduling module.

[0039] Specifically, the workflow of the deep learning decision-making module is as follows: receive the second environment data generated by the distributed control system, and preprocess the received second environment data; extract features from the preprocessed second environment data, fuse features of different modes, and generate a comprehensive feature representation; use the current deep learning model parameters to forward propagate the comprehensive feature representation to generate a preliminary decision output; calculate the loss value of the preliminary decision output; based on the loss value of the preliminary decision output, use the back propagation algorithm to update the deep learning model parameters; use the updated deep learning model parameters to infer the current environment state, and generate a first decision strategy based on the inference results; output the first decision strategy to the dynamic path planning module and the task scheduling module.

[0040] Furthermore, the workflow of the dynamic path planning module is as follows: receiving the second environmental data transmitted by the distributed control system and the first decision strategy output by the deep learning decision module; constructing an environmental model based on the second environmental data; predicting the trajectory of dynamic obstacles (such as moving people or vehicles) identified by the environmental model, and estimating their positions in the future; comprehensively considering the trajectory prediction results of static obstacles, terrain passability information and dynamic obstacles identified by the environmental model, and generating a cost map reflecting the difficulty of environmental passage; using the first decision strategy to set a framework for path search, including determining the starting and end points of path planning, and designing heuristics to guide path search. function, select a path search algorithm; apply the selected path search algorithm to search on the cost map to generate an initial path; use a smoothing algorithm (such as Bezier curve) to optimize the initial path to improve the smoothness and executability of the initial path; during the path execution process, continuously monitor environmental changes, and if any condition in the dynamic path adjustment trigger set is met, perform local path replanning to achieve dynamic path adjustment; evaluate the adjusted initial path to verify whether the initial path meets the requirements of the first decision-making strategy; if it meets the requirements of the first decision-making strategy, output the optimized and evaluated initial path as the first path to the task scheduling module and the motion control module.

[0041] Specifically, the construction of the environment model includes the following steps: converting the second environment data into a 2D or 3D grid map; performing semantic segmentation on the 2D or 3D grid map, and adding occupancy probability and semantic information to each grid unit; using a computer vision algorithm to identify static obstacles and dynamic obstacles in the environment based on the semantic segmentation results, and classify and label them; based on the grid map and obstacle information, analyzing terrain features (such as slope, roughness, etc.), generating terrain passability information, and providing terrain constraints for path planning; based on the grid map, constructing a topological graph representation of the environment, including key nodes and connection relationships.

[0042] Preferably, generating a cost map reflecting the difficulty of navigating the environment includes the following steps: mapping the identified static obstacles onto a grid map, and assigning a static obstacle cost value to each obstacle cell, the calculation formula of the static obstacle cost is:

[0043] C static (x,y)=k s ·O static (x,y)

[0044] Among them, (x, y) is the grid unit, C static (x,y) is the static obstacle cost, k s is the weight coefficient of the static obstacle, O static is the obstacle occupancy function, which is 1 if there is an obstacle at the grid cell (x, y), otherwise it is 0.

[0045] Furthermore, according to the difficulty of passing different terrain types, the terrain passability cost is calculated. The specific formula is:

[0046] C terrain (x,y)=k t T(x,y)

[0047] Among them, C terrain (x, y) is the terrain passability cost, k t is the terrain weight coefficient, T is the terrain passability function, and the return value range can be [0,1], where 0 means completely impassable and 1 means completely passable.

[0048] Furthermore, the trajectory prediction results are used to calculate the space that the dynamic obstacle may occupy in the future time period, and the dynamic obstacle cost is assigned. The specific formula of the dynamic obstacle cost is as follows:

[0049]

[0050] Among them, C dynamic (x, y, t) is the dynamic obstacle cost, k d is the dynamic obstacle weight coefficient, D is the number of dynamic obstacles, Pi (x, y, t) is the probability that the i-th dynamic obstacle occupies the grid cell (x, y) at time t, where t is the time.

[0051] Furthermore, the static obstacle cost, terrain passability cost and dynamic obstacle cost are weighted and fused to generate the total cost C total (x,y,t), the specific formula is as follows:

[0052] C total (x,y,t)=λ s ·C static (x,y)+λ t ·C terrain (x,y)+λ d ·C dynamic (x,y,t)

[0053] Among them, C terrain (x,y) is the terrain cost, C dynamic (x, y, t) is the dynamic obstacle cost, C static (x,y) is the static obstacle cost, λ s , t , d are the fusion weights of static obstacles, terrain and dynamic obstacles respectively.

[0054] Furthermore, the total cost is processed using a Gaussian smoothing function to avoid sudden changes in the cost value. The cost smoothing formula is:

[0055]

[0056] Among them, r is the smoothing radius, G(i,j) is the Gaussian kernel function value, C total (x,y,t) is the total cost,

[0057] C smooth (x,y,t) is the cost after smoothing.

[0058] Furthermore, the smoothed cost value is normalized, and the specific formula is as follows:

[0059]

[0060] Among them, C max , C min are the minimum and maximum values ​​of the smoothed cost, C norm (x,y,t) is the normalized cost.

[0061] Furthermore, for each grid cell (x, y) in the cost map, the Euclidean distance d(x, y) between it and the nearest obstacle and the buffer contribution value are calculated, and the buffer contribution value is added to the original cost. The specific formula is as follows:

[0062]

[0063] Among them, C norm (x, y, t) is the normalized cost, σ is the buffer range parameter (a positive real number, the unit is the same as the distance unit, usually between 0.5 and 2 meters), k b is the buffer strength coefficient (a positive real number, usually between 0.1 and 1), d(x,y) is the Euclidean distance between the grid cell (x,y) and the nearest obstacle, buffer_value is the buffer contribution value, C buffer (x, y, t) is the final cost value after adding the safety buffer.

[0064] It should be noted that when d(x, y) = 0 (i.e. the position is on the obstacle), the buffer contribution value buffer_value reaches the maximum value k b , as d(x,y) increases, buffer_value decays exponentially, and when d(x,y) is much larger than the buffer range parameter σ, buffer_value approaches 0.

[0065] Preferably, this method creates a continuous, smoothly transitioned high-cost region by adding a safety buffer around the obstacle, avoiding mutations in the cost map, and by adjusting the parameter k b and σ, the strength and range of the buffer zone can be flexibly controlled, and the tendency of the path planning algorithm to avoid obstacles is effectively increased, thus improving the safety of the planned path.

[0066] Maintaining the continuity of the cost map helps generate a smoother and more executable path.

[0067] Specifically, the dynamic path adjustment trigger set includes but is not limited to the following situations: a new obstacle enters the planned path; the deviation between the actual trajectory of the original dynamic obstacle and the predicted trajectory exceeds a preset threshold; an obstacle that was originally considered static moves; an emergency occurs in the environment (such as a traffic accident, road closure, etc.) that affects the original path; new sensor data is received, showing that certain areas of the original path become inaccessible; during path execution, the deviation between the actual position and the planned path exceeds the allowable range; the task objectives or constraints change, and the path needs to be replanned.

[0068] Furthermore, the workflow of the task scheduling module is as follows: receiving the first decision strategy output by the deep learning decision module and the first path output by the dynamic path planning module; decomposing the overall task into a set of subtasks according to the first decision strategy, each subtask having a clear goal and priority; evaluating the available hardware and software resources of the system; establishing a time frame for task execution according to the estimated execution time of the first path, and dividing the entire task cycle into multiple time windows; preliminarily sorting the subtasks according to their priorities, dependencies, and resource requirements; dividing the first path into multiple path segments corresponding to the subtasks, matching the sorted subtasks with the path segments, and allocating the required hardware and software resources to each subtask; identifying subtasks that can be executed in parallel, detecting and eliminating potential conflicts between subtasks, and formulating corresponding emergency plans for possible abnormal situations; estimating the execution time of subtasks and generating an execution schedule; verifying the feasibility of the execution schedule through simulation or rapid deduction, and adjusting and optimizing the execution plan according to the verification results; converting the task execution plan into a control instruction sequence, and outputting the control instruction sequence to the intelligent drive module.

[0069] Furthermore, in the communication layer, Industrial Ethernet is used to connect the hardware layer, software layer and application layer to achieve data transmission; the adaptive protocol module is used to dynamically select and configure the communication protocol that best suits the current network environment and task requirements to ensure communication stability and efficiency in different working scenarios; the data encryption module is used to encrypt and decrypt the transmitted data in real time to protect the security of sensitive information and prevent unauthorized access and data leakage.

[0070] Furthermore, in the application layer, the fault diagnosis module is used to monitor the operating status of each component of the system in real time, identify potential faults, and provide diagnostic results; the self-repair module is used to automatically perform repair operations based on the fault diagnosis results, thereby minimizing system downtime and the need for manual intervention.

[0071] Preferably, the design of the fault diagnosis and self-repair module greatly improves the reliability and continuous operation capability of the system, can timely detect and automatically repair potential problems, and minimizes the need for system failures and manual intervention.

[0072] Furthermore, the present embodiment also provides a modular, highly customizable robot control method, including a multimodal perception module collecting first environmental data; a distributed control system performing preliminary processing on the first environmental data to generate second environmental data; a deep learning decision module updating learning parameters according to the second environmental data, and generating a first decision strategy based on the updated learning parameters; a dynamic path planning module constructing an environmental model according to the second environmental data, and calculating a first path based on the environmental model and the first decision strategy; a task scheduling module generating a control instruction sequence according to the first decision strategy and the first path; an intelligent drive module executing the control instruction sequence; a fault diagnosis module monitoring the operating status of a main control module, a multimodal perception module, an intelligent drive module, a distributed control system, a deep learning decision module, a dynamic path planning module and a task scheduling module in real time, and generating a fault diagnosis result; a self-repair module receiving the fault diagnosis result, and automatically starting a corresponding self-repair program when the fault diagnosis result indicates an abnormality.

[0073] This embodiment also provides a computer device suitable for a modular, highly customizable robot control system, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a modular, highly customizable robot control system as proposed in the above embodiment.

[0074] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0075] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, a modular and highly customizable robot control system is implemented as proposed in the above embodiment.

[0076] In summary, the present invention adopts a modular design, which can be flexibly combined and expanded according to actual needs, and is highly customizable. Users can choose different hardware modules and software modules to combine and configure according to actual needs to meet different application requirements. At the same time, users can also customize and develop software modules as needed to achieve specific functions. Combined with artificial intelligence technology, intelligent control of robots is realized. By introducing machine learning, deep learning and other technologies, the system can autonomously learn and optimize control strategies, improve the adaptability and intelligence level of the robot, and combine with artificial intelligence technology to achieve intelligent control of robots.

[0077] Example 2

[0078] Reference Figure 1 to Figure 4 , which is the second embodiment of the present invention, this embodiment provides a modular, highly customizable robot control system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0079] In order to verify the effectiveness of this technical solution, a simulation experiment was conducted in a simulated factory environment. The experimental parameters are as follows: The simulation environment is a 50m x 50m factory workshop, containing 10 static obstacles and 5 dynamic obstacles. The experimental task is to control 5 robots to transport goods from the starting point to the end point while avoiding obstacles. The experiment was conducted 100 times and compared with the traditional A* algorithm path planning solution.

[0080] The experimental process includes the following steps: First, initialize the simulation environment and randomly place static and dynamic obstacles. Then assign a random starting point and end point to each robot. Then use this technical solution and the traditional A* algorithm for path planning and task execution. In each experiment, record the completion time, number of collisions, and path length. Repeat the above steps 100 times and finally calculate the average value.

[0081] The experimental results show that this technical solution is superior to the traditional A* algorithm in multiple key performance indicators. In terms of average completion time, this solution is 127.3 seconds, while the traditional algorithm is 186.5 seconds, an improvement of 31.7%. In terms of the average number of collisions, this solution is only 0.15 times, while the traditional algorithm is 1.87 times, a reduction of 92.0%. In terms of average path length, this solution is 63.8 meters, while the traditional algorithm is 71.2 meters, an optimization of 10.4%.

[0082] The experiment also tested the performance of this technical solution in environments of different complexity. In a low-complexity environment, the average completion time was 98.6 seconds, the average number of collisions was 0.05 times, and the average path length was 58.3 meters. In a medium-complexity environment, these indicators were 127.3 seconds, 0.15 times, and 63.8 meters, respectively. In a high-complexity environment, the indicators became 156.2 seconds, 0.32 times, and 69.7 meters. These data show that as the complexity of the environment increases, the performance of this technical solution has declined, but it still remains within an acceptable range, showing good adaptability.

[0083] These simulation experimental data demonstrate the significant advantages of this technical solution in improving the efficiency, safety and adaptability of robot control systems. Compared with traditional methods, this solution can complete tasks faster, significantly reduce collision risks, and optimize path length, thereby saving energy consumption. At the same time, this solution has shown stable performance in environments of different complexity, proving its good adaptability and robustness.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A modular, highly customizable robot control system, characterized by: include, The hardware layer includes a main control module, a multimodal sensing module and an intelligent driving module, wherein the main control module is connected to the multimodal sensing module and the intelligent driving module via a bus; The software layer includes a distributed control system, a deep learning decision module, a dynamic path planning module and a task scheduling module, wherein the output end of the distributed control system is connected to the input ends of the deep learning decision module, the dynamic path planning module and the task scheduling module respectively; The communication layer includes an industrial Ethernet, an adaptive protocol module and a data encryption module, wherein the industrial Ethernet is used to connect the hardware layer, the software layer and the application layer to realize data transmission; The application layer includes a fault diagnosis module and a self-repair module, which are connected to the distributed control system to monitor the system status and perform maintenance operations; Among them, the multimodal perception module collects the first environmental data; the distributed control system performs preliminary processing on the first environmental data to generate the second environmental data; the deep learning decision module updates the learning parameters according to the second environmental data, and generates the first decision strategy based on the updated learning parameters; the dynamic path planning module constructs the environmental model according to the second environmental data, and calculates the first path based on the environmental model and the first decision strategy; the task scheduling module generates a control instruction sequence according to the first decision strategy and the first path; the intelligent drive module executes the control instruction sequence; the fault diagnosis module monitors the operating status of the main control module, the multimodal perception module, the intelligent drive module, the distributed control system, the deep learning decision module, the dynamic path planning module and the task scheduling module in real time, and generates a fault diagnosis result; the self-repair module receives the fault diagnosis result, and when the fault diagnosis result indicates an abnormality, automatically starts the corresponding self-repair program.

2. The modular, highly customizable robotic control system of claim 1, wherein: The workflow of the distributed control system is as follows: Receiving first environmental data collected by the multimodal perception module, and parsing the first environmental data according to a predefined format; Performing time synchronization on the first environment data based on the timestamp information, and performing spatial registration on the first environment data having spatial information; Preprocessing the first environment data after time synchronization and spatial registration; Extract key features from the preprocessed first environment data, and perform preliminary multimodal data fusion on the extracted key features; Building and optimizing the environment representation model based on the fused key features, and compressing and encoding the optimized environment representation model; evaluating the quality of the compressed and encoded environment representation model, screening out the first environment data exceeding the quality threshold, and using it as the second environment data; The second environment data is output to the deep learning decision module, the dynamic path planning module and the task scheduling module.

3. The modular, highly customizable robotic control system of claim 1, wherein: The workflow of the deep learning decision module is as follows: receiving second environment data generated by the distributed control system, and preprocessing the received second environment data; Extract features from the preprocessed second environment data, fuse features from different modalities, and generate a comprehensive feature representation; Using the current deep learning model parameters, forward propagate the comprehensive feature representation to generate preliminary decision outputs; Calculate the loss value of the initial decision output; Based on the loss value output by the preliminary decision, the back-propagation algorithm is used to update the deep learning model parameters; Using the updated deep learning model parameters, reasoning about the current environment state, and generating a first decision strategy based on the reasoning results; The first decision strategy is output to the dynamic path planning module and the task scheduling module.

4. The modular, highly customizable robotic control system of claim 1, wherein: The workflow of the dynamic path planning module is as follows: Receiving second environment data transmitted by the distributed control system and a first decision strategy output by the deep learning decision module; Build an environment model based on the second environment data; predict the trajectory of the dynamic obstacles identified by the environment model and estimate the position within the first time; Comprehensively consider the static obstacles identified by the environment model, terrain passability information, and trajectory prediction results of dynamic obstacles to generate a cost map that reflects the difficulty of navigating the environment; Using a first decision strategy, setting a framework for path search, the framework including selecting a path search algorithm; Apply the selected path search algorithm to search on the cost map to generate an initial path; Use smoothing algorithm to optimize the initial path; During the path execution process, the environment changes are continuously monitored. If any condition in the dynamic path adjustment trigger set is met, local path replanning is performed and dynamic path adjustment is performed; Evaluate the adjusted initial path to verify whether the initial path meets the requirements of the first decision-making strategy; If the first decision-making strategy requirement is met, the optimized and evaluated initial path is output as the first path to the task scheduling module and the motion control module.

5. The modular, highly customizable robotic control system of claim 4, wherein: The construction of the environment model includes the following steps: Converting the second environment data into a 2D or 3D raster map; Perform semantic segmentation on 2D or 3D grid maps, adding occupancy probability and semantic information to each grid cell; Using computer vision algorithms, static and dynamic obstacles in the environment are identified based on semantic segmentation results, and classified and labeled; Based on the grid map and obstacle information, the terrain features are analyzed, terrain passability information is generated, and terrain constraints are provided for path planning; Based on the raster map, a topological representation of the environment is constructed.

6. The modular, highly customizable robotic control system of claim 1, wherein: The workflow of the task scheduling module is as follows: Receive a first decision strategy output by a deep learning decision module and a first path output by a dynamic path planning module; According to the first decision strategy, the overall task is decomposed into a set of subtasks; Evaluate the hardware and software resources available to the system; According to the estimated execution time of the first path, the entire task cycle is divided into a plurality of time windows; Perform a preliminary sorting of subtasks based on their priorities, dependencies, and resource requirements; Dividing the first path into a plurality of path segments corresponding to the subtasks, matching the sorted subtasks with the path segments, and allocating required hardware and software resources to each subtask; Identify subtasks that can be executed in parallel, detect and eliminate potential conflicts between subtasks, and develop corresponding emergency plans for potential abnormal situations; Estimate subtask execution time and generate execution schedule; Verify the feasibility of the execution plan through simulation or rapid deduction, and adjust and optimize the execution plan based on the verification results; The task execution plan is converted into a control instruction sequence, and the control instruction sequence is output to the intelligent drive module.

7. The modular, highly customizable robotic control system of claim 4, wherein: The specific formula of the cost map is as follows: Among them, C norm (x, y, t) is the normalized cost, σ is the buffer range parameter, k b is the buffer strength coefficient, d(x,y) is the Euclidean distance between the grid cell (x,y) and the nearest obstacle, buffer_value is the buffer contribution value, C buffer (x, y, t) is the final cost value after adding the safety buffer, and t is the time.

8. A modular, highly customizable robot control method, based on the modular, highly customizable robot control system according to any one of claims 1 to 7, characterized in that: Also includes, The multimodal perception module collects first environment data; The distributed control system performs preliminary processing on the first environment data to generate second environment data; The deep learning decision module updates the learning parameters according to the second environment data, and generates a first decision strategy based on the updated learning parameters; The dynamic path planning module constructs an environment model according to the second environment data, and calculates the first path based on the environment model and the first decision strategy; The task scheduling module generates a control instruction sequence according to the first decision strategy and the first path; The intelligent drive module executes the control instruction sequence; The fault diagnosis module monitors the operating status of the main control module, multimodal perception module, intelligent drive module, distributed control system, deep learning decision module, dynamic path planning module and task scheduling module in real time to generate fault diagnosis results; The self-repair module receives the fault diagnosis result, and automatically starts the corresponding self-repair program when the fault diagnosis result indicates that an abnormality exists.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the modular, highly customizable robot control system according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the modular, highly customizable robot control system according to any one of claims 1 to 7 are implemented.

Images