New-generation convenient robot chassis secondary development system based on ROS2.0 system

Abstract

The invention discloses a new-generation convenient robot chassis secondary development system based on an ROS2.0 system, and relates to the technical field of robots. The new-generation convenient robot chassis secondary development system comprises a multi-mode self-sensing self-adaption universal hardware interface module, a multi-mode self-sensing self-adaption universal hardware interface module and a multi-mode self-sensing self-adaption universal hardware interface module, the hierarchical ROS2.0 software architecture module fused with the spatio-temporal semantics comprises a driving layer unit; a functional layer unit; the application layer unit comprises a micro-service architecture subunit based on a knowledge graph; a block chain technology application subunit; the context awareness and evolutionary intelligent scene adaptation configuration management module comprises a context awareness model unit; an evolutionary algorithm unit; the multi-sensor fusion fault diagnosis and redundancy switching module comprises a multi-sensor fusion diagnosis unit; a redundancy switching strategy unit; through the multi-module design, the hardware universality, the software intelligence, the scene adaptability and the system reliability are improved, and the system is suitable for multiple fields of industry, agriculture, medical treatment and the like.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a new generation of convenient robot chassis secondary development system based on ROS2.0 system. Background Art

[0002] As robotics technology continues to develop, traditional robot chassis have exposed many deficiencies when facing complex application scenarios. In the field of industrial logistics, traditional chassis are difficult to adapt to the dynamic changes in the warehouse environment. They are inefficient and unreliable in tasks such as cargo handling and path planning, and are difficult to meet the needs of efficient operation of modern logistics. In smart agriculture scenarios, complex terrain and changing natural environments place high adaptability requirements on robot chassis. Traditional chassis have limited capabilities in sensor adaptation, data processing, and intelligent decision-making, and cannot accurately complete tasks such as crop monitoring and picking. In medical service scenarios, robots need to be highly intelligent and reliable. Traditional chassis have obvious shortcomings in multi-task collaboration, data security, and adaptability to medical environments, and cannot effectively assist medical staff in completing tasks such as drug delivery and material handling.

[0003] In view of this, this application is hereby filed. Summary of the invention

[0004] The purpose of the present invention is to provide a new generation of convenient robot chassis secondary development system based on ROS2.0 system to solve the problems mentioned in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides a new generation of convenient robot chassis secondary development system based on ROS2.0 system, including:

[0006] Multimodal self-sensing adaptive universal hardware interface module: including reconfigurable circuit unit: based on FPGA, it integrates multiple communication interfaces and signal processing modules, can automatically adapt communication protocols and interface parameters according to the connected sensors and actuators, and has built-in MEMS micro-environmental sensors to monitor temperature, humidity and electromagnetic interference in real time, and adjust interface performance accordingly; this reconfigurable circuit unit greatly improves the versatility and adaptability of the robot chassis hardware interface; through the reconfigurable characteristics of FPGA, it can quickly adapt to different types of sensors and actuators, reducing the cost and time of hardware development; at the same time, MEMS micro-environmental sensors monitor environmental factors in real time and adjust interface performance, ensuring the stability and reliability of the hardware interface in complex and changeable industrial environments, and reducing the risk of communication failures and equipment damage caused by environmental factors;

[0007] Hierarchical ROS2.0 software architecture module integrating spatiotemporal semantics: including:

[0008] Driver layer unit: Develop the driver program by combining event-driven and data buffering mechanisms with TSN technology, process data using callback functions, ensure data transmission time synchronization with TSN, and avoid data loss with a circular buffer; This design of the driver layer unit enables efficient and stable transmission of sensor data; The combination of event-driven and callback functions improves the timeliness and response speed of data processing; TSN technology guarantees the time synchronization of data transmission, which is particularly important for sensor data that requires accurate timestamps (such as lidar data), avoiding information errors caused by time asynchronization; The circular buffer effectively avoids data loss, ensures the integrity of system data, and provides a reliable data foundation for subsequent functional and application layers;

[0009] Functional layer unit: Includes motion control and navigation algorithm sub-units: Integrate algorithms based on spatio-temporal semantic maps, fuse lidar and camera data to construct maps containing spatial, temporal dynamics, and semantic information, use the A* algorithm combined with the dynamic window method to plan paths, and adjust trajectories in real-time according to the environment; The motion control and navigation algorithm sub-units based on spatio-temporal semantic maps significantly improve the navigation and motion control capabilities of the robot chassis; Spatio-temporal semantic maps contain richer information, such as spatial layout, temporal dynamic changes, and semantic information, enabling the robot to better understand the surrounding environment; The combination of the A* algorithm and the dynamic window method can perform both global path planning and local real-time trajectory adjustment, effectively avoiding dynamic obstacles, improving the robot's autonomous operation ability and safety in complex industrial logistics scenarios, and reducing the occurrence of collision accidents

[0010] Application layer unit: Build an industrial logistics knowledge graph based on the microservice architecture sub-unit of the knowledge graph, store it in a graph database, query and update through RESTful API interfaces, and call relevant services according to tasks and coordinate the workflow; Includes the blockchain technology application sub-unit: Adopt a consortium blockchain architecture, nodes have unique identity identifiers and public-private key pairs, record task, cargo, and equipment information, and use smart contracts to implement task allocation and settlement; The microservice architecture sub-unit based on the knowledge graph makes information management in industrial logistics more efficient and intelligent; The knowledge graph is stored in the form of a graph database, which can clearly represent the relationships between various entities, and is convenient for querying and updating through RESTful API interfaces, providing accurate information support for task scheduling and service coordination; The blockchain technology application sub-unit enhances the security and credibility of data; The consortium blockchain architecture and node identity identifiers ensure access control of data, and public-private key pair encryption ensures the confidentiality and integrity of data; Smart contracts automate task allocation and settlement, reduce manual intervention, and improve the transparency and efficiency of the logistics process;

[0011] Situation awareness and evolutionary intelligent scenario adaptation configuration management module: Includes

[0012] Situational Awareness Model Unit: A model is constructed based on deep reinforcement learning and transfer learning. After pre-training with DQN, it is transferred to the actual environment for online learning, analyzing sensor data in real time to judge the scenario and predict changes. The situational awareness model unit endows the robot chassis with powerful scenario awareness and prediction capabilities. The combination of deep reinforcement learning and transfer learning enables the model to quickly learn and adapt to different scenarios. The pre-training and online learning mechanisms of the DQN model utilize a large amount of data in the simulation environment for initial learning and can continuously optimize and adjust in the actual environment, accurately judging the current scenario and predicting its change trend in real time, enabling the robot to make corresponding decisions in advance, improving the adaptability and intelligence of the robot in complex and changing environments.

[0013] Evolutionary Algorithm Unit: Adopts a hybrid strategy combining genetic algorithm and particle swarm optimization algorithm to optimize the scenario-

[0014] configuration mapping table and adjusts the chassis configuration parameters according to the new scenario. Through the hybrid optimization strategy, the evolutionary algorithm unit can quickly and effectively optimize the scenario-

[0015] configuration mapping table. The combination of genetic algorithm and particle swarm optimization algorithm gives full play to the advantages of the two algorithms, with higher efficiency and accuracy in searching for the optimal configuration parameters. Automatically adjusting the chassis configuration parameters according to the new scenario enables the robot to better adapt to different working scenarios, improving the performance and efficiency of the robot and reducing the workload and cost of manual configuration adjustment.

[0016] Multi-Sensor Fusion Fault Diagnosis and Redundancy Switching Module: The multi-sensor fusion diagnosis unit fuses the data of lidar, camera, IMU, and ultrasonic sensors based on the evidence theory, preprocesses it to judge the chassis state, and monitors the faults of moving parts. When the redundancy switching strategy unit detects a fault, it switches the function of the faulty module to the standby module according to the preset strategy, records the fault information, and notifies the maintenance personnel. The multi-sensor fusion fault diagnosis and redundancy switching module greatly improves the reliability and maintainability of the robot chassis. The multi-sensor fusion diagnosis unit fuses the data of multiple sensors through the evidence theory, can more accurately judge the chassis state and detect the faults of moving parts, avoiding misjudgments that may occur with a single sensor. The redundancy switching strategy unit can quickly switch the function to the standby module when a fault is detected, ensuring the continuous operation of the robot and reducing the downtime caused by faults. At the same time, recording the fault information and notifying the maintenance personnel facilitates subsequent fault troubleshooting and repair work.

[0017] Furthermore, the reconfigurable circuit unit automatically adapts to the Ethernet UDP communication protocol for lidar, optimizes the data processing flow according to its operating frequency and data volume, adjusts the rotation speed of the cooling fan when the temperature is too high, and activates the anti-interference filtering circuit when the electromagnetic interference increases; this specific adaptation and optimization mechanism for lidar ensures efficient communication and stable operation between the lidar and the robot chassis; automatically adapting to the Ethernet UDP communication protocol enables the lidar to seamlessly access the system without complex manual configuration; optimizing the data processing flow according to the operating frequency and data volume improves the efficiency and speed of data processing, giving full play to the performance of the lidar; adjusting the rotation speed of the cooling fan when the temperature is too high and activating the anti-interference filtering circuit when the electromagnetic interference increases can effectively protect the lidar and the entire hardware interface, extend the service life of the device, and reduce performance degradation and failures caused by environmental factors.

[0018] Furthermore, the driver layer unit configures TSN rules for lidar data on the network switch, allocates high-priority transmission channels, and ensures that the transmission delay is within the microsecond level; configuring TSN rules for lidar data and allocating high-priority transmission channels ensure that lidar data can be transmitted quickly and accurately in the network; lidar data is crucial for the environmental perception and navigation of the robot, and the transmission delay within the microsecond level guarantees the real-time nature of the data, enabling the robot to obtain accurate environmental information in a timely manner, thus making quick and accurate decisions, and improving the operating safety and efficiency of the robot in complex environments.

[0019] Furthermore, when constructing the spatio-temporal semantic map, the motion control and navigation algorithm sub-unit filters, segments, and extracts features from lidar data through the point cloud processing algorithm, and performs semantic segmentation using camera image data; this method of constructing the spatio-temporal semantic map improves the accuracy and richness of the map; filtering, segmenting, and extracting features from lidar data through the point cloud processing algorithm removes noise and useless information, and extracts key environmental features, enabling the map to more accurately reflect spatial information; performing semantic segmentation using camera image data adds semantic information to the map, such as the category and attributes of objects, enabling the robot to better understand the surrounding environment, and thus plan paths and control movements more intelligently, improving the navigation ability and task execution efficiency of the robot in complex scenarios.

[0020] Furthermore, the microservice architecture sub-unit of the application layer unit based on the knowledge graph uses the Neo4j graph database to store the industrial logistics knowledge graph. Storing the industrial logistics knowledge graph in the Neo4j graph database gives full play to the advantages of the graph database in processing complex relational data. Neo4j can efficiently store and query the relationships between entities, enabling various information in industrial logistics (such as goods, tasks, equipment, etc.) to be represented and managed in an intuitive graph structure. This not only facilitates the construction and maintenance of the knowledge graph but also improves the efficiency and accuracy of information query, providing stronger support for the invocation of microservices and the coordination of work processes, and helping to improve the overall operation efficiency and management level of the industrial logistics system.

[0021] Furthermore, the situation awareness model unit pre-trains the DQN model with a large amount of scenario data in a simulation environment and then migrates it to the actual environment for online learning. Pre-training the DQN model in the simulation environment takes advantage of the fact that a large amount of scenario data can be generated and controlled in the simulation environment, enabling the model to learn rich scenario knowledge and decision-making strategies in a short time. Then, migrating the pre-trained model to the actual environment for online learning can quickly adapt to the characteristics and changes of the actual environment, reducing the time and cost of learning from scratch in the actual environment. This method of pre-training and transfer learning improves the learning efficiency and performance of the situation awareness model, enabling the robot to perceive and adapt to the actual working scenario faster and more accurately, and enhancing the intelligence and adaptability of the robot.

[0022] Furthermore, the multi-sensor fusion diagnosis unit of the multi-sensor fusion fault diagnosis and redundancy switching module preprocesses the sensor data, including filtering, noise reduction, and feature extraction, and judges the obstacle situation by calculating the credibility of sensor evidence according to the evidence theory. Preprocessing the sensor data, such as filtering, noise reduction, and feature extraction, improves the quality and usability of the data, removes noise and interference information, and makes subsequent fault diagnosis and obstacle judgment more accurate. Calculating the credibility of sensor evidence according to the evidence theory takes into account the information of multiple sensors, avoiding the limitations and uncertainties of a single sensor and improving the accuracy and reliability of obstacle situation and chassis fault judgment. This helps the robot to detect potential dangers and faults in a timely manner, take corresponding measures, and ensure the safe operation of the robot and the smooth execution of tasks.

[0023] Furthermore, each module in the system performs data transmission and collaborative work through the topic, service, and action mechanisms of ROS2.0 to ensure the efficient operation of the overall system. Using the topic, service, and action mechanisms of ROS2.0 for data transmission and collaborative work provides a standardized and flexible communication and collaboration method for each module in the system. The topic mechanism facilitates real-time data sharing between modules, and the service mechanism supports requests between modules -

[0024] For response interaction, the action mechanism is applicable to the management of long-running tasks; this mechanism enables each module to be developed and maintained independently while also enabling close collaboration, improving the scalability and maintainability of the system; in this way, the efficient operation of the overall system is ensured, enabling the robot chassis to fully utilize the functions of each module and achieve more intelligent and efficient task execution.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1. Improvement in the generality and stability of hardware interfaces: Through the reconfigurable circuit unit based on FPGA, various sensors and actuators can be quickly adapted, reducing the hardware development cost and time. At the same time, the built-in MEMS micro environmental sensor monitors and adjusts the interface performance in real time, ensuring the stable and reliable hardware interface in complex environments and reducing the risks of communication failures and equipment damage;

[0027] 2. Efficient and intelligent software architecture: The hierarchical ROS2.0 software architecture integrating spatio-temporal semantics ensures efficient and stable data transmission at the driver layer, improves the robot's navigation and motion control capabilities at the function layer, and realizes intelligent management of industrial logistics information and data security guarantee at the application layer. Each layer works in coordination to improve the overall intelligence level and task execution efficiency of the robot;

[0028] 3. Enhancement of scenario adaptability and intelligent decision-making: The situation awareness and evolutionary intelligent scenario adaptation configuration management module endows the robot with powerful scenario perception, prediction, and automatic configuration adjustment capabilities, enabling it to quickly adapt to different scenario changes, make advance decisions, and enhance its adaptability and intelligence in complex and changeable environments;

[0029] 4. Improvement in system reliability and maintainability: The multi-sensor fusion fault diagnosis and redundancy switching module uses the evidence theory to fuse multi-sensor data to accurately judge faults, quickly switches to the standby module during faults, and records fault information for convenient troubleshooting and repair, effectively improving the system reliability and reducing the downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic block diagram of a secondary development system for a new generation of convenient robot chassis based on the ROS2.0 system;

[0031] Figure 2 It is a flowchart of a secondary development system for a new generation of convenient robot chassis based on the ROS2.0 system. DETAILED DESCRIPTION OF THE INVENTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] Please refer to Figure 1-2 , the present invention provides a technical solution: a new generation of convenient robot chassis secondary development system based on the ROS2.0 system, which is applicable to robot application development in multiple scenarios such as industrial logistics, intelligent agriculture, and medical services.

[0034] Embodiment 1:

[0035] I. Implementation background:

[0036] In modern intelligent warehousing and logistics scenarios, the adaptability and intelligence of traditional robot chassis are insufficient in complex environments. This time, a common robot chassis is redeveloped based on the ROS2.0 system to improve its operation efficiency and reliability in warehousing and logistics.

[0037] II. Hardware basis:

[0038] A robot chassis with four-wheel differential drive is selected, with a maximum load of 500 kg, a maximum speed of 2 m / s, and is equipped with a 48V, 100Ah lithium battery, with a battery life of about 8 hours. The main control platform uses an industrial-grade computer, configured with an Intel Core i7-12700 processor, 16GB of memory, and a 512GB SSD. The sensors installed on the chassis include:

[0039] 3D lidar: Model is Velodyne VLP-16, scanning range is 360°, measurement distance is 100 m, and angular resolution is 0.1°.

[0040] Binocular camera: Model is Intel RealSense D435i, resolution is 1920×1080, and frame rate is 30 fps.

[0041] IMU: Model is InvenSense MPU-9250, with an accuracy of 0.1°.

[0042] Ultrasonic sensor: Model is HC-SR04, detection distance is 2 cm - 450 cm.

[0043] Robotic arm: 6 degrees of freedom, maximum load of 20 kg, repeat positioning accuracy of ±0.1 mm.

[0044] III. Detailed implementation and data of each unit module:

[0045] (1) Reconfigurable circuit unit:

[0046] 1. Hardware implementation:

[0047] The Xilinx Zynq UltraScale+MPSoC ZU7EV FPGA chip is adopted. The main frequency of the ARM processor of this chip is 1.5 GHz, and the FPGA logic resources include 1 million logic units. The size of the PCB board is 100 mm × 150 mm, which integrates CAN, RS485, Ethernet interfaces, as well as signal amplification, filtering and other processing modules.

[0048] 2. Automatic adaptation mechanism:

[0049] When connecting the Velodyne VLP-16 lidar, the FPGA detects that it uses the EthernetUDP communication protocol within 100 ms, and automatically adjusts the clock frequency of the interface to 100 MHz and the data bit width to 32 bits. According to the data volume of the lidar (about 1 MB / s), the size of the internal data buffer of the FPGA is configured to 2 MB.

[0050] 3. Environment perception and adjustment:

[0051] The built-in BME680 environmental sensor monitors environmental information in real time. When the environmental temperature reaches 40 °C (the normal operating temperature range is -20 °C - 60 °C), the FPGA controls the duty cycle of the PWM signal of the cooling fan to increase from 50% to 80%, and the fan speed increases from 1500 rpm to 2500 rpm. When the detected electromagnetic interference intensity reaches 100 dBμV / m (the normal range is 20 - 80 dBμV / m), the internal anti-interference filtering circuit is started, and the bit error rate of the signal is reduced from 0.1% to 0.01%.

[0052] (2) Driver layer unit:

[0053] 1. Event-driven and data buffering:

[0054] The lidar driver program is written in Python. In the ROS2 environment, the lidar collects point cloud data every 100 ms and triggers the callback function for processing. The size of the circular buffer is set to 10 data frames. When the buffer is full, the new data overwrites the old data. During the data transmission process, due to the event-driven mechanism and the existence of the buffer, the data loss rate is reduced from 5% without buffering to 0.1%.

[0055] 2. Application of TSN technology:

[0056] Configure TSN rules on the network switch to mark lidar data as high-priority traffic. Through the IEEE802.1Qbv time-aware shaper, allocate a 100 μs time window for the transmission of lidar data. In actual tests, the transmission delay of lidar data in the network is reduced from 10 - 20 ms in a normal network to 100 - 200 μs.

[0057] (III) Motion Control and Navigation Algorithm Sub-unit:

[0058] 1. Spatiotemporal Semantic Map Construction:

[0059] Lidar Data Processing: Use the PCL library to process lidar point cloud data. After filtering operations, the noise point removal rate reaches 90%. After segmentation operations, the point cloud data is segmented into different objects, and the segmentation accuracy rate reaches 85%. After feature extraction, features such as planes and edges can be accurately identified.

[0060] Camera Image Semantic Segmentation: Use the U-Net network to perform semantic segmentation on binocular camera images. In the training phase, use 1000 labeled warehouse logistics scenario images for training, and the training cycle is 50 times. In actual applications, the accuracy rate of image semantic segmentation reaches 80%. Fuse the processed lidar data and camera image data, and the accuracy of the constructed spatiotemporal semantic map reaches ±5 cm.

[0061] 2. Path Planning Algorithm:

[0062] Combine the A* algorithm with the DWA algorithm for path planning. In a simulated warehouse environment of 50 m × 50 m, the average path planning time for the robot from the starting point to the target point is 1 - 2 s. During actual operation, when encountering dynamic obstacles, the DWA algorithm can adjust the motion trajectory within 0.5 s, and the obstacle avoidance success rate reaches 95%.

[0063] (IV) Application Layer Unit:

[0064] 1. Microservice Architecture Sub-unit Based on Knowledge Graph:

[0065] Knowledge Graph Construction: Use the Neo4j graph database to construct an industrial logistics knowledge graph. Collect 1000 pieces of cargo information, 200 pieces of warehouse layout information, 50 pieces of task rule information, and 100 pieces of equipment information. The average query response time of the knowledge graph is 100 - 200 ms.

[0066] Microservice Invocation and Coordination: When receiving a goods handling task, the knowledge graph is queried through the RESTful API interface, and the time to obtain relevant information is 200 - 300 ms. The navigation service, goods identification service, and robotic arm control service are called, and the workflow of these services is coordinated. The entire task execution time is shortened by 30% compared to the traditional method.

[0067] 2. Blockchain Technology Application Sub - unit:

[0068] The Hyperledger Fabric consortium blockchain architecture is adopted, connecting 5 nodes (warehouse management system, 3 robotic chassis, and logistics scheduling center). During the goods handling process, task information, goods information, and equipment status information are recorded on the blockchain, and the recording time is 1 - 2 s. The automatic allocation and settlement of tasks are achieved through smart contracts, and the settlement accuracy rate reaches 100%.

[0069] (V) Situation Awareness Model Unit:

[0070] 1. Model Construction and Pre - training:

[0071] A deep Q - network (DQN) is used to construct the situation awareness model, which is implemented using the PyTorch framework. In the simulation environment, 5000 different warehousing and logistics scenario data are used to pre - train the DQN model, and the training cycle is 100 times. After pre - training, the decision accuracy rate of the model in the simulation environment reaches 70%.

[0072] 2. Transfer Learning and Online Learning:

[0073] The pre - trained DQN model is migrated to the actual robotic chassis. During the actual operation, the model analyzes sensor data in real - time. The accuracy rate of judging the current application scenario is 60% in the initial stage of operation. After 100 times of online learning, the accuracy rate is increased to 85%. It can predict the dynamic change trend of the scenario 5 - 10 s in advance.

[0074] (VI) Evolutionary Algorithm Unit:

[0075] 1. Hybrid Optimization Strategy:

[0076] A hybrid strategy combining genetic algorithm and particle swarm optimization algorithm is used to optimize the scenario -

[0077] configuration mapping table. The population size of the genetic algorithm is set to 50, the crossover probability is 0.8, and the mutation probability is 0.1. The number of particles in the particle swarm optimization algorithm is 30, and the inertia weight is 0.7.

[0078] 2. Algorithm Execution Process:

[0079] When the robot chassis enters a narrow passage scenario, the time to complete the task in the initial configuration is 60 s, and the energy consumption is 50 Wh. After adjusting the configuration parameters with the hybrid optimization algorithm, the time to complete the task is shortened to 40 s, and the energy consumption is reduced to 30 Wh.

[0080] (VII) Multi-sensor fusion fault diagnosis and redundancy switching module:

[0081] 1. Multi-sensor fusion diagnosis unit:

[0082] Data preprocessing: Use the Kalman filtering algorithm to filter the IMU data. The noise variance of the filtered data is reduced from 0.01 to 0.001. Perform operations such as histogram equalization and edge detection on the camera images. The contrast of the images is increased by 30%, and the accuracy of edge detection reaches 90%.

[0083] Evidence theory fusion: Use the Dempster-Shafer evidence theory to fuse the sensor data. When the lidar and the camera have inconsistent judgments on the front obstacle, through the calculation of the evidence theory, the credibility of the existence of the obstacle is increased from 70% of a single sensor to 90%.

[0084] 2. Redundancy switching strategy unit:

[0085] When a lidar failure is detected, the system switches the function to the standby ultrasonic sensor within 0.5 s. During the operation after the switch, the obstacle avoidance success rate of the robot chassis still remains above 90%. At the same time, the system records the fault information in the log and sends the fault situation to the maintenance personnel within 1 s.

[0086] In summary: Through the implementation of the above unit modules, the operation efficiency of the robot chassis in the warehousing and logistics scenario is significantly improved. The average task completion time is shortened by 30%, and the energy consumption is reduced by 20%. The reliability and stability of the system are improved, and the failure rate is reduced from 5% to 1%. The robot chassis can better adapt to the complex and changeable warehousing and logistics environment, bringing significant economic benefits to the enterprise.

[0087] Embodiment 2: Intelligent agriculture scenario:

[0088] Hardware basis:

[0089] Select a tracked robot chassis with good off-road performance, which can adapt to the complex terrain of farmland. The chassis is equipped with a large-capacity lithium battery, and the endurance time can reach 10 hours, and the maximum load is 300 kg. The installed sensors include:

[0090] Multi-spectral camera: Used to monitor the growth status and pest and disease conditions of crops, etc. The spectral range covers the visible light and near-infrared bands, and the resolution is 2048×1536.

[0091] Soil moisture sensor: Measures the soil moisture with an accuracy of ±2%.

[0092] Weather station: Real-time monitors meteorological information such as temperature, humidity, light intensity, wind speed, etc.

[0093] Robotic arm: Used for picking ripe crops, with 4 degrees of freedom, a maximum load of 5 kg, and a repeat positioning accuracy of ±0.2 mm.

[0094] Detailed implementation and data of each unit module:

[0095] Reconfigurable circuit unit:

[0096] Hardware implementation: Adopts an Altera Cyclone V FPGA chip, which has rich logic resources and can meet the interface requirements of various sensors and actuators. The PCB board integrates multiple communication interfaces such as SPI, I2C, and CAN, with a size of 80 mm × 120 mm.

[0097] Automatic adaptation mechanism: When connecting a multispectral camera, the FPGA recognizes its SPI communication protocol within 150 ms, automatically adjusts the clock frequency of the interface to 20 MHz, and the data bit width to 16 bits. According to the data volume of the camera (about 500 KB / s), the size of the internal data buffer area of the FPGA is configured to 1 MB.

[0098] Environmental perception and adjustment: The built-in temperature and humidity sensors real-time monitor environmental information. When the environmental humidity reaches 80% (the normal working humidity range is 20% - 70%), the FPGA activates the moisture-proof protection circuit and reduces the working voltage of the circuit board by 5% to prevent the circuit from being damaged by moisture. When the detected light intensity is too high, the rotation speed of the cooling fan is adjusted from 1200 rpm to 2000 rpm.

[0099] Driver layer unit:

[0100] Event-driven and data buffering: Writes a multispectral camera driver program using C++. In the ROS2 environment, the camera acquires image data every 200 ms and triggers a callback function for processing. The size of the circular buffer is set to 8 data frames, and the data loss rate is reduced from 3% without buffering to 0.05%.

[0101] TSN technology application: Configures TSN rules on the network switch to allocate a 200 μs time window for the transmission of multispectral camera data. In actual tests, the transmission delay of camera data in the network is reduced from 15 - 25 ms in a common network to 150 - 250 μs.

[0102] Motion control and navigation algorithm sub-unit:

[0103] Construction of Spatiotemporal Semantic Map:

[0104] Multi-spectral Camera Data Processing: Analyze the data collected by the multi-spectral camera to extract the growth characteristics of crops, such as chlorophyll content, leaf area index, etc. The accuracy rate of the processed data reaches 85%.

[0105] Environmental Information Fusion: Integrate the data from soil humidity sensors, weather stations and multi-spectral camera data to construct a spatiotemporal semantic map containing the growth status of crops, soil conditions and meteorological information. The map accuracy reaches ±10cm.

[0106] Path Planning Algorithm: In a 100-acre farmland, the robot needs to monitor and pick crops according to a preset route. The A* algorithm combined with the Dijkstra algorithm is used for path planning, and the average path planning time is 2 - 3s. When encountering obstacles such as ridges and ditches, it can adjust the motion trajectory within 1s, and the obstacle avoidance success rate reaches 90%.

[0107] Application Layer Unit:

[0108] Microservice Architecture Sub-unit Based on Knowledge Graph:

[0109] Knowledge Graph Construction: Use the JanusGraph graph database to construct an agricultural knowledge graph, collecting 800 pieces of crop information, 300 pieces of soil information, 200 pieces of meteorological information and 150 pieces of farming operation rule information. The average query response time of the knowledge graph is 150 - 250ms.

[0110] Microservice Invocation and Coordination: When the robot detects pests and diseases in the crops of a certain farmland, query the knowledge graph through the RESTful API interface to obtain relevant control solutions, and the query time is 300 - 400ms. Invoke the pesticide spraying service and the farming reminder service, and coordinate the work processes of these services. The entire processing process is shortened by 40% compared with the traditional method.

[0111] Blockchain Technology Application Sub-unit: Adopt the Corda consortium chain architecture to connect 4 nodes such as the farm management system, the robot chassis and the agricultural product seller. During the crop planting process, record the planting information, fertilization and pesticide application information and growth monitoring information on the blockchain, and the recording time is 1 - 2s. Achieve the quality traceability and sales settlement of agricultural products through smart contracts, and the settlement accuracy rate reaches 100%.

[0112] Situational Awareness Model Unit:

[0113] Model Construction and Pre-training: The situation awareness model is constructed using the policy gradient algorithm in deep reinforcement learning and implemented using the TensorFlow framework. In the simulation environment, the model is pre-trained with 3000 different farmland scenario data, and the training cycle is 80 times. After pre-training, the decision accuracy of the model in the simulation environment reaches 65%.

[0114] Transfer Learning and Online Learning: The pre-trained model is transferred to an actual robot chassis. During actual operation, the model analyzes sensor data in real time. The accuracy of judging the current farmland scenario is 60% at the initial stage of operation. After 80 times of online learning, the accuracy is increased to 80%. It can predict the growth trend of crops 3 - 5 days in advance.

[0115] Evolutionary Algorithm Unit:

[0116] Hybrid Optimization Strategy: A hybrid strategy combining the simulated annealing algorithm and the ant colony algorithm is used to optimize the scenario-configuration mapping table. The initial temperature of the simulated annealing algorithm is set to 1000, and the cooling rate is 0.95. The number of ants in the ant colony algorithm is 20, and the pheromone evaporation coefficient is 0.2.

[0117] Algorithm Execution Process: When the robot operates in farmland with different fertilities, the time required to complete a comprehensive monitoring under the initial configuration is 120 minutes, and the energy consumption is 60 Wh. After adjusting the configuration parameters through the hybrid optimization algorithm, the completion time is shortened to 90 minutes, and the energy consumption is reduced to 45 Wh.

[0118] Multi-sensor Fusion Fault Diagnosis and Redundancy Switching Module:

[0119] Multi-sensor Fusion Diagnosis Unit:

[0120] Data Preprocessing: The multi-spectral camera data is denoised and normalized. After processing, the signal-to-noise ratio of the data is increased by 20%. The soil moisture sensor data is filtered using the Kalman filter, and the fluctuation range of the filtered data is reduced from ±5% to ±1%.

[0121] Evidence Theory Fusion: The Dempster-Shafer evidence theory is used to fuse sensor data. When the multi-spectral camera and the soil moisture sensor have inconsistent judgments on the water shortage situation of crops, through evidence theory calculation, the credibility of judging the water shortage situation is increased from 75% of a single sensor to 92%.

[0122] Redundant switching strategy unit: When a malfunction is detected in the multispectral camera, the system switches the function to the standby ordinary RGB camera within 0.8 s. During the operation after the switch, although the monitoring accuracy decreases, it can still meet the basic requirements for monitoring crop growth, and the monitoring accuracy rate remains above 70%. At the same time, the system records the fault information in the log and sends the fault situation to the maintenance personnel within 1.5 s.

[0123] Embodiment 3: Medical service scenario:

[0124] Hardware basis:

[0125] A wheeled robot chassis is selected, which has flexible steering ability, a maximum speed of 1.5 m / s, a battery life of 6 hours, and a maximum load of 100 kg. The installed sensors include:

[0126] 3D depth camera: Used to identify the surrounding environment and human body position, with a resolution of 640×480 and a depth measurement accuracy of ±1 cm.

[0127] Infrared body temperature sensor: Measures human body temperature with an accuracy of ±0.1°C.

[0128] Air quality sensor: Monitors indoor air quality, including indicators such as PM2.5 and formaldehyde.

[0129] Robotic arm: Used for handling drugs and supplies, with 3 degrees of freedom, a maximum load of 3 kg, and a repeat positioning accuracy of ±0.15 mm.

[0130] Detailed implementation and data of each unit module:

[0131] Reconfigurable circuit unit:

[0132] Hardware implementation: An Lattice ECP5 FPGA chip is adopted, and its low-power characteristics are suitable for long-term operation in medical scenarios. The PCB board integrates communication interfaces such as USB, UART, and Ethernet, with a size of 60 mm×100 mm.

[0133] Automatic adaptation mechanism: When connecting the 3D depth camera, the FPGA recognizes its USB communication protocol within 120 ms, automatically adjusts the clock frequency of the interface to 48 MHz, and the data bit width to 8 bits. According to the data volume of the camera (about 300 KB / s), the size of the internal data buffer area of the FPGA is configured to 512 KB.

[0134] Environment perception and adjustment: The built-in air quality sensor monitors the environmental information in real time. When the indoor formaldehyde concentration exceeds 0.1 mg / m 3 (The national standard is 0.1 mg / m 3) When the FPGA starts the air purification module, it also reduces the running speed of the robot from 1.5 m / s to 1 m / s to reduce air disturbance. When the detected temperature is too high, it adjusts the fan speed of the heat sink from 1000 rpm to 1800 rpm.

[0135] Driver layer unit:

[0136] Event-driven and data buffering: Write a 3D depth camera driver program using Python. In the ROS2 environment, the camera captures depth data every 150 ms and triggers a callback function for processing. The size of the circular buffer is set to 6 data frames, and the data loss rate is reduced from 2% without buffering to 0.03%.

[0137] Application of TSN technology: Configure TSN rules on the network switch to allocate a 150 μs time window for the transmission of 3D depth camera data. In actual tests, the transmission delay of camera data in the network is reduced from 10 - 20 ms in a normal network to 100 - 200 μs.

[0138] Motion control and navigation algorithm sub-unit:

[0139] Construction of spatio-temporal semantic map:

[0140] Processing of 3D depth camera data: Process the data collected by the 3D depth camera to construct a three-dimensional map of the indoor environment. The accuracy of the map reaches ±5 mm, and it can accurately identify areas such as wards, corridors, and elevators.

[0141] Fusion of human body positions: Integrate the human body position information detected by the infrared body temperature sensor with the 3D depth camera data to construct a spatio-temporal semantic map containing personnel distribution and environmental information.

[0142] Path planning algorithm: In a hospital building with an area of 5000 square meters, the robot needs to transport medicines and supplies along a specified route. The Dijkstra algorithm combined with the A* algorithm is used for path planning, and the average path planning time is 1 - 2 s. When encountering crowded areas or obstacles, it can adjust the motion trajectory within 0.5 s, and the obstacle avoidance success rate reaches 92%.

[0143] Application layer unit:

[0144] Micro-service architecture sub-unit based on knowledge graph:

[0145] Construction of knowledge graph: Use the GraphDB graph database to construct a medical knowledge graph, collecting 1000 pieces of medicine information, 500 pieces of patient information, 300 pieces of department information, and 200 pieces of medical process rule information. The average query response time of the knowledge graph is 120 - 220 ms.

[0146] Microservice Invocation and Coordination: When the robot receives a drug delivery task, it queries the knowledge graph through the RESTful API interface to obtain the storage location of the drug, the target department, and patient information. The query time is 200 - 300 ms. It invokes the navigation service, material handling service, and information notification service, and coordinates the workflow of these services. The entire task execution time is shortened by 35% compared to the traditional method.

[0147] Blockchain Technology Application Sub - unit: It adopts the Hyperledger Sawtooth consortium chain architecture and connects 5 nodes such as the hospital information system, the robot chassis, and the pharmacy management system. During the drug delivery process, information such as drug information, delivery time, and receiving department is recorded on the blockchain, and the recording time is 1 - 2 s. Through smart contracts, inventory management and cost settlement of drugs are realized, and the settlement accuracy rate reaches 100%.

[0148] Situational Awareness Model Unit:

[0149] Model Construction and Pre - training: A recurrent neural network (RNN) is used to construct the situational awareness model, which is implemented using the Keras framework. In the simulation environment, 2000 different hospital scenario data are used to pre - train the model, and the training cycle is 60 times. After pre - training, the decision accuracy rate of the model in the simulation environment reaches 60%.

[0150] Transfer Learning and Online Learning: The pre - trained model is transferred to the actual robot chassis. During the actual operation, the model analyzes sensor data in real - time. The accuracy rate of judging the current hospital scenario is 55% at the initial stage of operation. After 60 times of online learning, the accuracy rate is increased to 75%. It can predict the needs of patients 2 - 3 minutes in advance.

[0151] Evolutionary Algorithm Unit:

[0152] Hybrid Optimization Strategy: A hybrid strategy combining genetic algorithm and particle swarm optimization algorithm is used to optimize the scenario - configuration mapping table. The population size of the genetic algorithm is set to 40, the crossover probability is 0.7, and the mutation probability is 0.08. The number of particles of the particle swarm optimization algorithm is 25, and the inertia weight is 0.6.

[0153] Algorithm Execution Process: When the robot operates in the hospital environment on different floors, the time required to complete a drug delivery task under the initial configuration is 15 minutes, and the energy consumption is 15 Wh. After adjusting the configuration parameters through the hybrid optimization algorithm, the completion time is shortened to 12 minutes, and the energy consumption is reduced to 12 Wh.

[0154] Multi - sensor Fusion Fault Diagnosis and Redundant Switching Module:

[0155] Multi - sensor Fusion Diagnosis Unit:

[0156] Data preprocessing: Filter and extract features from the 3D depth camera data, so that the processed data can more accurately identify objects. Calibrate the data of the infrared body temperature sensor, and the data error after calibration is reduced from ±0.2°C to ±0.05°C.

[0157] Evidential theory fusion: Use the Dempster-Shafer evidential theory to fuse the sensor data. When the 3D depth camera and the infrared body temperature sensor have inconsistent judgments on the personnel position and body temperature, through the calculation of the evidential theory, the credibility of the judgment result is increased from 70% of a single sensor to 90%.

[0158] Redundant switching strategy unit: When a failure of the 3D depth camera is detected, the system switches the function to the standby 2D camera within 0.6 s. During the operation after the switch, although the environmental perception ability decreases, it can still complete basic navigation and obstacle avoidance tasks, and the navigation accuracy rate remains above 80%. At the same time, the system records the fault information in the log and sends the fault situation to the maintenance personnel within 1 s.

[0159] In summary, the present invention provides a new generation of convenient robot chassis secondary development system based on the ROS2.0 system;

[0160] At the hardware level: The innovative design of the reconfigurable circuit unit in the multi-modal self-perception and adaptive general hardware interface module based on FPGA realizes the flexible integration of the communication interface and the signal processing module, can automatically adapt the protocol parameters according to the connected device, and adjusts the performance in real time with the help of the MEMS micro environmental sensor, which brings a qualitative leap to the hardware versatility and stability, and breaks through the limitation of the fixed configuration of the traditional hardware interface.

[0161] In the software architecture: The unique design of the hierarchical ROS2.0 software architecture module that fuses spatio-temporal semantics. The driving layer unit combines event-driven, data buffering and TSN technology to build an efficient and stable channel for data transmission; the motion control and navigation algorithm sub-units in the function layer innovatively fuse multiple data to build a spatio-temporal semantic map and use advanced algorithms to plan paths, significantly improving the robot's environmental understanding and motion control capabilities; the application layer uses knowledge graph and blockchain technology to realize the intelligent management of industrial logistics information and data security guarantee, creating a new model for logistics task management and data processing.

[0162] Facing complex and changeable application scenarios, the situation awareness and evolutionary intelligent scenario adaptation configuration management module, based on deep reinforcement learning, transfer learning and hybrid optimization algorithms, endows the robot chassis with powerful scenario awareness, prediction and automatic configuration adjustment capabilities, breaking through the bottleneck of the traditional robot's lagging response to scenario changes.

[0163] In addition, the multi-sensor fusion fault diagnosis and redundancy switching module diagnoses faults by fusing multi-sensor data based on the evidence theory and is equipped with a redundancy switching strategy, greatly improving the system reliability and maintainability and solving the problems of inaccurate fault diagnosis and untimely fault handling of traditional robots.

[0164] These technical means support and are organically combined with each other to jointly build a highly intelligent, flexible and reliable secondary development system for robot chassis, bringing innovative breakthroughs to the applications of robots in multiple fields such as industrial logistics, intelligent agriculture, and medical services, realizing the efficient, intelligent and reliable operation of the robot chassis in multiple scenarios, improving the versatility, adaptability and maintainability, and being of great significance to promoting the development of robot technology in complex scenarios.

Claims

1. A new generation of convenient robot chassis secondary development system based on ROS2.0 system, characterized by: include: Multi-modal self-sensing adaptive universal hardware interface module: including: Reconfigurable circuit unit: Based on FPGA, it integrates multiple communication interfaces and signal processing modules, and can automatically adapt the communication protocol and interface parameters according to the connected sensors and actuators. The built-in MEMS micro-environmental sensor monitors temperature, humidity and electromagnetic interference in real time, and adjusts the interface performance accordingly. Hierarchical ROS2.0 software architecture module integrating spatiotemporal semantics: including: Driver layer unit: The driver is developed by combining event-driven and data buffering mechanisms with TSN technology, and callback functions are used to process data. TSN ensures data transmission time synchronization, and the ring buffer avoids data loss. Functional layer unit: including motion control and navigation algorithm subunit: integrating algorithms based on spatiotemporal semantic maps, integrating lidar and camera data to build maps containing spatial, temporal dynamic and semantic information, using A* algorithm combined with dynamic window method to plan paths, and adjusting trajectories in real time according to the environment; Application layer unit: including: knowledge graph-based microservice architecture sub-unit: building industrial logistics knowledge graph, storing it in graph database, querying and updating it through RESTful API interface, calling related services according to tasks and coordinating workflows; blockchain technology application sub-unit: using alliance chain architecture, nodes have unique identity and public and private key pairs, record tasks, goods and equipment information, and realize task allocation and settlement through smart contracts; Situational awareness and evolutionary intelligent scene adaptation configuration management module: including: Contextual Awareness Model Unit: Builds models based on deep reinforcement learning and transfer learning, migrates DQN pre-training to the actual environment for online learning, and analyzes sensor data in real time to determine scenarios and predict changes; Evolutionary algorithm unit: uses a hybrid strategy combining genetic algorithm and particle swarm optimization algorithm to optimize the scenario-configuration mapping table and adjust chassis configuration parameters according to new scenarios; Multi-sensor fusion fault diagnosis and redundancy switching module: including: Multi-sensor fusion diagnosis unit: Based on evidence theory, it fuses the data of lidar, camera, IMU and ultrasonic sensors, judges the chassis status after pre-processing, and monitors the failure of moving parts; Redundant switching strategy unit: When a fault is detected, the function of the faulty module is switched to the standby module according to the preset strategy, the fault information is recorded and the maintenance personnel are notified.

2. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: The reconfigurable circuit unit automatically adapts to the Ethernet UDP communication protocol for the laser radar, optimizes the data processing flow according to its operating frequency and data volume, adjusts the speed of the cooling fan when the temperature is too high, and starts the anti-interference filter circuit when the electromagnetic interference is enhanced.

3. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: The driver layer unit configures TSN rules on the network switch for the lidar data, allocates high-priority transmission channels, and ensures that the transmission delay is within microseconds.

4. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: When constructing the spatiotemporal semantic map, the motion control and navigation algorithm subunit filters, segments and extracts features of the lidar data through a point cloud processing algorithm, and performs semantic segmentation using camera image data.

5. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: The knowledge graph-based microservice architecture subunit of the application layer unit uses the Neo4j graph database to store the industrial logistics knowledge graph.

6. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: The context-aware model unit pre-trains the DQN model with a large amount of scene data in a simulated environment, and then migrates it to an actual environment for online learning.

7. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: The multi-sensor fusion diagnosis unit of the multi-sensor fusion fault diagnosis and redundancy switching module pre-processes sensor data including filtering, noise reduction and feature extraction, and calculates the sensor evidence credibility according to the evidence theory to judge the obstacle situation.

8. The new generation convenient robot chassis secondary development system based on ROS2.0 system as claimed in claim 1, characterized in that: Each module in the system transmits data and works collaboratively through the topics, services and action mechanisms of ROS2.0.

Images