A method and system for cooperative control and communication scheduling of unmanned aerial vehicles and robotic vehicles

Abstract

The application discloses a kind of unmanned aerial vehicle and robot vehicle road cooperation control communication scheduling method and system, it is related to communication technical field.The method comprises the following steps: S101, construct the heterogeneous fusion communication architecture based on UWB ultra-wideband and V2X vehicle networking, and carry 6G short-distance high-speed transmission protocol;S102, real-time acquisition spectrum energy, error rate and retransmission rate, and the current industrial electromagnetic interference type is identified using lightweight convolutional neural network.By constructing the heterogeneous fusion communication architecture of UWB ultra-wideband and V2X vehicle networking, in combination with the interference identification of lightweight convolutional neural network and adaptive anti-interference, the superframe pre-allocation and micro-time slot preemption scheduling of double-scale deep reinforcement learning model, high-precision positioning, wide-area cooperation is realized Unified, high-reliable transmission of instruction, and extremely low latency and high priority burst service fast response, meet the real-time control needs of unmanned aerial vehicle and robot vehicle road cooperation operation.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method and system for communication scheduling of unmanned aerial vehicles and robots in vehicle-road cooperative control. Background Technology

[0002] With the rapid development of intelligent transportation and collaborative applications of unmanned systems, stringent requirements have been placed on the positioning accuracy, real-time performance, and anti-interference capabilities of wireless communication systems. Although current mainstream vehicle-to-everything (V2X) communication technologies can support data interaction over medium distances between vehicles and between vehicles and infrastructure, their positioning accuracy is relatively low, and latency fluctuations are significant in multi-device, highly dynamic scenarios, making it difficult to meet the millisecond to microsecond response requirements of industrial real-time control.

[0003] In terms of positioning technology, while UWB (Ultra-Wideband) technology can achieve centimeter-level high-precision positioning, its communication coverage is limited, making it difficult to independently undertake wide-area collaborative tasks. This technical characteristic results in significant coverage blind spots in existing systems when facing large-area, multi-node collaborative operation scenarios, requiring the deployment of additional positioning infrastructure, which increases the complexity and cost of system deployment.

[0004] In terms of data transmission, collaborative operations between drones and mobile robots place higher demands on communication bandwidth and latency. Existing technologies struggle to simultaneously achieve high-speed data transmission and high-precision positioning, resulting in time discrepancies between the issuance and execution of collaborative control commands, impacting the overall system response speed and operational efficiency. Furthermore, the complex electromagnetic environment in industrial settings leads to frequent data packet loss and significantly increased retransmission latency in conventional wireless communication systems, severely affecting the reliable transmission and execution of control commands.

[0005] To address these issues, the industry has begun exploring a technical approach that combines heterogeneous converged communication architecture with intelligent scheduling algorithms. However, existing solutions still have the following shortcomings in practical applications: First, communication resource allocation lacks multi-scale collaborative management of superframes and micro-slots, making it difficult to achieve fine-grained dynamic scheduling; second, electromagnetic interference identification and adaptive anti-interference capabilities are insufficient, failing to accurately filter out different types of interference; third, the application of deep reinforcement learning models in the field of communication scheduling is still in the exploratory stage, and the preemptive response mechanism for high-priority bursty services needs further improvement. Therefore, there is an urgent need to propose a communication scheduling method and system for UAV and robot vehicle-road cooperative control that balances high-precision positioning, extremely low-latency communication, and efficient collaborative control. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for communication scheduling in UAV and robot vehicle-road cooperative control. By constructing a heterogeneous fusion communication architecture of UWB and V2X vehicle networking, and combining interference recognition and adaptive anti-interference of lightweight convolutional neural networks with superframe pre-allocation and micro-slot preemptive scheduling of dual-scale deep reinforcement learning models, it achieves unified high-precision positioning and wide-area cooperation, highly reliable transmission of commands, and rapid response to extremely low latency and high-priority bursty services, thus meeting the real-time control requirements of UAV and robot vehicle-road cooperative operations.

[0007] In a first aspect, the present invention provides a method for coordinated control, communication, and scheduling of unmanned aerial vehicles (UAVs) and robots for vehicle-road cooperation, comprising: S101. Construct a heterogeneous converged communication architecture based on UWB ultra-broadband and V2X vehicle networking, and carry 6G short-range high-speed transmission protocol; S102. Real-time acquisition of spectrum energy, bit error rate and retransmission rate, and the use of lightweight convolutional neural network to identify the current industrial electromagnetic interference type; S103. Construct a dual-scale deep reinforcement learning model in the heterogeneous converged communication architecture. Based on the fusion data association table collected by UWB and V2X vehicle network, perform large-scale communication resource pre-allocation with superframe as the period, and perform small-scale resource dynamic adjustment and preemption scheduling with micro-time slot as the period. S104. Combine the scheduled communication resources output by the dual-scale deep reinforcement learning model to execute collaborative control commands between UAVs, ground mobile robots, and vehicle-road equipment.

[0008] Preferably, S101 includes: The devices participating in the collaborative communication initialize the UWB ultra-wideband communication unit and the V2X vehicle-to-everything (V2X) communication unit respectively, and load the 6G short-range high-speed transmission protocol stack. The devices participating in the collaborative communication include drones, ground mobile robots, and vehicle-to-infrastructure (V2I) equipment. Each UWB ultra-wideband communication unit of a participating device sends a timestamped pulse signal to other participating devices in a preset polling order. The receiving device records the arrival time of the pulse signal, calculates the signal flight time using a two-sided bidirectional ranging method, generates the straight-line distance between the devices, and packages the distance values ​​measured by each device into a positioning data frame; Each V2X vehicle-to-everything (V2X) communication unit of participating devices periodically sends out data packets containing its own identifier, location, speed, direction of movement, and task status during the allocated broadcast time slots, and continuously listens to the channel to receive V2X data packets sent by other devices and extract the collaborative information in the V2X data packets. The distance values ​​obtained by the UWB ultra-wideband communication unit and the location and motion information obtained by the V2X vehicle-to-everything communication unit are timestamped to form a fusion data association table indexed by the device ID.

[0009] Preferably, S102 includes: The spectral energy, bit error rate, and retransmission rate are concatenated into a three-dimensional feature vector, and after linear normalization, they are repeatedly arranged to form a single-channel feature map. A single-channel feature map is input into a lightweight convolutional neural network, and three feature mapping values ​​are output after passing through the convolutional layer. The three feature map values ​​output by the convolutional layer are flattened and then input into a fully connected layer, which outputs the probability distribution of each interference class. The fully connected layer uses the Softmax activation function. The category with the highest probability is selected as the current industrial electromagnetic interference type, the identification result is output, and the notch filter parameters are switched according to the identification result to dynamically adjust the transmission power and modulation coding scheme.

[0010] Preferably, the step of switching notch filter parameters and dynamically adjusting the transmit power and modulation coding scheme according to the identification result includes: If the identification result is impulse interference, then switch to a digital filter with notch characteristics, where the notch center frequency is set to the most recently counted interference frequency. If the identification result is narrowband interference, the frequency hopping mapping table is switched to a pre-stored set of spare frequency points; If the identification result indicates the presence of interference, the transmission power will be gradually reduced according to the preset step size until the interference power ratio is lower than the threshold. If the identification result is no interference, then restore the transmit power and modulation and coding scheme to their initial values.

[0011] Preferably, S103 includes: A dual-scale deep reinforcement learning model is constructed in a heterogeneous converged communication architecture, which includes a large-scale deep Q-network and a small-scale dual-delay deep deterministic policy gradient network. At the start of each superframe, the current queue length, average transmission latency in the most recent superframe, device motion speed and trajectory prediction values, and device service priority labels for each device are read from the fused data association table and concatenated to obtain a state vector; the expression of the state vector is as follows: ; In the formula, For the first The current queue length of each device For the first The average transmission latency of each device within the most recent superframe. For the first Predicted speed and trajectory values ​​of each device For the first Service priority tags for each device The total number of devices participating in collaborative communication; The state vector is input into a large-scale deep Q-network, and after forward propagation through multiple fully connected layers, the Q value corresponding to the allocation action of each time slot block is output. The action corresponding to the maximum Q value is selected as the pre-allocated time slot resource. The pre-allocated time slot resource includes the number of time slot blocks allocated to each device in the superframe, the starting time slot offset, and the length of the time slot block. The pre-allocated time slot resources are used as the initial reference allocation scheme for the small-scale dual-delay deep deterministic policy gradient network. The expected latency and throughput corresponding to the initial reference allocation scheme are used as the initial input of the value network, and the scheduled communication resources are generated through a small-scale dual-delay deep deterministic policy gradient network. At the end of each superframe, the small-scale dual-delay deep deterministic policy gradient network statistically analyzes the actual utilization rate of each micro-slot within the superframe and the backlog length of low-priority services caused by preemption. This data is then combined with the state collected by the large-scale deep Q-network at the beginning of the current superframe to form an empirical sample, which is stored in the empirical replay buffer. The expression for the reward function is as follows: ; In the formula, For the actual utilization rate of micro-slots, For low-priority business backlog length, For average transmission delay, , , These are the weighting coefficients; Once the number of samples in the experience replay buffer reaches a preset threshold, a batch of samples is randomly selected, the target Q value is calculated, and the parameters of the large-scale deep Q network are updated using the gradient descent method.

[0012] Preferably, the step of using the expected latency and throughput corresponding to the initial reference allocation scheme as the initial input to the value network, and generating the scheduled communication resources through a small-scale dual-latency deep deterministic policy gradient network, includes: The pre-allocated time slot resources are used as the initial reference allocation scheme for the small-scale dual-delay deep deterministic policy gradient network, and the expected delay and throughput corresponding to the initial reference allocation scheme are used as the initial inputs of the value network. At the beginning of each micro-slot, real-time channel state information and current interference type label are collected to detect whether high-priority burst services arrive. The real-time information and the allocation status of the corresponding micro-slots in the current pre-allocated time slot resources are spliced ​​together to form a small-scale state; The small-scale state is input into the policy network of the dual-delay deep deterministic policy gradient algorithm. The policy network outputs a continuous adjustment vector, which includes the adjustment amount for the current micro-slot allocation, the fine-tuning value of the transmit power, and the index change of the modulation and coding scheme. If a high-priority service is detected and the current micro-slot is already occupied by a low-priority service, the dual-delay deep deterministic policy gradient algorithm outputs a preemption action to generate scheduled communication resources. The preemption action reassigns the current micro-slot to the high-priority service and reschedules the preempted low-priority service to an idle micro-slot.

[0013] Preferably, S104 includes: According to the 6G short-range high-speed transmission protocol format, each control command generated based on the scheduled communication resources will be encapsulated into a short data frame. The receiving device parses the instruction content in the short data frame, converts the instruction content into a control signal that the actuator can recognize, and completes the execution of the instruction in the next control cycle.

[0014] Secondly, the present invention also provides a UAV and robot vehicle-road cooperative control communication scheduling system, which applies the UAV and robot vehicle-road cooperative control communication scheduling method described above. The system includes a construction module, an identification module, a scheduling module and an execution module. The building module is used to construct a heterogeneous converged communication architecture based on UWB ultra-wideband and V2X vehicle networking, and is equipped with a 6G short-range high-speed transmission protocol. The identification module is used to collect spectrum energy, bit error rate and retransmission rate in real time, and uses a lightweight convolutional neural network to identify the current type of industrial electromagnetic interference. The scheduling module is used to build a dual-scale deep reinforcement learning model in the heterogeneous converged communication architecture. Based on the fusion data association table collected by UWB and V2X vehicle networking, it performs large-scale communication resource pre-allocation with superframe as the period and small-scale resource dynamic adjustment and preemption scheduling with micro-time slot as the period. The execution module is used to combine the scheduled communication resources output by the dual-scale deep reinforcement learning model to execute collaborative control commands between UAVs, ground mobile robots, and vehicle-road equipment.

[0015] Thirdly, the present invention also provides a drone and robot vehicle-road cooperative control and communication scheduling device, the drone and robot vehicle-road cooperative control and communication scheduling device including a memory and at least one processor, the memory storing instructions; the processor is used to call the instructions in the memory, so that the device executes each step of the drone and robot vehicle-road cooperative control and communication scheduling method as described above.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. This invention constructs a heterogeneous converged communication architecture based on UWB ultra-wideband and V2X vehicle-to-everything (V2X) networks. It precisely aligns the device distance values ​​obtained from UWB ranging with the absolute position and motion status information obtained from V2X broadcasting by timestamp and merges them into a fused data association table with the device ID as the primary key. This achieves the organic unity of centimeter-level high-precision positioning and wide-area collaborative information interaction, overcoming the inherent contradiction of single technologies having high positioning accuracy but limited coverage, or wide coverage but insufficient accuracy. It can eliminate coverage blind spots in large-area, multi-node collaborative operation scenarios without the need for additional positioning infrastructure deployment, providing a precise, continuous, and reliable spatiotemporal reference for the collaborative control of UAVs, ground mobile robots, and vehicle-road equipment.

[0017] 2. This invention acquires spectrum energy, bit error rate, and retransmission rate in real time at the physical layer and uses a lightweight convolutional neural network to identify the current industrial electromagnetic interference type online. Based on the identification results, it selectively switches notch digital filtering, activates adaptive frequency hopping, or adjusts the transmission power and modulation coding scheme accordingly. This enables accurate identification and differentiated suppression of complex operating conditions such as pulse interference, narrowband interference, and multiple interference superposition, effectively reducing data packet loss and retransmission delay in complex electromagnetic environments, and ensuring highly reliable transmission and stable execution of collaborative control commands.

[0018] 3. This invention constructs a dual-scale deep reinforcement learning model in a heterogeneous converged communication architecture. It performs large-scale pre-allocation of communication resources with superframes as the period and small-scale dynamic adjustment and preemption scheduling of resources with micro-time slots as the period. It also encapsulates and executes cooperative control commands based on the scheduled communication resources and forms a continuous closed loop based on the confirmation frames. This enables refined cooperative management of communication resources at multiple time scales and rapid preemption response to high-priority bursty services. It significantly reduces the transmission latency and service backlog of cooperative control, improves the utilization rate of micro-time slot resources, and meets the stringent requirements of UAV and robot vehicle-road cooperative operations for extremely low latency and real-time control. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for collaborative control, communication, and scheduling between unmanned aerial vehicles (UAVs) and robots.

[0020] Figure 2 This is a structural block diagram of a vehicle-road cooperative control, communication, and scheduling system for drones and robots.

[0021] Figure 3 This is a schematic diagram of a device for the coordinated control, communication, and scheduling of unmanned aerial vehicles (UAVs) and robots. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0023] Example 1 Please see Figures 1-3 This embodiment takes the collaborative operation scenario of drones and mobile robots in an industrial plant as the application scenario. This scenario involves a collaborative operation area of ​​drone aerial inspection, mobile robot ground handling, and vehicle-road equipment information interaction. It aims to achieve high-precision positioning and wide-area collaboration through the heterogeneous converged communication architecture of UWB ultra-wideband and V2X vehicle networking, and to achieve dynamic scheduling of communication resources through a dual-scale deep reinforcement learning model.

[0024] This invention provides a method for scheduling communication and control of unmanned aerial vehicles (UAVs) and robots in a vehicle-to-infrastructure (V2I) cooperative system. At the hardware level, each device participating in the cooperative communication integrates a UWB ultra-wideband communication module and a V2X vehicle-to-everything (V2X) communication module. The UWB module uses the DW1000 chip, supports the 3.5GHz to 6.5GHz ultra-wideband frequency band, and achieves centimeter-level ranging accuracy. The V2X module is based on the C-V2X standard, supports direct PC5 communication, and has a maximum transmission rate of 27Mbps. The converged communication controller uses a high-performance embedded processor with an 8-core ARM Cortex-A72 architecture and a main frequency of 2.5GHz, providing computing power support for real-time inference of a dual-scale deep reinforcement learning model. At the software level, the method includes a construction module, a recognition module, a scheduling module, and an execution module. The method includes: S101. Construct a heterogeneous converged communication architecture based on UWB ultra-broadband and V2X vehicle networking, and carry 6G short-range high-speed transmission protocol; Specifically, S101 includes: In the communication architecture construction, for each device participating in collaborative communication, a hardware power-on self-test is first performed to confirm that the UWB ultra-wideband communication unit and the V2X vehicle-to-everything (V2X) communication unit are working properly. Then, the corresponding underlying driver firmware is loaded. Next, the 6G short-range high-speed transmission protocol stack is initialized in the system memory, and the time slot synchronization parameters, frame structure format, and data packet encapsulation rules in the protocol stack are configured to ensure that all devices use a unified set of communication parameters. Finally, the unique identifier ID of the device is written into the node management table of the protocol stack to establish the foundation for subsequent network communication. The device types include UAVs, ground mobile robots, and vehicle-to-everything (V2X) devices. In the UWB ranging function, each device's UWB ultra-wideband communication unit obtains transmission permission sequentially according to the system's preset polling order table. Within the authorized transmission time slot, the UWB ultra-wideband communication unit generates a high-precision timestamp, encapsulates this timestamp and the local device ID into a pulse signal frame, and broadcasts it to other devices in space via the UWB antenna. After transmission, the device immediately switches to a receiving and listening mode, waiting for responses or measurement feedback from other devices. Upon receiving the pulse signal, the receiving device records the arrival time of its local clock and parses the original transmission timestamp carried in the pulse signal. Subsequently, a bilateral bidirectional ranging method is used, which eliminates errors caused by clock offset and crystal oscillator drift by averaging two round trip measurements, and calculates the actual flight time of the signal in the air. The straight-line distance between the transmitting and receiving devices is calculated by dividing the product of the electromagnetic wave propagation speed and the flight time by two. Each device packages all distance values ​​obtained from its own ranging participation into a positioning data frame according to the target device ID index and temporarily stores it in the transmission buffer. In V2X information exchange, each device's V2X vehicle-to-everything (V2X) communication unit periodically generates a basic V2X safety message within the pre-allocated broadcast time slots of the system. This message sequentially includes the device's own identifier, position coordinates obtained via GPS or inertial navigation, instantaneous velocity vector, motion direction angle, and the current task status. The V2X V2X communication unit encodes and sends this message according to the V2X communication protocol, while continuously monitoring the channel. When it detects V2X data packets sent by other devices, it immediately receives and parses them, extracting collaborative information such as the other party's identifier, position, speed, direction, and task status, and storing it in a local temporary buffer. In data fusion, the set of distance values ​​between devices obtained by the UWB ultra-wideband communication unit and the absolute position and motion status information of each device obtained by the V2X vehicle-to-everything (V2X) communication unit are precisely aligned according to the timestamp of the data generation. That is, within the time window, data with a time difference less than a preset threshold are regarded as observations at the same moment. Then, using the ID of each device as the primary key, the distance values ​​from UWB and the position coordinates and motion parameters from V2X are merged into a single record to form a complete fused data association table. This table is stored in the shared memory of the fused communication controller and can be read as needed.

[0025] S102. Real-time acquisition of spectrum energy, bit error rate and retransmission rate, and the use of lightweight convolutional neural network to identify the current industrial electromagnetic interference type; Specifically, S102 includes: In feature extraction, a spectrum sensing unit continuously runs at the physical layer of the device, collecting the spectrum energy distribution of the current channel every 1ms, divided into low-frequency, mid-frequency, and high-frequency bands. Simultaneously, the bit error rate and retransmission rate of the most recent 100 data packets are statistically analyzed from the MAC layer. These three feature values—spectral energy, bit error rate, and retransmission rate—are combined to form a three-dimensional feature vector. Each component of the three-dimensional feature vector is linearly normalized, mapping its value range to... The interval; the normalization coefficients are dynamically updated from the historical maximum values; the normalized three-dimensional feature vectors are repeatedly arranged to form a single-channel feature map with a width of 3 and a height of 1; this feature map serves as the sole input to the lightweight convolutional neural network; In convolutional processing, the feature map passes through a one-dimensional convolutional layer with a kernel size of 1×3 and a stride of 1. The convolutional layer outputs three feature mapping values. Each feature mapping value represents the initial activation level for a certain interference mode, such as no interference, impulse interference, or narrowband interference. The three feature mapping values ​​output by the convolutional layer are flattened and input into a fully connected layer with three output nodes. The fully connected layer uses the Softmax activation function and outputs the probability distribution of each interference category. In interference classification, the category with the highest probability is selected as the current type of industrial electromagnetic interference; if the highest probability is less than 0.6, it is judged as uncertain, and the previous interference type remains unchanged. In adaptive anti-interference, if the identification result is impulse interference, the controller sends a command to the baseband processing unit to switch to a digital filter with notch characteristics, and the notch center frequency is set to the most recently statistically analyzed interference frequency. If the identification result is narrowband interference, the adaptive frequency hopping algorithm is activated, and the frequency hopping mapping table is switched to a pre-stored set of backup frequencies. If the identification result is that interference exists, the transmit power is gradually reduced according to a preset step size until the interference power ratio is lower than the threshold. At the same time, the modulation and coding scheme is downgraded from a high-order scheme such as 64QAM to a low-order scheme such as QPSK, and the coding rate is reduced. If the identification result is that there is no interference, the transmit power and modulation and coding scheme are gradually restored to their initial values, and the process is repeated. A complete adaptive anti-interference processing flow is executed once per cycle.

[0026] S103. Construct a dual-scale deep reinforcement learning model in the heterogeneous converged communication architecture. Based on the fusion data association table collected by UWB and V2X vehicle network, perform large-scale communication resource pre-allocation with superframe as the period, and perform small-scale resource dynamic adjustment and preemption scheduling with micro-time slot as the period. Specifically, S103 includes: In the construction of the dual-scale model, a dual-scale deep reinforcement learning model is constructed in a heterogeneous fusion communication architecture. The dual-scale deep reinforcement learning model includes a large-scale deep Q-network and a small-scale dual-delay deep deterministic policy gradient network. The superframe period length is determined to be 20ms, the micro-slot period length is 200μs, and the experience replay buffer is cleared. In large-scale resource pre-allocation, at the start of each superframe, the current queue length, average transmission latency in the most recent superframe, device motion speed and trajectory prediction values, and device service priority labels of each device are read from the fused data association table and concatenated to obtain a state vector; the expression of the state vector is as follows: ; In the formula, For the first The current queue length of each device For the first The average transmission latency of each device within the most recent superframe. For the first Predicted speed and trajectory values ​​of each device For the first Service priority tags for each device The total number of devices participating in the cooperative communication; the state vector is input into a large-scale deep Q network, and after forward propagation through multiple fully connected layers, the Q value corresponding to the allocation action of each time slot block is output. The action corresponding to the maximum Q value is selected as the pre-allocated time slot resource, where the pre-allocated time slot resource includes the number of time slot blocks allocated to each device in a superframe, the starting time slot offset, and the length of the time slot block; In the small-scale dynamic adjustment, the pre-allocated time slot resources are used as the initial reference allocation scheme of the small-scale dual-delay deep deterministic policy gradient network, and the expected delay and throughput corresponding to the initial reference allocation scheme are used as the initial input of the value network. At the beginning of each micro-time slot, real-time channel state information and current interference type labels are collected to detect whether high-priority burst services arrive. The real-time information and the allocation state of the corresponding micro-time slot in the current pre-allocated time slot resources are concatenated to form a small-scale state. The small-scale state is input into the policy network of the dual-delay deep deterministic policy gradient algorithm, and the policy network outputs a continuous adjustment vector, which includes the adjustment amount of the current micro-time slot allocation, the transmit power fine-tuning value, and the index change of the modulation and coding scheme. If a high-priority service is detected and the current micro-time slot is occupied by a low-priority service, the dual-delay deep deterministic policy gradient algorithm outputs a preemption action to generate the scheduled communication resources. The preemption action re-allocates the current micro-time slot to the high-priority service and re-arranges the preempted low-priority service to an idle micro-time slot. During model training and updates, at the end of each superframe, the small-scale dual-delay deep deterministic policy gradient network statistically analyzes the actual utilization rate of each micro-slot within the superframe and the backlog length of low-priority services caused by preemption. This data is then combined with the state collected by the large-scale deep Q-network at the beginning of the current superframe to form an empirical sample, which is stored in the empirical replay buffer. The expression for the reward function is as follows: ; In the formula, For the actual utilization rate of micro-slots, For low-priority business backlog length, For average transmission delay, , , The weight coefficients are used; when the number of samples in the experience replay buffer reaches a preset threshold, a batch of samples is randomly selected, the target Q value is calculated, and the parameters of the large-scale deep Q network are updated using the gradient descent method.

[0027] S104. Combine the scheduled communication resources output by the dual-scale deep reinforcement learning model to execute collaborative control commands between UAVs, ground mobile robots and vehicle-road equipment. Specifically, S104 includes: In the instruction encapsulation, the communication resources after scheduling by the dual-scale deep reinforcement learning model are obtained, and it is determined whether each device is authorized to send data in each micro-time slot in the next superframe, as well as the transmit power and modulation and coding parameters used. Based on the real-time positioning and status information obtained from UWB and V2X vehicle-to-everything (V2X) networks, the controller runs a preset cooperative control algorithm, such as following formation, obstacle avoidance and detour, and intersection passage order decision, to generate flight attitude instructions for UAVs, speed and direction instructions for mobile robots, and acceleration / deceleration or traffic light control instructions for vehicle-road equipment. In accordance with the 6G short-range high-speed transmission protocol format, each control instruction is encapsulated into a short data frame, with the frame header containing the target device ID, instruction type, and timestamp, and the frame payload containing the specific control parameters. During instruction execution, each device's transmitting module immediately sends out the encapsulated control instruction data frame via UWB or V2X channel upon arrival of its authorized micro-timeslot. During transmission, the transmit power and modulation / coding scheme specified in the scheduling timeslot table are strictly followed. All devices continuously monitor the channel and synchronize with the global clock at the start of each micro-timeslot. When a data frame destined for this device is received, it is immediately sent to the receive buffer. The receiving device parses the instruction content in the data frame, converts it into control signals recognizable by the underlying actuators, and completes instruction execution in the next control cycle. After instruction execution, the receiving device replies to the controller with a short acknowledgment frame in the next authorized transmission timeslot, containing the instruction execution result and current status measurements. The controller determines whether to retransmit or adjust the next coordinated control instruction based on the acknowledgment frame, thus repeatedly forming a continuous coordinated control closed loop.

[0028] Taking a collaborative operation scenario in an industrial plant as an example, assume there is one drone, two ground mobile robots, and three vehicle-to-everything (V2X) devices participating in collaborative communication. In step S101, each device completes the initialization of the heterogeneous fusion communication architecture. The UWB unit measures the distance between the drone and robot A as 5.2m, and the distance between robot A and robot B as 3.8m. The V2X unit broadcasts position information showing that the drone's flight altitude is 50m and its speed is 8m / s. A fusion data association table is formed through data fusion. In step S102, the lightweight convolutional neural network identifies the current interference type as no interference, and the system maintains normal transmission power and 64QAM modulation. In step S10... In step 3, the large-scale deep Q network outputs a pre-allocation scheme, dividing the 20ms superframe into 100 micro-time slots. The UAV is allocated 30 micro-time slots, Robot A is allocated 25 micro-time slots, Robot B is allocated 20 micro-time slots, and the vehicle-road equipment is allocated a total of 25 micro-time slots. The small-scale TD3 network fine-tunes the allocation according to the real-time channel status. In the 45th micro-time slot, a high-priority obstacle avoidance command from the UAV is detected, and a preemption action is output to reassign the micro-time slot to the UAV. In step S104, the controller generates obstacle avoidance turning commands, encapsulates and sends them. Each device executes the commands and replies with confirmation. The entire system runs in a 20ms cycle, achieving efficient and orderly collaborative operation.

[0029] Example 2 Please see Figures 1-3 This embodiment takes the scenario of drones and mobile robots cooperating to perform emergency tasks as the application scenario. This scenario is characterized by frequent occurrence of high-priority burst services and scarce channel resources. It aims to verify the effectiveness of the dual-scale deep reinforcement learning model of the present invention in the scenario of high-priority service preemption.

[0030] In this embodiment, it is assumed that during collaborative operation, the UAV suddenly detects an obstacle ahead and needs to immediately issue an emergency obstacle avoidance command. In the small-scale dynamic adjustment stage of step S103, the TD3 network's policy network detects the arrival of a high-priority burst service at the beginning of the 60th micro-time slot. The real-time channel state information shows that the current channel quality is good and the interference type label is no interference. The continuous adjustment vector output by the policy network is: time slot allocation adjustment amount. Transmit power fine-tuning value Modulation coding scheme index change Since a high-priority service was detected and the current micro-time slot was occupied by a low-priority service, the TD3 algorithm outputs a preemption action, snatching the 60th micro-time slot from the originally allocated robot B and reassigning it to the drone to issue obstacle avoidance commands. The preempted robot B's service is rescheduled to the 85th micro-time slot, which was originally allocated to vehicle-to-everything (V2X) equipment 2. In step S104, the obstacle avoidance command is successfully sent and received by the drone in the 60th micro-time slot. The drone performs obstacle avoidance maneuvers in the next control cycle and replies with an acknowledgment frame. The controller determines that the obstacle avoidance command was successfully executed based on the acknowledgment frame and does not need to be retransmitted.

[0031] This embodiment further verifies the scheduling effect in continuous high-priority service scenarios: Assuming that in the next 5 superframes, the UAV continuously detects dynamic obstacles and needs to continuously issue obstacle avoidance commands; the large-scale deep Q network learns the arrival pattern of high-priority services based on historical experience samples and automatically adjusts the pre-allocation scheme, increasing the number of pre-allocated time slot blocks for the UAV from 30 to 45, reducing the number for robot A from 25 to 20, robot B from 20 to 15, and vehicle-to-everything (V2X) equipment from 25 to 20; the TD3 network makes fine adjustments based on this pre-allocation, reducing the number of successful preemption times in each superframe from an average of 3.2 times to 0.8 times, significantly reducing service backlog caused by preemption; in the operation of the entire system for 10 consecutive superframes, the average utilization rate of micro-time slots reaches 92.3%, the average backlog length of low-priority services is controlled within 5 data packets, and the average transmission latency is 1.8ms, meeting the real-time requirements of urgent tasks.

[0032] Example 3 Please see Figures 1-3 This embodiment is designed for complex electromagnetic interference environments, including pulse interference, narrowband interference, and multiple interference superposition, and aims to verify the effectiveness of the interference identification and adaptive anti-interference mechanism of the present invention.

[0033] This embodiment simulates a pulse interference scenario generated by the start-stop of a large motor in an industrial plant: In the feature extraction step S102, the spectrum sensing unit collects the low-frequency spectrum energy as -45dBm, the mid-frequency band as -52dBm, and the high-frequency band as -48dBm; the bit error rate is statistically... The retransmission rate was 4.1%; the normalized 3D feature vector was input into a lightweight convolutional neural network, and the three feature mapping values ​​output by the convolutional layer were respectively... After flattening, the probability distribution is output through the Softmax activation function of a fully connected layer. Among the various interference types, the second type, impulse interference, had the highest probability, reaching 0.89. The system determined the current interference type to be impulse interference. In the adaptive anti-interference stage, the controller switched to a digital filter with notch characteristics, setting the notch center frequency to the most recently statistically observed interference frequency of 400MHz. The transmit power remained unchanged, and the modulation and coding scheme remained 64QAM. After anti-interference processing, the bit error rate of the subsequent 100 data packets decreased to [missing value]. The retransmission rate dropped to 0.3%, verifying the effectiveness of pulse interference filtering.

[0034] This embodiment further simulates a narrowband interference scenario: Assume a new frequency converter is added to the factory area, generating narrowband interference with a center frequency of 2.4 GHz and a bandwidth of 5 MHz; the spectrum sensing unit detects a sharp increase in the mid-frequency spectrum energy to -28 dBm, and the bit error rate rises to... The retransmission rate increased to 12.3%; the probability distribution of the output of the lightweight convolutional neural network is... Narrowband interference has the highest probability, reaching 0.83; the system determines the current interference type to be narrowband interference; the adaptive anti-interference module starts the adaptive frequency hopping algorithm, and the frequency hopping mapping table switches to the pre-stored set of backup frequencies. 5.5GHz was selected as the new operating frequency; after the switch, the bit error rate decreased to The retransmission rate dropped to 1.1%, effectively avoiding narrowband interference.

[0035] This embodiment also simulates a multi-interference superposition scenario: simultaneous pulse interference and narrowband interference exist, and the interference power is high, with the recognition probability distribution being as follows: The highest probability of pulse interference was 0.45, but it did not exceed the 0.6 threshold. The system determined it to be uncertain and maintained the previous interference type judgment. Simultaneously, dual anti-interference measures were activated: the transmit power was gradually reduced in 0.5dB increments, from the initial 23dBm to 20dBm, and the interference power ratio decreased from 2.1dB to 0.8dB, below the threshold. At the same time, the modulation and coding scheme was downgraded from 64QAM to 16QAM, and the coding rate was reduced from 0.85 to 0.65. After comprehensive anti-interference processing, the communication link maintained normal connection, and the bit error rate was controlled within [a certain range]. The following measures meet basic communication requirements.

[0036] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for coordinated control, communication, and scheduling of unmanned aerial vehicles (UAVs) and robots, characterized in that, include: S101. Construct a heterogeneous converged communication architecture based on UWB ultra-broadband and V2X vehicle networking, and carry 6G short-range high-speed transmission protocol; S102. Real-time acquisition of spectrum energy, bit error rate and retransmission rate, and the use of lightweight convolutional neural network to identify the current industrial electromagnetic interference type; S103. Construct a dual-scale deep reinforcement learning model in the heterogeneous converged communication architecture. Based on the fusion data association table collected by UWB and V2X vehicle network, perform large-scale communication resource pre-allocation with superframe as the period, and perform small-scale resource dynamic adjustment and preemption scheduling with micro-time slot as the period. S104. Combine the scheduled communication resources output by the dual-scale deep reinforcement learning model to execute collaborative control commands between UAVs, ground mobile robots, and vehicle-road equipment.

2. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 1, characterized in that, S101 includes: The devices participating in the collaborative communication initialize the UWB ultra-wideband communication unit and the V2X vehicle-to-everything (V2X) communication unit respectively, and load the 6G short-range high-speed transmission protocol stack. The devices participating in the collaborative communication include drones, ground mobile robots, and vehicle-to-infrastructure (V2I) equipment. Each UWB ultra-wideband communication unit of a participating device sends a timestamped pulse signal to other participating devices in a preset polling order. The receiving device records the arrival time of the pulse signal, calculates the signal flight time using a two-sided bidirectional ranging method, generates the straight-line distance between the devices, and packages the distance values ​​measured by each device into a positioning data frame; Each V2X vehicle-to-everything (V2X) communication unit of participating devices periodically sends out data packets containing its own identifier, location, speed, direction of movement, and task status during the allocated broadcast time slots, and continuously listens to the channel to receive V2X data packets sent by other devices and extract the collaborative information in the V2X data packets. The distance values ​​obtained by the UWB ultra-wideband communication unit and the location and motion information obtained by the V2X vehicle-to-everything communication unit are timestamped to form a fusion data association table indexed by the device ID.

3. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 2, characterized in that, S102 includes: The spectral energy, bit error rate, and retransmission rate are concatenated into a three-dimensional feature vector, and after linear normalization, they are repeatedly arranged to form a single-channel feature map. A single-channel feature map is input into a lightweight convolutional neural network, and three feature mapping values ​​are output after passing through the convolutional layer. The three feature map values ​​output by the convolutional layer are flattened and then input into a fully connected layer, which outputs the probability distribution of each interference class. The fully connected layer uses the Softmax activation function. The category with the highest probability is selected as the current industrial electromagnetic interference type, the identification result is output, and the notch filter parameters are switched according to the identification result to dynamically adjust the transmission power and modulation coding scheme.

4. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 3, characterized in that, The step of switching notch filter parameters and dynamically adjusting transmit power and modulation coding scheme based on the identification results includes: If the identification result is impulse interference, then switch to a digital filter with notch characteristics, where the notch center frequency is set to the most recently counted interference frequency. If the identification result is narrowband interference, the frequency hopping map table is switched to a pre-stored set of spare frequency points; If the identification result indicates the presence of interference, the transmission power is gradually reduced according to the preset step size until the interference power ratio is lower than the threshold. If the identification result is no interference, then restore the transmit power and modulation and coding scheme to their initial values.

5. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 3, characterized in that, S103 includes: A dual-scale deep reinforcement learning model is constructed in a heterogeneous converged communication architecture, which includes a large-scale deep Q-network and a small-scale dual-delay deep deterministic policy gradient network. At the start of each superframe, the current queue length, average transmission latency in the most recent superframe, device motion speed and trajectory prediction values, and device service priority labels for each device are read from the fused data association table and concatenated to obtain a state vector; the expression of the state vector is as follows: ； In the formula, For the first The current queue length of each device For the first The average transmission latency of each device within the most recent superframe. For the first Predicted speed and trajectory values ​​of each device For the first Service priority tags for each device The total number of devices participating in collaborative communication; The state vector is input into a large-scale deep Q-network, and after forward propagation through multiple fully connected layers, the Q value corresponding to the allocation action of each time slot block is output. The action corresponding to the maximum Q value is selected as the pre-allocated time slot resource. The pre-allocated time slot resource includes the number of time slot blocks allocated to each device in the superframe, the starting time slot offset, and the length of the time slot block. The pre-allocated time slot resources are used as the initial reference allocation scheme for the small-scale dual-delay deep deterministic policy gradient network. The expected latency and throughput corresponding to the initial reference allocation scheme are used as the initial input of the value network, and the scheduled communication resources are generated through a small-scale dual-delay deep deterministic policy gradient network. At the end of each superframe, the small-scale dual-delay deep deterministic policy gradient network statistically analyzes the actual utilization rate of each micro-slot within the superframe and the backlog length of low-priority services caused by preemption. This data is then combined with the state collected by the large-scale deep Q-network at the beginning of the current superframe to form an empirical sample, which is stored in the empirical replay buffer. The expression for the reward function is as follows: ； In the formula, For the actual utilization rate of micro-slots, For low-priority business backlog length, For average transmission delay, , , These are the weighting coefficients; Once the number of samples in the experience replay buffer reaches a preset threshold, a batch of samples is randomly selected, the target Q value is calculated, and the parameters of the large-scale deep Q network are updated using the gradient descent method.

6. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 5, characterized in that, The step of using the expected latency and throughput corresponding to the initial reference allocation scheme as the initial input to the value network, and generating the scheduled communication resources through a small-scale dual-latency deep deterministic policy gradient network includes: The pre-allocated time slot resources are used as the initial reference allocation scheme for the small-scale dual-delay deep deterministic policy gradient network, and the expected delay and throughput corresponding to the initial reference allocation scheme are used as the initial inputs of the value network. At the beginning of each micro-slot, real-time channel state information and current interference type label are collected to detect whether high-priority burst services arrive. The real-time information and the allocation status of the corresponding micro-slots in the current pre-allocated time slot resources are spliced ​​together to form a small-scale state; The small-scale state is input into the policy network of the dual-delay deep deterministic policy gradient algorithm. The policy network outputs a continuous adjustment vector, which includes the adjustment amount for the current micro-slot allocation, the fine-tuning value of the transmit power, and the index change of the modulation and coding scheme. If a high-priority service is detected and the current micro-slot is already occupied by a low-priority service, the dual-delay deep deterministic policy gradient algorithm outputs a preemption action to generate scheduled communication resources. The preemption action reassigns the current micro-slot to the high-priority service and reschedules the preempted low-priority service to an idle micro-slot.

7. The UAV and robot vehicle-road cooperative control, communication, and scheduling method according to claim 5, characterized in that, S104 includes: According to the 6G short-range high-speed transmission protocol format, each control command generated based on the scheduled communication resources will be encapsulated into a short data frame. The receiving device parses the instruction content in the short data frame, converts the instruction content into a control signal that the actuator can recognize, and completes the execution of the instruction in the next control cycle.

8. A UAV and robot vehicle-road cooperative control, communication, and scheduling system, characterized in that, The method for unmanned aerial vehicle (UAV) and robot vehicle-road cooperative control, communication, and scheduling as described in any one of claims 1-7, wherein the system comprises a construction module, an identification module, a scheduling module, and an execution module; The building module is used to construct a heterogeneous converged communication architecture based on UWB ultra-wideband and V2X vehicle networking, and is equipped with a 6G short-range high-speed transmission protocol. The identification module is used to collect spectrum energy, bit error rate and retransmission rate in real time, and uses a lightweight convolutional neural network to identify the current type of industrial electromagnetic interference. The scheduling module is used to build a dual-scale deep reinforcement learning model in the heterogeneous converged communication architecture. Based on the fusion data association table collected by UWB and V2X vehicle networking, it performs large-scale communication resource pre-allocation with superframe as the period and small-scale resource dynamic adjustment and preemption scheduling with micro-time slot as the period. The execution module is used to combine the scheduled communication resources output by the dual-scale deep reinforcement learning model to execute collaborative control commands between UAVs, ground mobile robots, and vehicle-road equipment.

9. A drone and robot vehicle-road cooperative control, communication, and scheduling device, characterized in that, The UAV and robot vehicle-road cooperative control and communication scheduling device includes a memory and at least one processor. The memory stores instructions. The processor is used to call the instructions in the memory, causing the device to execute each step of the UAV and robot vehicle-road cooperative control and communication scheduling method as described in any one of claims 1-7.

Images