Unmanned aerial vehicle and ground robot cooperative control method and system

Abstract

The invention relates to the technical field of intelligent robot cooperative control, and discloses an unmanned aerial vehicle and ground robot cooperative control method and system, and the method comprises the steps: obtaining the state data and environment data of an unmanned aerial vehicle and a ground robot, and generating a verified cooperative state feature; outputting a cooperative control strategy through a dynamic strategy generation model based on the cooperative state characteristics; executing the cooperative control strategy, collecting strategy execution data in real time, and calculating a cooperative efficiency index based on the strategy execution data; when the collaborative efficiency index is lower than a self-adaptive threshold value, parameters of a dynamic strategy generation model are adjusted based on historical strategy execution data; and regenerating a cooperative control strategy by using the adjusted dynamic strategy generation model, and updating strategy execution data. According to the method, the control precision and task execution efficiency of the system in a dynamic environment can be remarkably improved, and meanwhile, the self-adaptive capacity and the long-term operation stability of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent robot cooperative control, and particularly relates to a UAV and ground robot cooperative control method and system. BACKGROUND

[0002] Current UAV and ground robot cooperative systems mostly use preset rules or fixed strategies for control. Such methods have poor adaptability when encountering dynamic obstacles or task changes. The system often needs to be manually re-planned or adjusted, and cannot achieve true autonomous cooperative control, which seriously limits the application effect in complex environments.

[0003] On the other hand, the prior art lacks an effective fusion mechanism for multi-source heterogeneous data, resulting in temporal and spatial inconsistency of state information. At the same time, most systems lack real-time performance evaluation and strategy optimization capabilities, and cannot continuously improve the control strategy during task execution, affecting the overall cooperative efficiency and task completion quality of the system. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a UAV and ground robot cooperative control method, which can solve the problem of scene adaptability defects.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a UAV and ground robot cooperative control method, comprising: acquiring state data of the UAV and the ground robot, and performing temporal and spatial consistency verification on the state data to generate verified cooperative state features; Based on the cooperative state features, a cooperative control strategy is output by a dynamic strategy generation model, including motion trajectory parameters, task allocation schemes and communication scheduling instructions; The cooperative control strategy is executed, and strategy execution data is collected in real time, and a cooperative performance index is calculated based on the strategy execution data; When the cooperative performance index is lower than the adaptive threshold, a strategy optimization mechanism is triggered, and the parameters of the dynamic strategy generation model are adjusted based on historical strategy execution data; The cooperative control strategy is regenerated using the adjusted dynamic strategy generation model, and the strategy execution data is updated, forming a closed-loop control process.

[0007] As a preferred scheme of the UAV and ground robot cooperative control method of the present application, the positioning data, attitude data and speed data of the UAV, and the joint state data, wheel speed data and environment perception data of the ground robot are collected. The time stamp alignment method is used for synchronously processing the positioning data and joint state data, and eliminating the time sequence dislocation caused by the collection time delay. The state features in the unified coordinates are subjected to the abnormal value detection and elimination, and the data fusion is carried out based on the Kalman filtering, so as to generate the verified cooperative state features.

[0008] As a preferred scheme of the unmanned aerial vehicle and ground robot cooperative control method, the cooperative state features are input into the dynamic strategy generation model, the dynamic strategy generation model comprises a task analysis module, a motion planning module and a communication coordination module. The task analysis module is used for decomposing the overall task into a subtask sequence, and an appropriate execution robot is allocated to each subtask. The motion planning module is used for generating the flight trajectory parameters of the unmanned aerial vehicle and the movement path parameters of the ground robot based on the subtask sequence. The communication coordination module is used for generating the communication scheduling instruction based on the trajectory parameters and the path parameters, and specifying the data transmission time and bandwidth allocation.

[0009] As a preferred scheme of the unmanned aerial vehicle and ground robot cooperative control method, the cooperative control strategy is executed, and the strategy execution data is collected in real time, and the cooperative efficiency index is calculated based on the strategy execution data, and the steps are as follows: The unmanned aerial vehicle and the ground robot are controlled to execute the task according to the motion trajectory parameters and the task allocation scheme, and the actual trajectory deviation, the task completion time and the communication success rate are recorded; The trajectory tracking accuracy is calculated based on the actual trajectory deviation, the task execution efficiency is calculated based on the task completion time, and the communication reliability is calculated based on the communication success rate; The trajectory tracking accuracy, the task execution efficiency and the communication reliability are weighted and summed to generate the cooperative efficiency index.

[0010] As a preferred scheme of the unmanned aerial vehicle and ground robot cooperative control method, when the cooperative efficiency index is lower than the adaptive threshold value, the key features leading to the low efficiency are extracted from the historical strategy execution data; The gradient of each parameter in the dynamic strategy generation model is calculated by using the gradient back propagation algorithm, and the parameter value is adjusted according to the gradient direction; The dynamic strategy generation model is updated in an incremental learning manner.

[0011] As a preferred scheme of the unmanned aerial vehicle and ground robot cooperative control method, the current cooperative state features are input into the adjusted dynamic strategy generation model, and the updated cooperative control strategy is output; The updated cooperative control strategy is executed, and new strategy execution data is collected; Add new policy execution data to historical policy execution data for subsequent model adjustment; Repeat the steps of policy generation, execution, evaluation and optimization to realize continuous improvement of closed-loop control.

[0012] As a preferred scheme of the unmanned aerial vehicle and ground robot collaborative control method, a policy execution traceability mechanism is constructed, when the collaboration failure is detected, the key link in the policy execution chain is automatically analyzed and optimization suggestions are generated; Based on the optimization suggestions, the generation logic of the collaborative control strategy is adjusted to improve the robustness of the strategy.

[0013] As a preferred scheme of the unmanned aerial vehicle and ground robot collaborative control method, the decision nodes and execution results in the policy execution process are recorded; When the collaboration failure occurs, the decision chain is traced in reverse to identify the key decision nodes that cause the failure; Based on the historical data of the key decision nodes, the possible results under different decision paths are simulated to generate optimization suggestions; The optimization suggestions are fed back to the dynamic policy generation model to guide the subsequent policy generation.

[0014] In a second aspect, the present application provides an unmanned aerial vehicle and ground robot collaborative control system, comprising: a feature generation unit for obtaining state data of the unmanned aerial vehicle and the ground robot, and performing spatiotemporal consistency verification on the state data to generate verified collaborative state features; A policy output unit is used to output a collaborative control strategy based on the collaborative state features through a dynamic policy generation model, the collaborative control strategy including motion trajectory parameters, task allocation schemes and communication scheduling instructions; An index calculation unit is used to execute the collaborative control strategy, collect policy execution data in real time, and calculate collaborative efficiency indexes based on the policy execution data; A parameter adjustment unit is used to trigger a policy optimization mechanism when the collaborative efficiency indexes are lower than an adaptive threshold, and adjust parameters of the dynamic policy generation model based on historical policy execution data; A policy generation unit is used to regenerate the collaborative control strategy using the adjusted dynamic policy generation model, and update the policy execution data to form a closed-loop control process.

[0015] In a third aspect, the present application provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements an unmanned aerial vehicle and ground robot collaborative control method when executing the program.

[0016] Compared with the prior art, the present application has the following advantages: The application generates a cooperative state feature through space-time consistency verification, fuses heterogeneous data from different platforms, different rates and different coordinate systems into a single fact source which is unified, reliable and has strict space-time correlation, and lays a solid foundation for subsequent high-precision cooperative control from the information source, and reduces the control error caused by data conflict. Secondly, the application uses a dynamic strategy generation model to output integrated strategy, and performs cross-field cooperation and coupling optimization on task allocation, motion planning and communication scheduling. The model can generate integrated strategy which is matched and enhanced in motion trajectory, task division and communication resources based on unified cooperative state feature. The internal consistency makes the actions of unmanned aerial vehicles and ground robots highly cooperative, thereby systematically improving the smoothness and overall efficiency of task execution in dynamic environment. Finally, the application introduces a closed-loop optimization mechanism with cooperative performance index as feedback and model parameter online adjustment as core. This makes the system an intelligent agent that can learn and continuously improve from actual execution effect. When the environment changes or the internal state changes, the system can automatically trigger optimization and adjust its decision model, so that the next decision can better adapt to the new scene. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A schematic diagram of a cooperative control method of unmanned aerial vehicles and ground robots is provided for the present application. Figure 2 A schematic diagram of a cooperative control system of unmanned aerial vehicles and ground robots is provided for the present application. Figure 3 A schematic diagram of an electronic device is provided for the present application. DETAILED DESCRIPTION

[0019] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0020] Embodiment 1, refer to Figure 1For the first embodiment of the present application, the embodiment provides a method for cooperative control of unmanned aerial vehicles and ground robots, comprising: S1, acquiring state data of unmanned aerial vehicles and ground robots, and performing spatio-temporal consistency verification on the state data to generate verified cooperative state features; S2, based on the cooperative state features, outputting a cooperative control strategy through a dynamic strategy generation model, the cooperative control strategy including motion trajectory parameters, task allocation schemes and communication scheduling instructions; S3, executing the cooperative control strategy and collecting real-time strategy execution data, and calculating cooperative performance indicators based on the strategy execution data; S4, when the cooperative performance indicators are lower than the adaptive threshold, triggering a strategy optimization mechanism, and adjusting the parameters of the dynamic strategy generation model based on historical strategy execution data; S5, using the adjusted dynamic strategy generation model to regenerate the cooperative control strategy, and updating the strategy execution data to form a closed-loop control process.

[0021] The specific step S1 includes the following sub-steps: Collecting positioning data, attitude data and speed data of the unmanned aerial vehicles, and joint state data, wheel speed data and environmental perception data of the ground robots; Synchronizing the positioning data and joint state data using a timestamp alignment method to eliminate the time sequence misalignment caused by collection delay; Performing outlier detection and elimination on the state features in the unified coordinate, and performing data fusion based on Kalman filtering to generate the verified cooperative state features.

[0022] Specifically, in actual application, unmanned aerial vehicles and ground robots are respectively equipped with different types of sensors. Due to differences in sampling frequency, transmission delay and coordinate systems, the collected data may have spatio-temporal inconsistency problems. For example, the GPS positioning data of the unmanned aerial vehicle is updated at a frequency of 10Hz, while the laser radar data of the ground robot is updated at a frequency of 20Hz, resulting in a time synchronization problem between the two. At the same time, the unmanned aerial vehicle uses a global coordinate system, while the ground robot uses a local coordinate system, resulting in a problem of spatial coordinate inconsistency.

[0023] The positioning data, attitude data and speed data of the unmanned aerial vehicle, and the joint state data, wheel speed data and environment perception data of the ground robot are collected. Among them, the positioning data of the unmanned aerial vehicle is obtained through GPS and inertial navigation system, and the accuracy can reach centimeter level; the attitude data is obtained through gyroscope and accelerometer, including pitch angle, roll angle and yaw angle; the speed data is obtained through airspeed meter and ground speed meter. The joint state data of the ground robot is obtained through encoder, including the angle and angular velocity of each joint; the wheel speed data is obtained through wheel encoder; the environment perception data is obtained through laser radar, camera and ultrasonic sensor, including obstacle distance, terrain features and environment texture and other information.

[0024] The time stamp alignment method is used to synchronize the positioning data and joint state data, and eliminate the time sequence dislocation caused by the collection delay. In specific implementation, high-precision time stamps are added to all sensor data, and linear interpolation method is used to unify different frequency data to the same time sequence. For example, the positioning data of the unmanned aerial vehicle at 10 Hz is interpolated to 20 Hz, which is synchronized with the sensor data of the ground robot. At the same time, the sliding window method is used to cache the data, ensuring the real-time performance of data processing, and the window size is set to 500 milliseconds, which can effectively eliminate the influence of transmission delay.

[0025] The positioning data of the unmanned aerial vehicle is converted to the local coordinate system of the ground robot through spatial coordinate transformation, and the state features in the unified coordinate are generated. First, a local coordinate system with the initial position of the ground robot as the origin is established, and then the global coordinates of the unmanned aerial vehicle are converted to the local coordinate system through coordinate transformation matrix. The coordinate transformation includes two steps of translation and rotation, the translation vector is the coordinates of the initial position of the ground robot in the global coordinate system, and the rotation matrix is determined by the initial orientation of the ground robot. After coordinate transformation, the position data of the unmanned aerial vehicle and the ground robot can be processed and displayed uniformly in the same coordinate system.

[0026] The state features in the unified coordinate are subjected to outlier detection and elimination, and data fusion is performed based on Kalman filtering to generate the verified collaborative state features. The outlier detection adopts the 3σ criterion, and the data exceeding three times the standard deviation of the mean is eliminated. Kalman filtering fuses multi-source data through state equation and observation equation. The state equation is based on the kinematic model of the unmanned aerial vehicle and the ground robot, describing the dynamic evolution law of the position, velocity and other states over time. The observation equation establishes the mathematical relationship between the system state and the measurement values of various sensors (such as GPS, IMU, wheel speed meter, etc.). The core of this data fusion process is that the Kalman filter, as an optimal estimator, can dynamically and optimally combine the inaccurate model prediction and noisy sensor observations through the recursive cycle of "prediction-update". Specifically, in each step, the filter first makes a prediction based on the state equation to obtain a priori estimate with large uncertainty. Then, using the actual sensor observation values at the current time, the clean data obtained by the above-mentioned abnormal elimination is mainly used to correct the priori estimate to obtain a posteriori optimal estimate with significantly reduced uncertainty. Through this continuous optimization process of weighted average of prediction and observation, the system can effectively smooth the high-frequency random noise of the sensor, and utilize the complementary characteristics of different sensors (such as the long-term stability of GPS and the short-term accuracy of IMU) to correct each other, so as to suppress and enhance the random error of a single sensor inherent to ±2 meters to ±0.3 meters of collaborative positioning accuracy. This provides accurate, smooth and reliable state information for subsequent collaborative control.

[0027] The specific step S2 includes the following sub-steps: The collaborative state features are input into a dynamic strategy generation model, which includes a task analysis module, a motion planning module, and a communication coordination module. The task analysis module decomposes the overall task into a sub-task sequence and assigns an appropriate execution robot to each sub-task. The motion planning module generates the flight trajectory parameters of the unmanned aerial vehicle and the movement path parameters of the ground robot based on the sub-task sequence. The communication coordination module generates communication scheduling instructions based on the trajectory parameters and path parameters to specify data transmission timing and bandwidth allocation.

[0028] Specifically, the dynamic strategy generation model adopts a deep neural network architecture, including an input layer, three hidden layers, and an output layer. The input layer receives a 128-dimensional cooperative state feature vector, the hidden layers contain 256, 512, and 256 neurons respectively, and use ReLU activation functions. The output layer outputs the corresponding control strategy parameters according to the task requirements. The model uses a dataset containing 100,000 samples in the training phase, uses the Adam optimizer with a learning rate of 0.001, and after 100 epochs of training, the accuracy on the test set reaches more than 95%.

[0029] The cooperative state features are input into the dynamic strategy generation model, which includes a task analysis module, a motion planning module, and a communication coordination module. The task analysis module uses an attention mechanism and can automatically identify key task elements according to the current environmental features and task requirements. The motion planning module generates optimal motion trajectories while considering the robot's dynamics constraints based on the A algorithm and the dynamic window method. The communication coordination module uses the TDMA (Time Division Multiple Access) protocol to allocate communication time slots for different robots to avoid communication conflicts. These three modules work together to ensure that the generated cooperative control strategy meets the task requirements and conforms to the actual motion capabilities of the robots.

[0030] The task analysis module decomposes the overall task into a sequence of subtasks and assigns appropriate execution robots to each subtask. For example, in the embankment inspection task, the overall task is decomposed into embankment patrol, hazard identification, and hazard disposal subtasks. The task analysis module assigns tasks based on the capabilities of each robot. Drones are responsible for large-scale patrol and preliminary hazard identification (such as piping, cracks, and leakage points) due to their high-altitude perspective and rapid maneuverability. Ground robots are responsible for approaching suspected hazard points, conducting close-range detailed inspections, marking, and performing preliminary disposal actions (such as placing markers and throwing slope-stabilizing materials) upon instruction. The task assignment considers factors such as the remaining battery life and sensor performance of the robots, and uses the Hungarian algorithm to find the optimal allocation scheme to maximize task execution efficiency.

[0031] The motion planning module generates the flight trajectory parameters of the drone and the movement path parameters of the ground robot based on the sequence of subtasks. For drones, a Bezier curve trajectory containing parameters such as waypoint coordinates, flight height, and speed is generated. For ground robots, a spline curve path containing parameters such as path points, turning angles, and speed is generated. When planning the trajectory, obstacle avoidance constraints are considered, and the artificial potential field method is used to generate an attractive force at the target point and a repulsive force at the obstacle to ensure the safety of the trajectory. The generated trajectory is smooth and continuous, with a maximum curvature that does not exceed the limit of the robot's motion capability.

[0032] Based on trajectory and path parameters, the communication coordination module generates communication scheduling instructions, specifying data transmission timing and bandwidth allocation. Communication scheduling employs an adaptive time slot allocation mechanism, dynamically adjusting communication priorities according to task urgency. For example, during the target identification phase, the UAV needs to transmit image data in real time, allocating a higher communication bandwidth (2MHz) and frequent communication time slots (once every 100ms); during the cruise phase, only status information needs to be transmitted, allocating a lower communication bandwidth (500kHz) and sparser communication time slots (once every 1s). This adaptive scheduling mechanism ensures timely transmission of critical data while conserving communication resources.

[0033] The specific step S3 includes the following sub-steps: Control drones and ground robots to execute tasks according to motion trajectory parameters and task allocation schemes, and record actual trajectory deviations, task completion times and communication success rates; The trajectory tracking accuracy is calculated based on the actual trajectory deviation, the task execution efficiency is calculated based on the task completion time, and the communication reliability is calculated based on the communication success rate. The collaborative performance index is generated by weighting and summing the trajectory tracking accuracy, task execution efficiency, and communication reliability.

[0034] Specifically, the collaborative effectiveness index objectively evaluates the execution effect of collaborative control strategies through a structured, multi-level quantitative indicator system. This system decomposes the abstract collaborative effect into measurable and calculable specific indicators, and aggregates them into a comprehensive score through weighted summation. Taking a regional search task as an example: the task execution effectiveness weight is 0.4. Among its sub-indicators, the task completion rate scores 1.0 for successfully searching all target areas, execution accuracy scores 0.89 for correctly identifying 16 out of 18 preset targets, and timeliness scores 0.82 for the actual completion time of 85 seconds exceeding the baseline time of 70 seconds. The arithmetic average of these three is approximately 0.90, which, multiplied by the weight, yields a contribution value of 0.36. The resource utilization effectiveness weight is 0.3. Energy efficiency scores 0.80 because total energy consumption is 125% of the baseline value, and communication efficiency scores 0.95 because 95% of key data packets are successfully transmitted. The average of these two is approximately 0.875, which, multiplied by the weight, yields a contribution value of approximately 0.26. The collaborative effectiveness weight is 0.3. Among its sub-indicators, trajectory tracking accuracy scores 0.92 due to its average error being within 92% of the allowable threshold; obstacle avoidance success rate scores 1.0 due to avoiding all 10 obstacles; and data consistency scores 0.98 due to a map matching degree of 98%. The average of these three is approximately 0.967, and after multiplying by the weight, the contribution value is approximately 0.29. Adding the contribution values ​​of these three primary indicators, the final collaborative effectiveness index value for this task execution is 0.91. This value directly reflects the execution effect of the collaborative strategy. If the system's adaptive threshold is set at 0.85, the strategy evaluation is excellent, and the relevant data will be saved as positive samples. If the index value drops below 0.75, the strategy optimization mechanism will be triggered. The system can further trace and analyze the specific links where effectiveness is low, such as resource utilization efficiency being only 0.60, and pinpointing energy consumption efficiency as low as 0.50 as the key reason. This allows for precise and targeted optimization of the strategy, demonstrating the effective support of the evaluation system for the system's closed-loop optimization process.

[0035] The system controls the drone and ground robot to execute tasks according to motion trajectory parameters and task allocation schemes, and records the actual trajectory deviation, task completion time, and communication success rate. The actual trajectory deviation is calculated by comparing the root mean square error of the planned trajectory and the actual trajectory, using high-precision RTK-GPS for trajectory sampling at a sampling frequency of 10Hz. The task completion time is the time interval from task start to task end, precisely timed using the system clock. The communication success rate is calculated by statistically analyzing the proportion of successfully transmitted data packets out of the total number of transmitted data packets, using CRC checksum to ensure data accuracy.

[0036] Tracking accuracy is calculated based on actual trajectory deviation, task execution efficiency is calculated based on task completion time, and communication reliability is calculated based on communication success rate. The formula for calculating trajectory tracking accuracy is: 1 / (1+RMSE), where RMSE is the root mean square error of the trajectory, in meters. The formula for calculating task execution efficiency is: baseline time / actual completion time, where the baseline time is the ideal completion time statistically derived from historical data. Communication reliability is directly calculated using the communication success rate. All these indicators are normalized to the range of 0-1 for easier subsequent weighted calculation.

[0037] A collaborative performance index is generated by weighting and summing trajectory tracking accuracy, task execution efficiency, and communication reliability. The weights of the three indicators are determined using an expert scoring method, and are 0.4, 0.35, and 0.25, respectively. The calculation formula for the collaborative performance index is: 0.4 × trajectory tracking accuracy + 0.35 × task execution efficiency + 0.25 × communication reliability. A higher index value indicates better collaborative performance. When the index value is below 0.6, the collaborative performance is considered unsatisfactory, requiring the triggering of a strategy optimization mechanism. In practical applications, this index can accurately reflect the collaborative performance of the system and provide a quantitative basis for strategy optimization.

[0038] The specific step S4 includes the following sub-steps: When the collaborative performance index is lower than the adaptive threshold, extract the key features that lead to low performance from historical strategy execution data; The gradient of each parameter in the dynamic strategy generation model is calculated by using the gradient backpropagation algorithm, and the parameter values ​​are adjusted according to the gradient direction. The dynamic policy generation model is updated using an incremental learning approach.

[0039] Specifically, the adaptive threshold is dynamically adjusted based on task type and environmental complexity. In simple environments, the threshold is set to 0.7; in complex environments, it is set to 0.6. Environmental complexity is comprehensively evaluated using indicators such as obstacle density, the number of dynamic obstacles, and the intensity of communication interference. When the collaborative performance indicator falls below the threshold for three consecutive control cycles, a strategy optimization mechanism is triggered. This setting ensures both timely system response to performance degradation and avoids frequent optimizations caused by momentary fluctuations.

[0040] When the collaborative performance index falls below the adaptive threshold, key features leading to low performance are extracted from historical strategy execution data. Historical strategy execution data is stored in a circular buffer, containing the 1000 most recent records. Each record includes state characteristics, control strategy, execution result, and performance index. Principal component analysis (PCA) is used to extract key features, identifying the top 10 feature dimensions with the greatest impact on performance index. Simultaneously, a decision tree algorithm is used to analyze the decision paths leading to low performance, identifying weaknesses in the strategy. These analyses provide clear direction for subsequent model parameter optimization.

[0041] The gradients of each parameter in the dynamically generated model are calculated using the gradient backpropagation algorithm, and the parameter values ​​are adjusted according to the gradient direction. Batch gradient descent is used for gradient calculation, with a batch size of 32. An adaptive adjustment strategy is employed, with an initial value of 0.001, which is halved if there is no improvement after five consecutive iterations. A momentum term is added during parameter updates, with a momentum coefficient set to 0.9 to accelerate the convergence process. Simultaneously, gradient clipping is used to limit the gradient norm to within 1.0 to prevent gradient explosion. These measures ensure the stability and efficiency of parameter optimization.

[0042] Incremental learning is employed to update the dynamic policy generation model. Incremental learning fine-tunes the model using new policy execution data while retaining existing knowledge. Specifically, the parameters of the first few layers are frozen, and only the parameters of the last two layers are updated. This approach absorbs new experience while avoiding catastrophic forgetting. The model is updated online, with an incremental update performed every 100 new valid data points. During the update process, an early stopping mechanism is used to prevent overfitting; training stops when the validation set loss no longer decreases after 10 consecutive iterations.

[0043] The specific step S5 includes the following sub-steps: The current cooperative state features are input into the adjusted dynamic policy generation model, and the updated cooperative control policy is output. Execute the updated collaborative control strategy and collect new strategy execution data; Add new strategy execution data to the historical strategy execution data for subsequent model adjustments; Repeatedly execute the steps of strategy generation, execution, evaluation, and optimization to achieve closed-loop control for continuous improvement.

[0044] Specifically, the adjusted model needs to be validated before deployment. The validation process uses a reserved test dataset containing 1000 samples, covering various typical scenarios. Validation metrics include policy generation time, policy execution success rate, and performance improvement. The requirements are: policy generation time no more than 50ms, policy execution success rate no less than 90%, and performance improvement of at least 10% compared to before optimization. Only models that meet these conditions will be deployed to the actual system.

[0045] The current cooperative state features are input into the adjusted dynamic policy generation model, which outputs an updated cooperative control policy. The model inference process is executed on an embedded GPU to ensure real-time performance. The input state features are first standardized and then forward-propagated through each layer of the neural network. The output layer uses the Softmax function to generate the probability distribution of the policy parameters and selects the policy with the highest probability as the output. The entire inference process is completed within 30ms, meeting the requirements of real-time control. The output control policy includes detailed motion and communication parameters and can be directly sent to the actuator.

[0046] The updated collaborative control strategy is executed, and new strategy execution data is collected. During strategy execution, the robot's status data, environmental data, and control commands are recorded in real time. Data acquisition frequency is 20Hz, and timestamps are used to ensure data synchronization. Simultaneously, key events in strategy execution, such as task start, subtask switching, and anomaly occurrences, are recorded. This data is used not only to calculate performance metrics but also as historical data for subsequent model optimization. Data storage adopts a hierarchical structure: raw data is stored for 30 days, feature data for 90 days, and statistical metrics are permanently stored.

[0047] New strategy execution data is added to the historical strategy execution data for subsequent model adjustments. Quality checks are performed during data addition to remove outliers and duplicates. Simultaneously, the data is labeled, including information such as task type, environmental conditions, and execution results. The total amount of historical data is maintained at around 10,000 records, and an in-first-out (FIFO) strategy is used for updates. Historical data is periodically statistically analyzed to calculate the average performance index for each type of task, serving as a benchmark for system performance evaluation. This data provides rich training samples for continuous optimization.

[0048] The system iteratively executes strategy generation, execution, evaluation, and optimization steps to achieve continuous improvement through closed-loop control. Each control cycle is set to 500ms, including 100ms for strategy generation, 300ms for strategy execution, and 100ms for evaluation and optimization. The system maintains a control cycle counter, which is reset every 24 hours. Through this closed-loop control mechanism, the system can continuously adapt to environmental changes and optimize its control strategy. Experiments show that after 50 control cycles of learning, the system's performance indicators on the same task can be improved by more than 25%.

[0049] In some embodiments, the collaborative control method between a drone and a ground robot may further include the following steps: Construct a strategy execution traceability mechanism that automatically analyzes key links in the strategy execution chain and generates optimization suggestions when a coordination failure is detected. The generation logic of the collaborative control strategy is adjusted based on optimization suggestions to improve the robustness of the strategy.

[0050] Specifically, the strategy execution tracing mechanism is based on a distributed log system, recording data from the entire process from strategy generation to execution. Log data includes timestamps, robot IDs, operation types, input parameters, and output results. The log system uses structured storage, supporting rapid querying and analysis. When a collaborative failure occurs, the system can quickly pinpoint the source of the problem, providing a basis for optimization. The implementation of the tracing mechanism reduces problem diagnosis time from hours to minutes.

[0051] In some embodiments, the collaborative control method between a drone and a ground robot may further include the following steps: Record the decision-making nodes and execution results during the strategy execution process; When a collaborative failure occurs, the decision-making chain is traced in reverse to identify the key decision-making nodes that led to the failure. Based on historical data of key decision-making nodes, simulate possible outcomes under different decision-making paths and generate optimization suggestions; The optimization suggestions are fed back to the dynamic policy generation model to guide subsequent policy generation.

[0052] Specifically, the system records decision nodes and execution results during strategy execution. Decision nodes include task allocation decisions, path planning decisions, and communication scheduling decisions. Each decision node records information such as decision time, decision basis, and decision result. Execution results include actual trajectory data, task completion status, and resource consumption data. These data are linked through timestamps to form a complete decision-execution chain. Data recording uses a lightweight format, with the record size of a single decision node not exceeding 1KB, ensuring that system overhead remains within acceptable limits.

[0053] When a collaborative failure occurs, the decision-making chain is traced backward to identify the key decision-making nodes that led to the failure. Backward tracing begins with observed anomalies and follows the data flow backward until the initial erroneous decision is found. For example, when a collision occurs, the trajectory data at the collision point is started, and the path planning decisions and obstacle detection data are traced backward to find the root cause of the collision. Key decision-making nodes are determined through impact analysis, calculating the weight of each decision-making node's influence on the final outcome, and selecting the node with the highest weight as the key node.

[0054] Based on historical data from key decision-making nodes, the possible outcomes under different decision paths are simulated, generating optimization suggestions. The simulation process uses digital twin technology to build virtual models of the robot and its environment. The decision-making process is then re-executed in this virtual environment, trying different decision options and observing their impact on the final result. Optimization suggestions include specific parameter adjustment schemes, algorithm improvement suggestions, and strategy modification opinions. For example, when path planning is found to be too aggressive, increasing the safety distance parameter is suggested; when task allocation is found to be unreasonable, adjusting the utility function weights is recommended.

[0055] The optimized suggestions are fed back to the dynamic policy generation model to guide subsequent policy generation. The feedback mechanism includes two methods: direct parameter adjustment and indirect empirical learning. Direct parameter adjustment modifies model parameters directly based on the optimized suggestions, such as adjusting the weight coefficients of the cost function. Indirect empirical learning transforms the optimized suggestions into training samples and updates the model through incremental learning. Simultaneously, a suggestion effectiveness evaluation mechanism is established to statistically analyze the implementation effects of different types of suggestions and continuously improve the suggestion generation algorithm. Through this feedback mechanism, the system can continuously accumulate experience and improve decision-making quality.

[0056] Example 2, refer to Figure 2 This is a second embodiment of the present invention, which provides a collaborative control system for a UAV and a ground robot, comprising: a feature generation unit, used to acquire state data and environmental data of the UAV and the ground robot, and to perform spatiotemporal consistency verification on the state data to generate verified collaborative state features; a strategy output unit, used to output a collaborative control strategy based on the collaborative state features through a dynamic strategy generation model, the collaborative control strategy including motion trajectory parameters, task allocation schemes, and communication scheduling instructions; an index calculation unit, used to execute the collaborative control strategy, and to collect strategy execution data in real time, and to calculate a collaborative performance index based on the strategy execution data; a parameter adjustment unit, used to trigger a strategy optimization mechanism when the collaborative performance index is lower than an adaptive threshold, and to adjust the parameters of the dynamic strategy generation model based on historical strategy execution data; and a strategy generation unit, used to regenerate the collaborative control strategy using the adjusted dynamic strategy generation model, and to update the strategy execution data to form a closed-loop control process.

[0057] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a method for collaborative control of a drone and a ground robot.

[0058] like Figure 3 As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communication bus 640. The processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions from the memory 630 to execute a collaborative control method between a drone and a ground robot.

[0059] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute a method for collaborative control of a drone and a ground robot.

[0061] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a method for collaborative control of an unmanned aerial vehicle and a ground robot.

[0062] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

Claims

1. A method for collaborative control of a drone and a ground robot, characterized in that, include: Acquire state data from the UAV and ground robot, perform spatiotemporal consistency verification on the state data, and generate verified collaborative state features; Based on the aforementioned collaborative state characteristics, a collaborative control strategy is output through a dynamic strategy generation model. The collaborative control strategy includes motion trajectory parameters, task allocation scheme, and communication scheduling instructions. The collaborative control strategy is executed, and strategy execution data is collected in real time. Based on the strategy execution data, a collaborative performance index is calculated. When the collaborative performance index is lower than the adaptive threshold, the strategy optimization mechanism is triggered to adjust the parameters of the dynamic strategy generation model based on historical strategy execution data. The adjusted dynamic strategy generation model is used to regenerate the collaborative control strategy and update the strategy execution data to form a closed-loop control process.

2. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, The steps of acquiring state data from the UAV and ground robot, performing spatiotemporal consistency verification on the state data, and generating verified collaborative state features are as follows: Collect positioning, attitude, and speed data from drones, as well as joint status, wheel speed, and environmental perception data from ground robots; The positioning data and the joint status data are synchronized using a timestamp alignment method to eliminate timing misalignment caused by acquisition delay; Outlier detection and removal are performed on the state features under unified coordinates, and data fusion is performed based on Kalman filtering to generate the verified collaborative state features.

3. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, Based on the aforementioned cooperative state characteristics, the steps for outputting a cooperative control strategy through a dynamic strategy generation model are as follows: The cooperative state features are input into the dynamic strategy generation model, which includes a task parsing module, a motion planning module, and a communication coordination module. The overall task is decomposed into a sequence of sub-tasks by the task parsing module, and a suitable execution robot is assigned to each sub-task. The motion planning module generates flight trajectory parameters for the UAV and movement path parameters for the ground robot based on the sub-task sequence. The communication coordination module generates communication scheduling instructions based on trajectory parameters and path parameters, specifying the timing of data transmission and bandwidth allocation.

4. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, The steps of executing the collaborative control strategy, collecting strategy execution data in real time, and calculating collaborative performance indicators based on the strategy execution data are as follows: Control the drone and ground robot to execute tasks according to the motion trajectory parameters and task allocation scheme, and record the actual trajectory deviation, task completion time and communication success rate; The trajectory tracking accuracy is calculated based on the actual trajectory deviation, the task execution efficiency is calculated based on the task completion time, and the communication reliability is calculated based on the communication success rate. The collaborative performance index is generated by weighted summation of the trajectory tracking accuracy, the task execution efficiency, and the communication reliability.

5. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, When the collaborative performance index falls below the adaptive threshold, a strategy optimization mechanism is triggered. The steps involved adjusting the parameters of the dynamic strategy generation model based on historical strategy execution data are as follows: When the collaborative performance index is lower than the adaptive threshold, key features that lead to low performance are extracted from historical strategy execution data. The gradient of each parameter in the dynamic strategy generation model is calculated using the gradient backpropagation algorithm, and the parameter values ​​are adjusted according to the gradient direction. The dynamic policy generation model is updated using an incremental learning approach.

6. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, The steps for regenerating the cooperative control strategy using the adjusted dynamic strategy generation model and updating the strategy execution data to form a closed-loop control process are as follows: The current cooperative state features are input into the adjusted dynamic policy generation model, and the updated cooperative control policy is output. Execute the updated collaborative control strategy and collect new strategy execution data; The new strategy execution data is added to the historical strategy execution data for subsequent model adjustments; Repeatedly execute the steps of strategy generation, execution, evaluation, and optimization to achieve closed-loop control for continuous improvement.

7. The method for collaborative control of a UAV and a ground robot as described in claim 1, characterized in that, The method further includes: Construct a strategy execution traceability mechanism that automatically analyzes key links in the strategy execution chain and generates optimization suggestions when a coordination failure is detected. The generation logic of the collaborative control strategy is adjusted based on optimization suggestions to improve the robustness of the strategy.

8. The method for collaborative control of a UAV and a ground robot as described in claim 7, characterized in that, The specific steps for building a policy execution tracing mechanism are as follows: Record the decision-making nodes and execution results during the strategy execution process; When a collaborative failure occurs, the decision-making chain is traced in reverse to identify the key decision-making nodes that led to the failure. Based on historical data of the key decision nodes, simulate possible outcomes under different decision paths and generate optimization suggestions; The optimization suggestions are fed back to the dynamic policy generation model to guide subsequent policy generation.

9. A collaborative control system for unmanned aerial vehicles (UAVs) and ground robots, characterized in that, Implementing a collaborative control method for a drone and a ground robot as described in any one of claims 1 to 8, comprising: The feature generation unit is used to acquire state data of the UAV and the ground robot, perform spatiotemporal consistency verification on the state data, and generate verified collaborative state features. The strategy output unit is used to output a collaborative control strategy based on the collaborative state characteristics through a dynamic strategy generation model. The collaborative control strategy includes motion trajectory parameters, task allocation scheme and communication scheduling instructions. The indicator calculation unit is used to execute the collaborative control strategy, collect strategy execution data in real time, and calculate collaborative performance indicators based on the strategy execution data. The parameter adjustment unit is used to trigger the strategy optimization mechanism when the collaborative performance index is lower than the adaptive threshold, and adjust the parameters of the dynamic strategy generation model based on historical strategy execution data. The strategy generation unit is used to regenerate the collaborative control strategy using the adjusted dynamic strategy generation model and update the strategy execution data to form a closed-loop control process.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a collaborative control method for unmanned aerial vehicles and ground robots as described in any one of claims 1 to 8.

Images