Tracked robot intelligent shooting control system with dynamic target locking function

Abstract

The invention relates to the technical field of robots, and discloses a tracked robot intelligent shooting control system with a dynamic target locking function, which dynamically adjusts the working mode of a visual sensor based on real-time target position change and environmental parameters. The optimal shooting angle is calculated according to the adjusted visual sensor data, and real-time regulation and control are carried out; correcting shooting strength according to dynamic data of wind speed and wind direction to adapt to external environment change; and the priority is determined according to a multi-target motion trail prediction algorithm, and the shooting sequence is optimized. Through the scheme of the embodiment of the invention, the problem of how to regulate and control the shooting angle according to the real-time target position change so as to solve the problem of insufficient shooting precision can be solved; visual sensor parameters are regulated and controlled according to ambient light and reflection characteristics so as to solve the problem of low target recognition accuracy, and shooting strength is regulated and controlled according to dynamic data of wind speed and wind direction so as to solve the problem that the hit rate is reduced due to external environment influence.

Description

Technical Field

[0001] The invention relates to the field of robotics, and more specifically, to an intelligent shooting control system for a tracked robot with a dynamic target locking function. Background Art

[0002] An intelligent shooting control method for a tracked robot with a dynamic target locking function aims to achieve efficient tracking and accurate shooting of dynamic targets through intelligent algorithms and technical means. This method combines visual sensors, an environmental perception system, and a motion prediction algorithm to strive for higher shooting accuracy and efficiency in a changing environment.

[0003] However, this method faces several key challenges: First, there are still deficiencies in real-time adjustment of the shooting angle to adapt to the rapid changes in the target position, which will affect the final shooting accuracy; Second, the performance of the visual sensor is unstable under different light conditions, making it difficult to improve the target recognition accuracy; Third, in an environment with wind interference, the fluctuations in wind speed and direction will affect the shooting force, thereby reducing the hit rate.

[0004] In addition, when facing multiple moving targets, the existing multi-target motion trajectory prediction algorithms may lead to a lag in shooting selection and an inability to quickly make the optimal shooting decision; Finally, during continuous shooting, the cumulative error in the shooting feedback data will gradually increase the deviation of the aiming system. To address these problems, targeted regulation mechanisms and technical solutions must be developed to ensure that this shooting control method has higher reliability and effectiveness in actual combat applications.

[0005] Practical Information Content

[0006] To solve the above technical problems, the invention provides the following technical solutions:

[0007] An intelligent shooting control system for a tracked robot with a dynamic target locking function, comprising:

[0008] Dynamically adjust the working mode of the visual sensor based on the real-time target position change and environmental parameters, calculate the optimal shooting angle according to the adjusted visual sensor data and perform real-time regulation, correct the shooting force according to the dynamic data of wind speed and direction to adapt to external environmental changes, and determine the priority and optimize the shooting order according to the multi-target motion trajectory prediction algorithm;

[0009] Obtain the real-time target position parameters (X, Y, Z), where (X, Y) are the target plane coordinates and Z is the height. Based on the formula \(\theta_{new}=\theta_{old}+\Delta X\cdot K_1+\Delta Y\cdot K_2+\Delta Z\cdot K_3\), adjust the shooting angle \(\theta\) in real time. Here, \(K_1\), \(K_2\), and \(K_3\) are the horizontal, lateral, and vertical adjustment coefficients respectively. Adjust the light intensity threshold P of the vision sensor according to the ambient light sensor data. If the target recognition rate R is lower than the threshold T, i.e., \(R < T\), adjust the light compensation factor F and recalculate the light intensity threshold.

[0010] Preferably, detect the light reflection intensity I and the target surface reflectivity \(\gamma\) in real time, and update the contrast C of the vision sensor. Set the contrast gain G to enhance the target distinguishability in the high-reflection area. Reset the visual contrast C based on the formula \(C_{new}=C\times(G + I)\), and dynamically correct the ballistic curve equation according to the wind speed v and wind direction a.

[0011] Preferably, capture the multi-target trajectory prediction information and the corresponding risk assessment matrix \(\mathbf{M}\) for making decisions on the optimal priority sequence S. Sort the sequence S through an iterative optimization algorithm until the condition \(\sum w_i\times(\text{Distance}_{gi}(T))\leq D_{max}\) is satisfied. Here, \(w_i\) is the weighted weight of target i, \(\text{Distance}_{gi}(T)\) represents the predicted minimum approach distance within the given time T, and \(D_{max}\) is the set safety spacing limit. After each shooting operation, record the feedback error \(\varepsilon\) and accumulate it into the total deviation \(\Delta\).

[0012] If \(\Delta\) exceeds the pre-determined threshold \(\Omega\), trigger the adaptive learning model to fine-tune the calculation method of the next shelling parameter set \(\varPhi\), i.e., \(\varPhi=\varPhi+\beta\times\varepsilon\), where \(\beta\) is the system sensitivity adjustment factor.

[0013] Preferably, record the hit point coordinates H and the offset vector d = (\(\Delta x\), \(\Delta y\)) of the ideal aiming point A for each shooting event.

[0014] Define the error function \(E(k)=\sum_{i = k}^n{((\Delta x_i-\Delta x_k)+(\Delta y_i-\Delta y_k))^2}\) to calculate the relative deviation between the k-th bullet and the previous N consecutive shootings, and use this value as the compensation basis.

[0015] Dynamically adjust the basic yaw correction value \(\varLambda\) for the next round of firing according to the result of the previous step.

[0016] If the expected hit probability Pr(Δ|F_prev) of the current round is less than η, the energy output U of the next fired bullet is reduced to reduce unnecessary consumption while ensuring accurate hit. Here, F_prev represents the previous feedback situation, η is the acceptable low-precision tolerance level, and U controls the launch kinetic energy level.

[0017] Preferably, when all tasks of the current round are completed or a specific situation is encountered, switch to the standby mode and clear all temporary calculation data;

[0018] Use the Kalman filter to smooth the noise interference part in the difference image between the new and old visual frames, so as to improve the target tracking accuracy and enhance the robustness;

[0019] Use the three-dimensional point cloud mapping method to model the motion trend in a complex background, so that the subsequent shooting is more in line with the expected trend of dynamic changes;

[0020] Introduce the time-weighted moving average (Twma) model, and calculate the new displacement correction term Δ according to the formula \[Delta x=\mu\Delta x+(1\mu)\Delta x], so as to refine the specific position selection scheme for each shot. Here, μ∈(0,1) is used as the weighted ratio coefficient to ensure that past performance has enough influence on future decisions and is not overly sensitive.

[0021] Preferably, the reinforcement learning framework guides the uncertain factors in the automatic calibration process of machine learning, such as unpredictable variables such as changes in lighting conditions or probability distributions of accidental disturbances;

[0022] For each observed action response, give positive and negative reward scores, and establish a Q-Value state-action value table for subsequent training to improve performance;

[0023] In the face of unknown or extreme weather conditions, build a more accurate and reliable environmental simulation prediction through a random forest regression tree ensemble model, especially for the analysis of the impact of wind;

[0024] When a significantly large shooting dispersion phenomenon is found and accompanied by a significant deviation from the original planned route, that is, when σ>(σ_0+r*δσ) holds (here, σ_0 is the baseline deviation fluctuation, r is the factor determining the increment ratio, and δσ describes the increase amplitude of the standard deviation), additional protective measures should be initiated to limit the occurrence of dangerous situations.

[0025] Preferably, a multi-channel laser rangefinder is equipped to obtain the distance information D_set of multiple different reflecting surfaces in the front space in real time, and a high-resolution map image is generated accordingly to provide a reliable geographical reference system for subsequent shooting preparation;

[0026] Use the particle swarm optimization algorithm to solve the relative benefit score e of each potential strike point in a complex environment, and select the set of local optimal solutions;

[0027] Adjust the PSO parameters according to the characteristics of the actual test environment, including parameters such as the maximum flight speed \(v_{\text{max}}\), the trust coefficient gbest_factor, and the interaction frequency \(n_{\text{iter}}\);

[0028] A standardized update rule based on the formula \([n_j=(v^{T}_{j} / \parallel v_j\parallel)_{\text{clip}}]\) is set to regulate the position of individual particles, so that the overall evolution direction can be more scientifically, reasonably and orderly advanced to the vicinity of the global optimal solution. Here, \(v\) is the velocity vector in each dimension and \(\parallel.\parallel\) represents the Euclidean norm, and \(()_{\text{clip}}\) is limited to the interval \((-1,1)\).

[0029] Preferably, combine the deep belief network DBN pre-training initialization strategy to improve the convergence efficiency of the deep neural network structure, and ensure that the model can quickly enter a good state for application;

[0030] Add a long short-term memory module LSTM to capture the changing characteristics of sequence dependence and non-stationary behavior patterns;

[0031] Guide the activation threshold \(m\) of the LSTM output unit to gradually approach the most suitable state. Set the threshold \(\tau\) and the incremental rate \(\varphi\) according to experience. When \(t > (1+\varphi)^{k\tau}\), increase the output activation degree;

[0032] Try a variety of possible architecture combination experiments and verifications, such as hybrid types like double-layer BiLSTM plus convolutional pooling kernels, etc., and continuously search for the ideal configuration form most suitable for target positioning and shooting control tasks, and record it for direct reference in future similar projects.

[0033] Preferably, construct a multi-dimensional situation awareness library CAS, which is composed of three parts: the information \(W_{\text{fore}}\) returned by the weather forecast API interface, the data \(Mil_{\text{info}}\) uploaded by the military situation monitoring station, and the user-defined preference Profile;

[0034] Regularly synchronize and integrate the latest external data to update the internal database, and ensure that all important factors at the macro and micro levels can be fully considered in the combat planning and decision-making process;

[0035] Extract key indicators to construct an evaluation system KPI, online monitor the changes of each element and give periodic evaluation feedback on the overall effect;

[0036] Use the rule chain Rule_chain to parse the complex linkage scenario rule set, and infer the next operation instruction sequence Command_Set according to the comprehensive evaluation score score. Whenever the situation of score < \((Th+\rho)*avg\_score\) occurs.

[0037] Preferably, a fuzzy logic system is introduced to assist in adjusting the shooting accuracy, especially effectively improving the quality of the final result in the case of high uncertainty and difficulty in precise quantification.

[0038] A series of input / output membership functions are confirmed to describe the closeness of the relationship between various dimensions and the flexibility of boundary conversion. For example, forms of representation translated from natural language words such as near, far, bright, dark, and left deviation, right deviation, etc.

[0039] A series of inference rules in the form of IF THEN formulated in cooperation with expert experience are used to implement the fuzzy transformation rule table Table_F, which intuitively and clearly shows what specific action instructions should be taken in various situations, greatly improving the applicability of the system.

[0040] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0041] 1. Using advanced image processing and motion estimation algorithms to accurately capture the position of a fast-moving or variably accelerating target object. The system can continuously monitor the position of the target and can react in the shortest time to reset the shooting angle. This enables precise strikes even in a high-speed dynamic environment. In addition, through the instant evaluation of the shooting effect each time (i.e., shooting result feedback), the future shooting instructions are continuously calibrated, thereby minimizing the risk of missing the target due to sudden changes in the direction or speed of the target.

[0042] 2. By adjusting functions such as the gain, exposure time, and color calibration of the imaging device. It also introduces an adaptive enhancement processing method to make the imaging closer to the real color and reduce the possibility of shadow interference. More importantly, the machine learning method is adopted to allow the machine to self-train according to empirical data, so as to better distinguish various complex texture structures or hidden objects under a camouflaged background, and can automatically select the best imaging mode suitable for the current environmental conditions to ensure good detection performance at any time.

[0043] 3. By integrating a miniaturized meteorological detection station device, relatively accurate weather forecast data, especially the local wind field distribution characteristics, can be obtained before shooting, and key parameters such as the launch time and initial velocity are determined in advance based on this as a reference. On this basis, a complete simulation platform is established using the principles of fluid mechanics for simulation testing. A series of experimental verifications show that this solution can effectively alleviate the deviation phenomenon caused by natural factors. In addition, considering the situation that the wind speed and direction may fluctuate greatly over time, an online fine-tuning link is added before each trigger to further reduce random errors, and finally ensure that the shooting hit accuracy is always at a relatively ideal level. Brief Description of the Drawings

[0044] Figure 1 This is a schematic diagram of the specific process of the present invention;

[0045] Figure 2 This is a schematic diagram of the process for obtaining the target position parameters of the present invention;

[0046] Figure 3 This is a schematic diagram of the real-time light reflection intensity process of the present invention;

[0047] Figure 4 This is a schematic diagram of the process for capturing the target trajectory of the present invention;

[0048] Figure 5 This is a schematic diagram of the shooting record process of the present invention;

[0049] Figure 6 This is a schematic diagram of the process for calculating situation data of the present invention;

[0050] Figure 7 This is a schematic diagram of the learning and automatic calibration process of the present invention;

[0051] Figure 8 This is a schematic diagram of the image processing process of the present invention;

[0052] Figure 9 This is a schematic diagram of the network policy processing process of the present invention;

[0053] Figure 10 This is a schematic diagram of the process for collecting the situation awareness library of the present invention;

[0054] Figure 11 This is a schematic diagram of the process for assisting shooting accuracy of the present invention. Detailed implementation manners

[0055] Next, the technical solutions in the embodiments of the invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the invention. Obviously, the described embodiments are only a part of the embodiments of the invention, rather than all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the invention.

[0056] Please refer to Figure 1 - Figure 11, A smart shooting control system for a tracked robot with dynamic target locking function, including several core steps, which aim to improve the intelligent shooting accuracy and response speed of the tracked robot. Firstly, it is to adjust the working mode of the vision sensor based on the real-time target position change and environmental parameters. Specifically, when the vision sensor receives an instruction, it not only needs to continuously track the identified target, but also can quickly identify and adapt to the changes of suddenly emerging targets or backgrounds in the new environment. For example, if the target moves from indoors to a place with weaker outdoor light, the algorithm will command the sensor to automatically reduce the gain and increase the exposure time to ensure clear imaging. On the contrary, when the light is too strong, it will correspondingly increase the ISO value and use a shorter shutter interval to reduce noise and maintain image sharpness.

[0057] After clearly understanding the specific situation of the target, the next step is how to obtain the ideal launch angle based on the adjusted data and be able to fine-tune it at any time. This link calculates the most appropriate angle by integrating the Kalman filter and the deep neural network prediction model to achieve the best strike effect and reduce deviation. For example, in one embodiment, once the system detects that the enemy unit is approaching or moving away from its own position quickly in a preset direction, the program will calculate the corresponding elevation angle and azimuth angle increment based on this information, and accurately control the rotating chassis and lifting mechanism through the motor to make the barrel point accurately. At the same time, the software platform is also constantly monitoring the position relationship of its own position relative to the reference point to ensure that the correction parameters are always up-to-date and avoid the cumulative error caused by long-term combat.

[0058] At the same time, it cannot be ignored that external meteorological factors, especially the changes in wind speed and wind direction, will have a significant impact on the actual trajectory of the projectile. Therefore, it is also crucial to scientifically utilize this part of the data to adjust the launch power. To achieve this, the developers specifically designed a set of data sets transmitted by a real-time optical meteorological station based on the Bayesian classification method, and then gave reasonable initial velocity modification amount suggestions for the control system to refer to. Imagine such a scenario: if there is a crosswind in the desert area, then according to the calculation results of the formula, the powder loading can be appropriately increased to push the bullet to overcome the cross drift; and under the strong sea breeze blowing head-on on the plain, the energy should be reduced to make the flight path stable so as not to fly too high and cause missed targets or even self-injury.

[0059] Another noteworthy aspect is how to effectively sort shooting priorities in a multi-target environment. Here, an improved version of the A* path search technology is adopted, supplemented by long short-term memory recurrent units to form a hybrid model to estimate various possible action trends within the next second. Simply put, it's like when playing a video game and being besieged by a group of enemies, one has to first take out the nearest, easiest-to-defeat opponent with a relatively high threat level. More specifically, by calculating multiple important factors such as the distance, speed, and the level of protective devices carried by each potential hit individual through weighted averaging, a comprehensive evaluation criterion is obtained as the basis for ranking. Finally, the object ranked first is designated as the immediate firing target, and only when all high-risk items are cleared will it turn to the next priority to deal with the remaining enemies.

[0060] Finally, to overcome the problem of the inevitable cumulative small displacements caused by consecutive operations, which lead to a decrease in overall accuracy, we fully considered the shooting feedback information and proposed a compensation and calibration scheme to self-adjust the internal parameter configuration to ensure stability. That is to say, whether it is the vibration generated before and after each launch or the slight deformation of the gaps between parts caused by sudden temperature changes and other reasons, it can rely on post-statistical analysis to find the pattern, form a compensation formula, and then send it back to the CPU to wait for the opportunity to be applied. For example, in a simulated drill of a long-lasting battle, after five consecutive shots, it was found that the ballistic trajectory was slightly deflected to the left. Then this function was activated, and after comparing the difference in measurements before and after, timely correction was made, avoiding subsequent misjudgments and wasting unnecessary consumption of materials and resources, while greatly enhancing the reliability level and allowing the user to focus more on strategic arrangements rather than worrying about the lack of basic functions.

[0061] Next, it is described that the tracked robot of the present invention continuously obtains the precise position parameters (X, Y, Z) of the target in space through the equipped sensing device. Among them, (X, Y) are two-dimensional plane coordinates to locate the specific positions in the horizontal and depth directions where the target is located; Z represents the height value, which determines the vertical position of the target relative to the robot. This data set ensures that the shooting system can make precise responses to moving objects.

[0062] Subsequently, after receiving new position information, a shooting angle adjustment operation is performed according to the formula \(\theta_{\text{new}}=\theta_{\text{old}}+\Delta X\cdot K_1+\Delta Y\cdot K_2+\Delta Z\cdot K_3\). Here, \(\Delta X\), \(\Delta Y\), and \(\Delta Z\) are the target displacement change amounts obtained from two consecutive measurements respectively; \(K_1\) (optimal range: 0.1 - 0.4), \(K_2\) (0.15 - 0.5), and \(K_3\) (0.05 - 0.3) are proportionality coefficients that control the adjustment degree of their respective corresponding axes and ensure appropriate sensitivity in all directions. For example, in one embodiment, when a large vertical distance change is detected during the movement of the drone in the air, a smaller \(K_3\) value allows for more delicate angle correction without causing excessive interference to other orientations.

[0063] In addition, the light intensity threshold \(P\) for adjusting the sensitivity setting of the image vision sensor is also adjusted based on the input provided by the surrounding environmental fiber optic sensor. This is to ensure a better balance between recognition efficiency and quality. If the target recognition rate is lower than the preset standard \(T\) (such as 75%), at this time, by optimizing the light compensation factor \(F\) and resetting the working conditions of the vision sensor, the image processing effect can be further improved. Specifically, if the robot is operating in a relatively dim or uneven light source distribution situation, this step will help it better perceive the target features and thus maintain stable tracking.

[0064] Furthermore, in this system, the first step is to real-time detect the ambient light reflection intensity \(I\) and the target surface reflectivity \(\gamma\), and use this information to update the contrast \(C\) adopted by the force sense sensor. This step means that the robot can adapt to different light environments and targets with different color textures, ensuring that it can accurately capture and lock the target regardless of how the conditions change. The reflection intensity \(I\) depends on the external lighting conditions, while the reflectivity \(\gamma\) is determined by the material properties of the target. The change ranges of these two parameters are quite broad, but to achieve the best results, usually, the reflection intensity is ensured to be within the linear range that the sensor can perceive, and the reflectivity should cover different types of surface features as much as possible.

[0065] Subsequently, a contrast gain \(G\) is set, aiming to increase the degree of difference between the areas with high-intensity light reflection in the visual system and the surrounding environment. The contrast value \(C\) is re-determined using the formula \(C_{\text{new}} = C\times(G + I)\). In this calculation, \(G\) is an artificially introduced factor used as a weight value to emphasize the characteristics of a specific scene; the reasonable value of the gain \(G\) needs to be based on the specific application scenario. In the optimal case, it should maximize the visual sensing performance without causing over-saturation or information loss. For example, in a strong light environment, a higher \(G\) value can better highlight the target object, but in low light, a lower \(G\) may be selected to maintain sensitivity.

[0066] Then, it enters the third processing step, which involves mathematical modeling for ballistic compensation according to the current meteorological data, including the actually measured wind speed \(v\) and direction \(a\), in order to avoid excessive ballistic deviation caused by external factors. In this step, a suitable correction formula is deduced based on fluid mechanics and archery principles. For example, when operating a shooting outdoors in an open area (such as during a shooting range test or a military operation), if there is a strong wind, the distortion of the flight path of the projectile caused by these variables must be considered to ensure that the final shooting hit point is close to the predetermined coordinate point.

[0067] The next fourth part is to evaluate whether the environmental impact exceeds the permitted limit. The system will check whether the wind force will interfere with the aiming and adjust the decision-making strategy accordingly. Once the adverse change amount of the wind on the shooting accuracy exceeds the critical threshold \(T_v\), that is, when \(v\sin(\alpha)>T_v\) holds, an alarm signal will be issued to draw attention, and measures will be taken to slowly reduce the correction intensity to prevent reducing the overall combat ability due to overreaction. This mechanism aims to balance the relationship between the response speed and accuracy, neither causing waste of resources due to premature intervention nor increasing the failure probability due to delayed response.

[0068] Furthermore, when the tracked robot is in an urban search and rescue mission and uses an intelligent shooting control system to combat the risk of explosion caused by fire, the robot first needs to obtain the on-site light reflection information through the daylight / light detection element integrated on the camera, and adjust its own observation acuity accordingly. If you encounter a target made of high-brightness reflective material such as an emergency evacuation sign, setting an appropriately high \(G) will make the sign more conspicuous and easy to identify. In addition, if a large airflow suddenly blows towards the robot's route, it will automatically determine the angle of the gust and the degree of displacement that may be caused based on the pre-input map data and sensor readings. Assuming that the calculation shows that the impact of the wind is strong enough to trigger the warning threshold, the robot will immediately slow down the action frequency to avoid yaw, while keeping communication open and waiting for subsequent instructions to resume normal operation. In this way, it is ensured that each shot can accurately strike the preset enemy or danger source within the expected range.

[0069] In the tracked robot intelligent shooting control method with dynamic target locking function, multi-target trajectory prediction information and its corresponding risk assessment matrix \(mathbf{M}) are first captured. This information helps to construct the optimal priority sequence \(S) to determine the order of attacking targets. Specifically, in a multi-target environment, the system can predict the movement path of each potential attack target and evaluate the possible danger it may bring. Each entry in the risk assessment matrix represents the risk value of different targets under specific conditions. By analyzing these factors, the tracked robot provides itself with an orderly target processing order, thereby improving execution efficiency.

[0070] Next, based on the establishment of the above sequence, there will be a process of continuous optimization of the sorting to ensure that the overall security meets the standards. The iterative algorithm adjusts the weight ratio of each target in the sequence \(S) \(w_i) and the expected minimum distance between them and their own equipment at the future time (represented by the symbol \(text{Distance}_{gi}(T))) so that the product accumulation sum of all selected attack targets does not exceed the set maximum safety interval limit \(D_{max}). The formula is \[sum w_i*(text{Distance}_{gi}(T))≤D_{max}], where \(w_i\in(0,1]) reflects the degree of importance attached to a certain target, \(text{Distance}_{gi}(T)\geq0) represents the shortest approach distance calculated over time, and \(D_{max}>0) is determined by the mission scenario requirements. This ensures that even when facing complex targets at the same time, there is always enough reaction time and space for effective response actions.

[0071] After each operation, the actual hitting situation is checked for a gap relative to the expected value - the so-called feedback error. This difference is accumulated as the total deviation Δ and added to the overall consideration. For example, after a tracked device launches a round of attacks on hostile drones within a predetermined coordinate, if it is found that the impact position deviates from the original center by more than the allowable range, then this additional displacement will be counted into Δ. This can record the cumulative consequences of a series of consecutive operations and determine whether the warning level is reached based on this. If the increase in Δ exceeds a previously established standard Ω, then the built-in learning mode is activated to self-adjust the subsequent action guidelines.

[0072] I. Once it is detected that the cumulative deviation has reached the limiting condition, a fine-tuning program is launched to change the subsequent firing parameter set Φ. The calculation formula is: Φ = Φ + β * ε. In this relationship, β is a preset coefficient used to measure the system sensitivity and determines how to respond to external error changes; while ε corresponds to the immediate deviation component newly added in a single cycle. In one embodiment, if a combat drill shows that the bullet landing point often appears at the edge of the pre-aiming area, then according to this mechanism, the aiming direction and force are automatically adjusted to gradually reduce the error range until the accuracy requirement is met. This continuous adaptation and optimization method enables the entire system to perform excellently under actual combat conditions and have good growth and scalability.

[0073] For each shooting event, record the coordinate H of the hitting point and the offset vector d = (Δx, Δy) of the ideal aiming point A. Specifically, when the tracked robot makes a shot, record the differences in the X and Y directions between the actual hitting position and the expected ideal target point. For example, in a certain experiment, if the hitting point is (15, 20) and the ideal aiming point is set at (18, 23), then the generated offset vector will be recorded as (-3, -3).

[0074] To more precisely evaluate the multi-shot performance over a period of time, an error function \[E(k)=\sum_{i = k}^{n}{((\Delta x_i-\Delta x_k)+(\Delta y_i-\Delta y_k))^2}\] is defined. This function is used to analyze the relative deviation relationship between the projectile fired at the \(k\)-th shot and the most recent \(N\) shots, to help judge the shooting consistency and discover potential trend changes. In this formula, \(n\) represents the total number of historical shots involved in the calculation; \(k\) is the serial number of the most recently included shot in this calculation (ranging from 1 to infinity); and \((\Delta x_i,\Delta y_i)\) correspond to the specific horizontal and vertical direction errors of each individual shooting case, and these values are usually within a reasonable range to maintain the stable convergence of the system. This error function can highlight the location of abnormally deviated points through the cumulative sum of squares, which is particularly crucial for ensuring long-term stable precision improvement. Selecting this specific form can magnify larger errors, so that even a small number of serious mistakes will not be overlooked.

[0075] Dynamically adjust the basic yaw correction value \(\Lambda\) during the next round of firing according to the above obtained results. That is to say, adjust the future aiming angle by combining the data calculated by the error function, so as to gradually approach the most ideal result and optimize the shooting path and precision control logic. Suppose it is calculated that there is a phenomenon of being left of the due south continuously for multiple times, then the subsequent shooting instructions can be appropriately adjusted to the right by a certain angle to compensate for the cumulative error. This can make the shooting more accurate, gradually eliminate the influence caused by random factors, and then improve the overall hit quality.

[0076] If the expected hit probability \(Pr(\Delta|F_{prev})<\eta\) for the current round holds, then it is necessary to appropriately reduce the bullet launch kinetic energy \(U\) to avoid unnecessary resource waste while ensuring the hit performance requirements. Among them, \(F_{prev}\) represents the summary of the effects of the previous several shots and their corresponding parameters as a reference condition; \(\eta\) is set as a predefined number less than 1 and used as the boundary for measuring whether the minimum tolerable threshold is met. For example, suppose it is predicted at this stage that the next bullet may hit outside the area more than the radius \(r\) away from the bull's-eye, and the historical feedback information indicates that this is not an isolated accident but a continuous problem. At this time, the shooting intensity should be appropriately reduced, that is, the energy setting should be lowered until the high-level operation range is restored. Doing so can not only reduce consumption but also help maintain the reliability and stable operation state of the system.

[0077] In one embodiment, a series of live-fire exercise scenarios are considered. While the sensors of the tracked robot monitor its own movement, external weather conditions, and other dynamic factor changes, they can accurately perceive the specific location of the target. According to the established algorithm process, the operations of the foregoing steps are first performed, that is, immediately after each shooting action, the gap between the actual and predicted impact points is obtained. Subsequently, by introducing a non-linear error evaluation mechanism, abnormal deviations are screened and necessary compensation measures are determined; finally, the shooting force is flexibly controlled according to the actual situation. Such an iterative update process enables the intelligent shooting control system to continuously learn and accumulate experience to achieve self-evolution and improvement.

[0078] Further, when a round of tasks is completed or a specific situation is encountered, the system switches to the standby mode, and all temporary data will be cleared in this state. This step is to ensure that it will not be affected by the residual information of the previous operation when facing new challenges. For example, when conducting reconnaissance in a new combat environment, the interference factors such as the previous terrain and targets can be completely removed.

[0079] After switching the mode, the Kalman filter is used to process the visual information. This filter is used to analyze the change differences between the new and old image frames and effectively reduce the unstable signals caused by factors such as external noise. In this way, the tracking accuracy is greatly improved and the overall performance becomes more stable and reliable. Taking a specific embodiment as an example, when searching for a moving enemy in a forest camouflage background, through this technology, the relationship between the target and the background can be more clearly distinguished, and the probability of misidentification can be reduced. Among them, the change difference between the new and old image frames here refers to the part of the pixel intensity change between two consecutive images. The changes of these pixels can help determine the position and morphological changes of the dynamic target. The noise interference is usually composed of image noise caused by natural light flickering, the characteristics of the sensor itself, or other irrelevant movements.

[0080] To better predict and respond to the behavioral changes of the environment and the target, in the modeling link, the three-dimensional point cloud mapping method is used to make a detailed interpretation of the complex spatial layout around and the dynamic characteristics of the internal objects. Specifically, when tracking a humanoid unit that makes a quick turn or suddenly starts in the urban streets and alleys, the possible action trends of the other party can be judged in advance according to the trajectory. This method not only increases the possibility of grasping the shooting opportunity but also reduces the risk of pre-aiming mistakes.

[0081] Finally, the Time-Weighted Moving Average (Twma) model is incorporated to calculate the position deviation correction. In the formula \(\Delta x=\mu*\Delta x_{prev}+(1 - \mu)*\Delta x_{next}\), \(\mu\) is a value within the range of 0 to 1. The parameter \(\mu\) represents the importance comparison weight between the displacement difference in the previous moment (denoted as prev) and the expected position in the upcoming next moment. Setting such a coefficient \(\mu\) within the range of 0 to 1 is to enable the historical shooting performance to appropriately influence future operations without being overly biased towards one side. For example, when adjusting the aiming direction during confrontation, if the previous projectiles are slightly left-biased, the subsequent launches can be made more concentrated and accurate by increasing the right correction value; in the opposite case, similar adjustments can be made to maintain a good attack posture. Here, the value of \(\mu\) generally depends on the specific requirements in the actual application environment and the experimental verification results to obtain the optimal solution, and usually needs to be finely adjusted in combination with different types of combat environments and the motion characteristics of the targets faced. In this example, if it is found through multiple rounds of experiments that choosing \(\mu\) to be around 0.7 can achieve the best effect in a specific environment, the shooting accuracy will be optimized based on this in this scenario.

[0082] Furthermore, a smart shooting control system for a tracked robot with a dynamic target locking function, in which the reinforcement learning framework guides the machine to automatically adjust the uncertain parameter effects caused by the dependent external force. This means that when environmental conditions (such as lighting changes or unpredictable factors) change, the robot uses this framework to self-adjust its performance to maintain efficiency and accuracy. Specifically, whenever a new scenario response is encountered, the actions will be evaluated, and positive or negative reward scores will be given according to the results, thereby optimizing the strategy; in this way, experience is continuously accumulated and the Q-Value state-action value table is updated to ensure better performance in future similar situations. The Q-Value will directly affect the selection tendency of the same decision in the next similar event, which is particularly important during the shooting control process.

[0083] For example, in the sudden situation where the light suddenly weakens or an external object hits the tracked robot, causing its posture to deviate from the ideal track, the above mechanism can timely adjust parameters such as direction and speed to reposition the target, ensuring accurate hitting of the target while reducing the misjudgment probability caused by environmental noise.

[0084] When faced with elusive or extreme climate change, an ensemble learning model based on the random forest algorithm is constructed to enhance the depth of understanding of the external environment. In particular, when analyzing the trend of trajectory changes under different types of wind speeds that may occur in a large area, it provides strong support for improving prediction accuracy. In this scenario, the training set should cover as much meteorological element information recorded in the past as possible as input vectors to capture potential patterns. The number of trees in the random forest can be adjusted according to the actual situation, and the default recommended value ranges from 10 to 100, and the optimal configuration number under the best performance is found through cross-validation; this can significantly improve the reliability of the system to operate stably under unfamiliar conditions.

[0085] When it is detected that there is an over-dispersion phenomenon in the shooting result, that is, the dispersion area exceeds the expected safety range, and there is a serious deviation in the movement route beyond the allowable threshold, it indicates that there are uncontrollable factors interfering in the current environment. At this time, special safety mechanisms need to be intervened to protect the system. According to the formula σ > (σ_0 + r * δσ), where σ represents the measured dispersion degree (the distance difference between points within a unit length), the base limit value σ_0 defines the minimum error level that can be accepted under normal conditions, and the range usually starts from zero; r represents a coefficient used to determine the amplification magnitude ratio, and its ideal range should be selected at an appropriate position between 0 and 2 to achieve the maximum efficiency. If it is too large, the conditions will be too loose, and if it is too small, the response will be slow, increasing the probability of false alarms; finally, δσ reflects the standard deviation increment caused by random perturbations. If the calculated result is greater than or equal to the critical boundary, it is considered necessary to activate additional constraint measures to prevent the spread of potential damage, such as automatically triggering the braking function, decelerating and emitting warning sound signals, etc. Specifically, if the tracked unmanned combat vehicle encounters a strong lateral gust of wind during the execution of the precise attack command, causing a large drift in the turret pointing angle, the corresponding protection mechanism should be immediately activated to adjust the attitude until it meets the standard before resuming the normal operation process.

[0086] A multi-channel laser rangefinder is equipped at the front end of the tracked robot to continuously, quickly and accurately measure the distance information D_set of multiple reflecting surfaces in the front space. These data are integrated and analyzed and then converted into a high-resolution map to ensure that accurate and reliable geographical information is available as a reference for subsequent steps. For example, in the military exercise site test, this device can quickly capture the positions of surrounding obstacles such as rocks, vegetation or temporary structures and depict them on the electronic map for further evaluation of the positions and characteristics of potential shooting targets.

[0087] Based on the above-obtained distance information and other relevant data sources, input the particle swarm optimization algorithm to calculate the relative effectiveness score \(e\) brought about by implementing strike actions at each possible threat object. This score depends on various factors, such as the difficulty of the strike, the risk level, and the possible consequences, etc., thus constituting a series of alternative options for the system to decide which areas or points to prioritize for handling. In one embodiment, considering the shooting difficulty caused by the enemy's hiding place and the loss probability of accidentally hitting key facilities, different values are obtained to characterize their value. Then, a set formed by selecting the local best answers is prepared for subsequent judgment.

[0088] Due to the diverse and complex application scenarios, the internal parameters of PSO must be dynamically adjusted according to specific situations rather than being fixed. The key items involved are the maximum flight speed \(v_m\), the trust coefficient \(gbest\_factor\), and the interaction frequency \(n\_iter\) (representing the number of iterative updates). Specifically speaking, when facing the search and attack tasks in a narrow and variable environment like an urban block compared to an open desert scenario, a differentiated parameter set should be set to ensure the exertion of adaptability.

[0089] Then, integrate the long short-term memory module (LSTM) on the established basis. This processing method aims to introduce a mechanism to capture the temporal correlation between data and its changing characteristics, and is particularly suitable for analyzing the characteristics of behavior sequences evolving over time. When it comes to the task of tracking and shooting multiple dynamic targets that are constantly moving, LSTM can record the trajectory information of the target in the past period of time and predict the position at the next moment. For example, when a patrol vehicle suddenly changes its traveling path and accelerates away or approaches a certain direction, this algorithm can help calculate a reasonable firing window and reduce misjudgment.

[0090] Subsequently, guide the output unit activation threshold \(m\) of LSTM to approach the most suitable working state. To achieve this goal, a threshold \(\tau\) and an incremental rate \(\varphi\) are set to adjust the output sensitivity to ensure that the output can be reasonably enhanced as the conditions change. In the empirical formula mentioned here, \(t\) is the real-time step count or the system cycle sequence number, used to measure how long it has been since the last state update; \(k\) is obtained from the fitting of experimental data, reflecting the speed of growth; and setting that when \(t\gt(1 + \varphi)^{\text{k}\tau}\), increasing the output activation means that as the iteration progresses, the reaction intensity of the system gradually tends to an optimal level - if it is too sensitive, it may cause frequent but meaningless operations, and if it is too dull, it cannot respond to new situations immediately. Continuing with the above example, during the continuous tracking of the vehicle, the detection range will be gradually strengthened or fire support measures will be prepared due to the increase in surrounding interference until the safety strike accuracy reaches the ideal standard.

[0091] Finally, different hybrid architectures are tried to find the best solution. For example, a two-layer bidirectional long short-term memory neuron (BiLSTM) is combined with some specific structures such as convolutional kernels or pooling units to test which combination is more suitable for specific business requirements. For dynamic aiming weapons, this may mean selecting a set that can operate efficiently and stably under multi-perspective input conditions by comparing multiple algorithm arrangements in a simulated complex combat situation, and recording these valuable results for future project reference. For example, a proprietary image perception and decision logic link is developed for the complex and variable terrain characteristics in urban block search operations, which not only improves the adaptability of the overall system but also accumulates transferable experience and knowledge.

[0092] Furthermore, an intelligent shooting control system for a tracked robot with dynamic target locking function includes constructing a multi-dimensional situation awareness library CAS, updating the internal database by regularly synchronizing the latest external data, building an evaluation system KPI from the key indicators extracted from these data, and using a rule chain Rule_chain to deduce an appropriate sequence of operation instructions Command_Set. During this period, if the evaluation score is lower than the threshold, a special warning signal will be triggered and intervention suggestions will be provided to the commander.

[0093] The first step is to construct a multi-dimensional situation awareness library CAS. This step aims to integrate multiple information sources to ensure that the robot system can obtain a comprehensive information background to support combat planning. Specifically, CAS covers the information provided by the weather forecast API interface (W_fore), such as influencing factors like wind level, rainfall, cloud thickness, etc.; data uploaded from military situation monitoring stations (Mil_info), which includes battlefield geographical location characteristics, the number and movement of enemies, etc.; and also includes user-defined preferences (Profile). For example, in one embodiment, to better prepare a robot for a counter-terrorism mission in the mountains, the rainfall probability in the area in the last three days is collected as 45%, and the temperature ranges from 18 degrees to 23 degrees; the enemy armed strongholds are active at night and tend to use motorcycles to transport equipment, and accordingly, the robot's tactical deployment mode and search path planning logic are adjusted.

[0094] Following that is to keep the multi-dimensional situation awareness library CAS up-to-date. By regularly integrating the latest weather forecast API information and on-site military intelligence, the internal stored data is continuously refreshed to formulate more accurate combat guidelines. Specifically, at each integer time unit, the system will automatically execute an update process to ensure that all available information is taken into account. For example, when it is known that a sandstorm is about to hit or the enemy's reinforcement force has arrived, the combat plan should be modified in advance to avoid unnecessary losses.

[0095] Next, key performance indicators (KPIs) are extracted from the dataset collected by the CAS, and a mechanism for real-time tracking of the changes in each element and overall performance level assessment is established. Each KPI should be able to reflect the impact of the corresponding environmental or operational characteristics on the entire operation. For example, in one embodiment, the target recognition accuracy, firing rate efficiency, and remaining energy ratio may be set as the three main monitoring items, and based on this, a comprehensive scoring assessment is carried out once an hour.

[0096] Subsequently, the function of parsing the complex situation processing process is realized by means of the rule chain Rule_chain, and then a series of instructions are generated based on the overall score and combined into Command_Set to guide the actual action route of the tracked robot. Different weight parameters are assigned according to the analysis results of the above-mentioned conditions, and an optimal practice plan to be followed in the current state is calculated. Once the score score is lower than (Th + ρ) * avg_score - where Th represents a fixed threshold coefficient usually set around 0.8, ρ is used as a floating correction term with a value range of approximately ±0.2, and avg_score represents the historical average score. This setting is intended to ensure that a certain combat effectiveness and safety performance standard can still be maintained even in extremely harsh environments, while reserving a certain amount of flexible space for human judgment. For example, if the current score is 6.7, the average score is 8.0, and (0.8 + 0.2) * 8.0 = 8.0, a warning signal needs to be sent at this time to remind the operator to review the rationality of the current strategy, consider whether there is a better solution that can be adopted, and then decide whether to issue an instruction to continue moving forward.

[0097] The shooting control method is based on a tracked robot, and a fuzzy logic system is introduced in the dynamic target locking process to enhance the shooting accuracy, especially in the case of high uncertainty and difficulty in quantification. A new dimension is added to the traditional shooting control system.

[0098] First, the introduction of the fuzzy logic system is executed, which means that the system can simulate the human thinking mode for decision-making, especially showing advantages when facing complex non-linear conditions. In this process, the conversion from natural language to computer-recognizable signals for many concepts such as near, far, bright, dark, left deviation, and right deviation is completed by defining various input / output membership functions. These terms correspond to a set of parameterized functions, such as distance d (unit: meter), brightness I (unit: cd), and deviation angle θ (unit: degree), with ranges of 0 - ∞ meters, 0 - infinite brightness units, and -90 to 90 degrees respectively; setting reasonable boundary values is crucial for actual operations, and the optimal values vary according to different application scenarios.

[0099] Specifically, after designing the membership function, it is necessary to construct a fuzzy inference rule base based on expert experience. The rules are written in the IF-THEN format. For example, if the target distance is far and the light is dim, then increase the laser aiming error window and improve the sensitivity of the image sensor to maintain the tracking success rate, and finally form the rule table Table_F. Such settings ensure reasonable selection even in the face of changing and complex battlefield situations, greatly improving the shooting hit rate and safety.

[0100] For example, assume that the task environment of a tracked robot is in a forest where the dense trees interfere with the line of sight and cause the visual system to frequently lose tracking. At this time, the above-mentioned fuzzy logic auxiliary function will come into play. When it is determined that there is a suspected enemy movement ahead but there is a large uncertainty in positioning due to its being in a shrub occlusion area, an additional wide-angle detection program will be started to capture suspicious signals, and at the same time, the precise firing threshold will be lowered until it is confirmed again without error before allowing the firing authorization.

[0101] An intelligent shooting control system for a tracked robot with a dynamic target locking function according to the present invention includes:

[0102] The main process of the intelligent shooting control system is divided into the following steps:

[0103] 1. Initialization settings: When the robot starts, the system will comprehensively initialize the visual sensor and the shooting module to ensure that each subsystem is in the best working state. Configure the working mode of the visual sensor according to the initial environmental parameters (such as lighting conditions, temperature, etc.).

[0104] 2. Real-time environmental monitoring and perception: Obtain the target position change and the surrounding environmental conditions (including wind speed, wind direction, light change, etc.) in real time through high-frequency sensing units. These data will provide an accurate information source for subsequent processing.

[0105] 3. Multi-sensor fusion analysis and decision-making: Combine image processing technology and motion estimation algorithms to identify and track the target from the image information obtained by the visual sensor; at the same time, integrate the data of the inertial measurement unit to judge the terrain situation. For multiple target tracking in complex environments, a deep learning network will be used to classify each potential target and predict its future movement based on the historical behavior model.

[0106] 4. Generation of intelligent control strategies:

[0107] Automatically switch sensor parameters - adjust the image contrast, brightness, and focus range in a timely manner according to the detected changes to optimize the imaging quality;

[0108] Calculate and update the best projectile angle - formulate corresponding instructions to the actuator after comprehensively analyzing the inputs from various parts;

[0109] 5. Execution instruction feedback correction loop formation: Immediately collect flight trajectory information after launch for subsequent improvement; if there is a deviation, quickly feedback it to the controller for parameter adaptive correction until the ideal accuracy is achieved.

[0110] How to solve specific technical problems

[0111] Regarding Problem 1: How to adjust the shooting angle according to the real-time target position change to solve the problem of insufficient shooting accuracy;

[0112] In the present invention, a set of efficient real-time target position tracking mechanisms is designed, which uses advanced image processing and motion estimation algorithms to accurately capture the positions of fast-moving or variably accelerating target objects. The system can continuously monitor the position of the target and can react within the shortest time to reset the shooting angle. This enables precise strikes even in a high-speed dynamic environment. In addition, through the immediate evaluation of the shooting effect of each shot (i.e., shooting result feedback), the future shooting instructions are continuously calibrated, thereby minimizing the risk of missing the target due to the sudden change of the target's direction or speed.

[0113] -Regarding Problem 2: How to adjust the visual sensor parameters according to the environmental light and reflection characteristics to solve the problem of low target recognition accuracy;

[0114] To improve the ability of the visual sensor to stably and reliably recognize targets under different lighting conditions, the present invention adopts the method of dynamically adjusting functions such as the gain, exposure time, and color calibration of the imaging device. It also introduces an adaptive enhancement processing means to make the imaging closer to the real color and reduce the possibility of shadow interference. More importantly, the machine learning method is adopted to enable the machine to self-train according to empirical data, so as to better distinguish various complex texture structures or hidden objects under the camouflage background, and can automatically select the best imaging mode suitable for the current environmental conditions to ensure good detection performance at any time.

[0115] Regarding Problem 3: How to adjust the shooting force according to the dynamic data of wind speed and wind direction to solve the problem of reduced hit rate caused by external environmental influences;

[0116] The present invention integrates miniaturized meteorological detection station equipment, which can obtain more accurate weather forecast data, especially the local wind field distribution characteristics, before shooting, and use this as a reference to predetermine key parameters such as the launch timing and initial velocity. On this basis, a complete simulation platform is established using the principles of fluid mechanics to carry out simulation tests. A series of experimental verifications have shown that this solution can indeed effectively alleviate the offset phenomenon caused by natural factors. In addition, considering that the wind speed and wind direction may fluctuate greatly over time, an online fine-tuning step is added before each trigger to further reduce random errors, ultimately ensuring that the accuracy of shooting is always at a relatively ideal level.

[0117] Regarding question 4: How to adjust the priority according to the multi-target motion trajectory prediction algorithm to solve the problem of selective shooting lag in complex scenarios;

[0118] The present invention creatively adopts the artificial intelligence technology path based on the Markov Decision Process (MDP) framework to plan attack strategies for multiple mobile entities. First, a probability prediction model that conforms to physical characteristics and motion laws is constructed with the help of a deep neural network. Then, the value attributes of different objects are sorted to determine which unit should be responded to first as the target point. Then, the central control system sends instructions to the relevant weapon operation platform to complete the specified action, realizing a major transformation from passive response to active attack, improving the efficiency of emergency response while taking into account the goal of maximizing the overall mission effectiveness.

[0119] Regarding question 5: How to adjust the aiming system deviation compensation parameters according to the shooting feedback data to solve the problem of increasing cumulative error in continuous shooting;

[0120] In order to ensure a high hit probability after multiple shots, the invention particularly emphasizes a closed-loop control concept. That is, after each bullet is fired, specific information about the landing coordinates is quickly collected, and then a comparative difference analysis is carried out against the preset target point. Once any sign of deviation is found, the correction plan is immediately initiated - by inverting the mathematical equations and reverse reasoning to find the internal or external interference factors that cause the misalignment, and simultaneously pre-processing other pending events to ensure that subsequent shots do not repeat similar errors. The entire process involves the application of many high-end algorithms such as Kalman Filter, Least Squares Fitting and other scientific calculation methods, which all help to significantly improve the consistency and smoothness of continuous shooting.

[0121] Although embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, and the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A tracked robot intelligent shooting control system with dynamic target locking function, characterized in that: Including: Dynamically adjust the working mode of the vision sensor based on the real-time target position change and environmental parameters, calculate the optimal shooting angle according to the adjusted vision sensor data and perform real-time regulation, correct the shooting force based on the dynamic data of wind speed and wind direction to adapt to the external environment change, determine the priority according to the multi-target motion trajectory prediction algorithm and optimize the shooting order; Obtain the real-time target position parameters (X, Y, Z), where (X, Y) is the target plane coordinate and Z is the height. Adjust the shooting angle θ in real time based on the formula \(\theta_{new}=\theta_{old}+\Delta X\cdot K_1+\Delta Y\cdot K_2+\Delta Z\cdot K_3\), where \(K_1\), \(K_2\) and \(K_3\) are the horizontal, lateral and vertical adjustment coefficients respectively. Adjust the light intensity threshold P of the vision sensor according to the environmental light sensor data. If the target recognition rate R is lower than the threshold T, i.e., \(R < T\), adjust the light compensation factor F and recalculate the light intensity threshold.

2. According to claim 1, a tracked robot intelligent shooting control system with dynamic target locking function is characterized in that: Detect the light reflection intensity I and the target surface reflectivity γ in real time, and update the contrast C of the vision sensor. Set the contrast gain G to enhance the target distinguishability in the high-reflection area. Reset the visual contrast C based on the formula \(C_{new}=C\times(G + I)\). Dynamically correct the ballistic curve equation according to the wind speed v and wind direction a.

3. According to claim 1, a tracked robot intelligent shooting control system with dynamic target locking function is characterized in that: Capture the multi-target trajectory prediction information and the corresponding risk assessment matrix \(\mathbf{M}\) for making decisions on the optimal priority sequence S. Sort the sequence S through the iterative optimization algorithm until the condition \(\sum_{i}w_i\cdot(\text{Distance}_{gi}(T))\leq D_{max}\) is satisfied, where \(w_i\) is the weighted weight of target i, \(\text{Distance}_{gi}(T)\) represents the predicted minimum approach distance within the given time T, and \(D_{max}\) is the set safety spacing limit. After each shooting operation, record the feedback error ε and accumulate it into the total deviation Δ; If Δ exceeds the pre-determined threshold Ω, trigger the adaptive learning model to fine-tune the calculation method of the next shelling parameter set Φ, i.e., Φ = Φ + β·ε, where β is the system sensitivity adjustment factor.

4. According to claim 1, a tracked robot intelligent shooting control system with dynamic target locking function is characterized in that: Record the hit point coordinates H and the offset vector d = (Δx, Δy) of the ideal aiming point A for each shooting event; Define the error function \(E(k)=\sum_{i = k}^{n}{((\Delta x_i-\Delta x_k)+(\Delta y_i-\Delta y_k))^2}\) to calculate the relative deviation between the k-th bullet and the previous N consecutive shootings, and use this value as the compensation basis; Dynamically adjust the basic yaw correction value Λ for the next round of firing according to the result of the previous step; If the expected hit probability of the current round Pr(Δ|F_prev)<η, then reduce the energy output U of the next bullet to reduce unnecessary consumption while ensuring accurate hits, where F_prev represents the previous feedback, η is the acceptable low-precision tolerance level, and U controls the launch kinetic energy level.

5. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: When all tasks are completed in the current round or a specific situation occurs, the system switches to the preparation mode and clears all temporary calculation data; The Kalman filter is used to smooth the noise interference part in the difference image between the new and old visual frames to improve the target tracking accuracy and enhance the robustness; Use 3D point cloud mapping to model the motion trend in complex backgrounds, so that subsequent shooting is more in line with the expected trend of dynamic changes; The time-weighted exponential average Twma model is introduced, and the new displacement correction term Δ is calculated according to the formula \[Delta x=\mu\Delta x+(1\mu)\Delta x], so as to fine-tune the specific position selection plan for each shooting, where μ∈(0,1) is used as the weighted weight coefficient to ensure that past performance has sufficient influence on future decisions and is not overly sensitive.

6. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: The reinforcement learning framework guides machine learning to automatically calibrate the process of uncertainty, such as unpredictable variables such as changes in lighting conditions or probability distributions of unexpected disturbances; For each observed action response, a positive or negative reward score is given, and a QValue state action value table is established for subsequent training to improve performance; In the face of unknown or extreme weather conditions, a more accurate and reliable environmental simulation forecast is constructed through the random forest regression tree ensemble model, especially for wind impact analysis; When an abnormally large shooting dispersion is found accompanied by a significant deviation from the originally planned route, that is, when σ>(σ_0+r*δσ) holds (where σ_0 is the baseline deviation fluctuation, the r factor determines the incremental ratio, and δσ describes the increase in the standard deviation), additional protective measures should be initiated to limit the occurrence of dangerous situations.

7. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: Equipped with a multi-channel laser rangefinder to obtain the distance information D_set of multiple different reflective surfaces in the front space in real time, and generate high-resolution map images based on this, providing a reliable geographic reference system for subsequent shooting preparations; The particle swarm optimization algorithm is used to solve the relative benefit score e of each potential strike point in a complex environment and select the local optimal solution set; Adjust PSO parameters according to the actual test environment characteristics, including maximum flight speed v_{text{max}}, trust factor gbest_factor, and interaction frequency n_iter; A standardized update rule based on the formula \[n_j=(v^{text{T}}_{j} / ∥v_j∥)_\text{clip}] is set to regulate the position of individual particles, so that the overall evolution direction can be advanced more scientifically, reasonably and orderly to the vicinity of the global optimal solution. Here v is the velocity vector in each dimension and ∥.∥ represents the Euclidean norm, and ()clip is limited to the interval (1,1).

8. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: Combined with the deep belief network DBN pre-training initialization strategy, the convergence efficiency of the deep neural network structure is improved to ensure that the model can quickly enter a good state for use; Add long short-term memory module LSTM to capture the changing characteristics of sequence dependency and non-stationary behavior patterns; Guide the LSTM output unit activation threshold m to gradually approach the most suitable state, set the threshold τ and the increasing rate φ based on experience, and increase the output activation degree when t>\((1+φ)^{text{kτ}}); We tried to conduct experimental verification on various possible architecture combinations, such as a double-layer BiLSTM plus a convolutional pooling kernel and other mixed types, and continuously searched for the ideal configuration that was most suitable for target positioning and shooting control tasks, and recorded it for direct reference and use in similar projects in the future.

9. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: Construct a multi-dimensional situational awareness library CAS, which collects three parts: the information W_fore returned by the weather forecast API interface, the data Mil_info uploaded by the military situation monitoring station, and the user-defined preference Profile; Regularly synchronize and integrate the latest external data to update the internal database to ensure that all important factors at the macro and micro levels are fully considered in the operational planning decision-making process; Extract key indicators to build the evaluation system KPI, monitor the changes of various factors online and make periodic evaluation feedback on the overall effect; With the help of rule chain Rule_chain, the complex linkage scenario rule set is parsed, and the next operation instruction sequence Command_Set is inferred based on the comprehensive evaluation score score. Whenever score<(Th+ρ)*avg_score occurs.

10. The tracked robot intelligent shooting control system with dynamic target locking function according to claim 1, characterized in that: Introducing a fuzzy logic system to assist in adjusting shooting accuracy, especially in situations where uncertainty is high and difficult to quantify accurately, effectively improving the quality of the final result; Identify a series of input / output membership functions to describe the closeness of the relationship between dimensions and the flexibility of boundary transformation, such as near, far, bright, dark, left, right, etc., which are translated from natural language vocabulary; A series of inference rules IFTHEN formulated in conjunction with expert experience are used to implement the fuzzy transformation rule table Table_F, which intuitively and clearly shows what specific actions should be taken in various situations, greatly improving the applicability of the system.

Images