An artificial intelligence-based high-precision driving control system and method for a servo helmet

Abstract

The application discloses a high-precision driving and control system and method of a follow-up helmet based on artificial intelligence, relates to the technical field of driving control, and collects original data of a target by using a target tracking sensor, and collects motion state data of the helmet itself by using a body state sensor arranged in the helmet; a transformation matrix from a sensor coordinate system to a stable world coordinate system is generated, the original data of the target is processed by using the transformation matrix to obtain real data of the target; a prediction model is obtained by training the neural network by inputting historical real data into the neural network, and a predicted target instruction is calculated by using the real data; a final control instruction is obtained by fusing a feedforward control amount and a feedback control amount; the final control instruction is compensated by a compensator to obtain an optimal control instruction; and a prediction error is fed back to the prediction model for optimized training.

Description

Technical Field

[0001] This invention relates to the field of drive control technology, specifically to a high-precision drive control system and method for a servo helmet based on artificial intelligence. Background Technology

[0002] Early helmet homing systems were primarily based on simple sensors and mechanical control principles. For example, some pilot helmet homing systems used gyroscopes to measure the pilot's head tilt angle, transmitting this data to the control system, which then drove servos to control corresponding mechanical structures to perform following movements. However, the accuracy and functionality of such systems were relatively limited. Smart helmets integrate multiple sensors, such as accelerometers, gyroscopes, and magnetometers (inertial measurement units), enabling precise measurement of head posture and motion. Furthermore, the application of visual sensors such as cameras provides data support for target recognition and tracking. With the development of artificial intelligence (AI) technology, devices can understand and respond to changes in their surrounding environment. In homing helmets, AI can be used for speech recognition, target recognition and tracking, posture calculation, etc., optimizing performance through learning algorithms to improve system accuracy and real-time performance.

[0003] However, most traditional servo systems employ open-loop or simple closed-loop control based on a "sensing-decision-execution" model. Their core problem is latency. From the moment the sensor acquires target information, to the controller's calculations, and finally to the motor driving the helmet's rotation, any delay in any of these steps will cause tracking lag, especially for high-speed, irregularly moving targets. Summary of the Invention

[0004] The purpose of this invention is to provide a high-precision drive and control system and method for a servo helmet based on artificial intelligence, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A high-precision drive and control method for a servo helmet based on artificial intelligence, the method comprising the following steps:

[0007] S100. When the helmet captures a target, it uses a target tracking sensor to collect the target's raw data and uses a body state sensor installed in the helmet to collect the helmet's own motion state data.

[0008] Furthermore, the specific steps for collecting the helmet's own motion state data using the body state sensors installed in the helmet are as follows:

[0009] S101. Collect the target's raw data Z_t[k]=(r, θ, d, v_m) using a target tracking sensor, where r represents the distance between the helmet and the target, θ represents the target's azimuth angle, d represents the target's pitch angle, v represents the target's velocity, including velocities in three directions; k represents the k-th time moment; collect the helmet's own motion state data S_h[k]=(α, β, γ, ω_α, ω_β, ω_γ, (x, y, z), a_x, a_y, a_z), α, β and γ represent the current roll angle, pitch angle, and yaw angle of the helmet, respectively; ω_α, ω_β, and ω_γ represent the angular velocities of the current roll angle, pitch angle, and yaw angle of the helmet, respectively; a_x, a_y, and a_z represent the linear velocities of the helmet itself in three-dimensional coordinates, respectively; and (x, y, z) represent the three-dimensional position coordinates. The original target data is data in the sensor coordinate system, and the helmet's own motion state data is data in the carrier coordinate system. The relative positions between the sensor coordinate system and the carrier coordinate system remain fixed.

[0010] Simultaneously collecting the target's raw data and the helmet's own motion state data avoids tracking deviations caused by insufficient data dimensions and provides complete input for subsequent calculations; clarifying that the target data is based on the sensor coordinate system and the helmet state data is based on the carrier coordinate system, and that the relative positions of the two are fixed, provides a stable premise for subsequent coordinate transformation and reduces errors caused by coordinate system confusion.

[0011] S200. The Kalman filter algorithm is used to optimize the motion state data of the helmet itself to obtain the optimal state estimate. The optimal state estimate is used to generate a transformation matrix from the sensor coordinate system to the stable world coordinate system. The transformation matrix is ​​used to process the original target data to obtain the target's true data.

[0012] Furthermore, the specific steps for processing the original target data using the transformation matrix to obtain the true target data are as follows:

[0013] S201. Convert the roll, pitch, and yaw angles from the helmet's motion data into quaternion representations q = [q0, q1, q2, q3], where q0 represents the scalar part and q1, q2, and q3 represent the vector part. Extract the position coordinates to construct a position vector p = [x, y, z]. Extract the linear velocity coordinates to construct a velocity vector v = [a_x, a_y, a_z]. Extract the angular velocities of the helmet's current roll, pitch, and yaw angles to construct an angular velocity vector ω = [ω_α, ω_β, ω_γ]. Construct the helmet state vector X_ka = [q, p, ω, v]. T The helmet state vector is input into the Kalman filter algorithm for filtering optimization to obtain the optimal state estimate.

[0014] S202. Regarding the sensor coordinate system S, the carrier coordinate system B, and the world coordinate system W, the sensor coordinate system has its origin at the sensor itself, and the carrier coordinate system has its origin at the helmet itself. The three-dimensional coordinates of the origin of the sensor coordinate system in the carrier coordinate system are p_S. B =[x_S B ,y_S B ,z_S B ] T The angle data of the sensor origin in the carrier coordinate system is collected and converted into quaternions, and then the quaternions are converted into a rotation matrix R_S. B The transformation matrix from the sensor coordinate system S to the carrier coordinate system B is constructed using the following formula:

[0015] ;

[0016] In the formula, T_S B This represents the transformation matrix from the sensor coordinate system S to the carrier coordinate system B. The transformation matrix is ​​a 4×4 matrix, and the three-dimensional coordinates p_S of the origin of the sensor coordinate system in the carrier coordinate system are... B It is a 3×1 matrix, 03 T Let represent a 1×3 zero vector, where 1 represents a scalar;

[0017] Convert the quaternions in the optimal state estimation into a rotation matrix R_B W The 3×1 carrier position coordinates extracted from the optimal state estimate are p_B W The transformation matrix from the sensor coordinate system B to the carrier coordinate system W is constructed using the following formula:

[0018] ;

[0019] In the formula, T_B w The transformation matrix from sensor coordinate system B to carrier coordinate system W is represented by the transformation matrix T_S. Using these two transformation matrices, the transformation matrix from sensor coordinate system S to carrier coordinate system W is obtained. W =T_B w ×T_S B The transformation matrices are all 4×4 matrices.

[0020] S203. Calculate the homogeneous coordinates of the target in the sensor coordinate system using the target's original data. The formula is:

[0021] ;

[0022] In the formula, P_t S This represents the homogeneous coordinates of the target in the sensor coordinate system. The true coordinates of the target in the world coordinate system are calculated using the transformation matrix from the sensor coordinate system S to the carrier coordinate system W. The formula is:

[0023] ;

[0024] In the formula, P_t W The true coordinates of the target in the world coordinate system are given. The true velocity of the target in the world coordinate system is calculated using the following formula:

[0025] ;

[0026] In the formula, V_t W The target's true velocity in the world coordinate system is represented by a 3×1 velocity matrix with three dimensions; v_m represents the target's position coordinates in the original data, which is also a 3×1 matrix; ω_B W v_B represents the helmet angular velocity vector extracted from the optimal state estimation, and is a 3×1 matrix; W This represents the helmet velocity vector extracted from the optimal state estimation, which is a 3×1 matrix;

[0027] By combining the actual coordinates and actual velocity, we obtain the actual data of the target in the world coordinate system.

[0028] Kalman filtering is used to optimize the helmet motion data, filtering out sensor measurement noise and obtaining the optimal state estimate, thus improving data reliability. A three-level transformation matrix is ​​constructed from sensor to carrier to world coordinate system to convert the original target data in the sensor coordinate system into real data in the world coordinate system, eliminating the influence of the helmet's own motion on target positioning and achieving "decoupling" and accurate calculation of target position and velocity.

[0029] S300: Based on time series calculation, obtain the real data of the target's change over time in history, input the real historical data into the neural network to train the prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target command.

[0030] Furthermore, the specific steps for calculating the predicted target instruction using real data are as follows:

[0031] S301. Based on time series calculation, the real data of the target changing over time in history is obtained. The real historical data is input into the neural network to train a prediction model. The prediction time is set to Δt. The prediction model is used to predict the real data after Δt in the future as X_p[k+Δt]. The prediction time = helmet system information processing time + execution delay time.

[0032] S302. Perform an inverse transformation on the transformation matrix from sensor coordinate system B to carrier coordinate system W to obtain the transformation matrix T_W from sensor coordinate system W to carrier coordinate system B. BExtract the predicted true coordinates from the predicted true data, add scalar 1s to form a 4×1 matrix, and then use the transformation matrix T_W B Transform the predicted true coordinates to coordinates p_p in the carrier coordinate system B. B = (x_p, y_p, z_p), using coordinates p_p B Calculate the azimuth and pitch angles of the target being tracked by the helmet-mounted system, where the formula for calculating the azimuth angle is:

[0033] ;

[0034] In the formula, θ_c represents the azimuth angle of the target being tracked by the helmet-mounted system, atan2 represents the two-parameter arctangent function, y_p represents the y-axis coordinate of the target in the vehicle coordinate system, and x_p represents the x-axis coordinate of the target in the vehicle coordinate system; the pitch angle calculation formula is: In the formula, d_c represents the pitch angle of the target being tracked by the helmet drive, and z_p represents the z-axis coordinate of the target in the carrier coordinate system.

[0035] The azimuth and pitch angles of the target being tracked by the helmet-driven system are combined to form the target prediction command (θ_c, φ_c)[k+Δt].

[0036] A neural network prediction model is trained based on historical real data from time series to accurately predict the target state in the future Δt, fundamentally solving the "lag tracking" problem caused by delays in traditional servo systems. The predicted target data in the world coordinate system is inversely transformed to the carrier coordinate system to calculate the azimuth and pitch angles required for helmet actuation, ensuring that the prediction results are directly adapted to the helmet control logic and reducing command conversion losses.

[0037] S400: The helmet calculates the feedforward control quantity based on the predicted target command, and calculates the feedback control quantity based on the original target data and the helmet's own motion state data. The feedforward control quantity and the feedback control quantity are then fused to obtain the final control command.

[0038] Furthermore, the specific steps for fusing the feedforward control quantity and the feedback control quantity to obtain the final control command are as follows:

[0039] S401. Differentiate the predicted target command to obtain the desired acceleration and velocity during helmet-driven operation. Use the desired acceleration and velocity to generate the feedforward control quantity, as shown in the formula: In the formula, U_ff represents the feedforward control quantity, K_a represents the acceleration proportional gain, K_v represents the velocity proportional gain, a_c represents the desired acceleration, and v_c represents the desired velocity.

[0040] S402. Extract the current target angle (θ_m, d_m)[k] and the actual helmet angle (θ_h, d_h)[k] in the carrier coordinate system, and calculate the error between the target and the helmet using the following formula: In the formula, e[k] represents the current angle error between the target and the helmet; the feedback control quantity is calculated using the angle error, and the formula is:

[0041] ;

[0042] In the formula, U_fb represents the feedback control quantity, and K_p, Ki, and K_d represent the angle error, error integral, and error differential coefficients, respectively.

[0043] The feedforward control quantity and the feedback control quantity are fused to obtain the final control command, and the formula is: U=U_ff+U_fb, where U represents the final control command.

[0044] Based on the expected velocity and acceleration predicted by the target command, a feedforward control quantity is generated to drive the helmet to move towards the target's future position in advance, significantly improving the system's response speed to dynamic targets and avoiding the lag of "passive tracking". By calculating the angular error between the current target and the helmet, a feedback control quantity is generated in combination with a PID algorithm to correct the prediction deviation in the feedforward control in real time, achieving a dual guarantee of "prediction + correction". Feedforward control solves the response speed problem, and feedback control solves the accuracy deviation problem. The combination of the two makes the control command both "fast" and "accurate", balancing dynamic response and steady-state accuracy.

[0045] S500: A compensator is installed in the helmet. The final control command is input into the compensator, and the compensator compensates for the final control command to obtain the optimal control command.

[0046] Furthermore, the specific steps for the compensator to compensate for the final control command to obtain the optimal control command are as follows:

[0047] S501. The compensator is a nonlinear compensator. The nonlinear compensator is used to compensate for friction and tooth groove in the final control command and output the optimal control command.

[0048] By using a nonlinear compensator to specifically address mechanical nonlinear errors such as friction and cogging effect in the helmet drive mechanism, the final control command is corrected to the optimal command with "no loss", ensuring that the control signal is accurately converted into mechanical motion.

[0049] S600: The helmet rotates to track the target according to the optimal control command. After the control ends, the helmet collects the target prediction command and the actual position of the target, calculates the prediction error, and feeds the prediction error back to the prediction model for optimization training.

[0050] Furthermore, the specific steps for feeding the prediction error back into the prediction model for optimization training are as follows:

[0051] S601. Use the best control command to control the helmet drive to track the target. After the control is completed, the helmet collects the target prediction command and the actual position of the target. Calculate the difference between the predicted target command and the actual position of the target as the prediction error. Input the prediction error into the prediction model for optimization training.

[0052] The deviation between the predicted target command and the actual position is fed back to the neural network model, continuously optimizing the model parameters so that the prediction accuracy improves with the number of uses, adapts to the motion characteristics of different targets, and achieves "self-learning" evolution.

[0053] A high-precision drive and control system for a servo helmet based on artificial intelligence, comprising a data acquisition module, a data transformation module, a prediction module, an instruction generation module, an instruction optimization module, and an update module;

[0054] The data acquisition module is used to collect the target's raw data using the target tracking sensor and to collect the helmet's own motion state data using the body state sensor installed in the helmet when the helmet captures the target.

[0055] The data transformation module is used to optimize the helmet's own motion state data using the Kalman filter algorithm to obtain the optimal state estimate, generate a transformation matrix from the sensor coordinate system to the stable world coordinate system using the optimal state estimate, and process the original target data using the transformation matrix to obtain the target's true data.

[0056] The prediction module is used to calculate the real data of the target's changes over time in history based on time series calculations, input the real historical data into a neural network to train a prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target instructions.

[0057] The instruction generation module is used for the helmet to calculate the feedforward control quantity based on the predicted target instruction, calculate the feedback control quantity based on the original target data and the helmet's own motion state data, and fuse the feedforward control quantity and the feedback control quantity to obtain the final control instruction.

[0058] The instruction optimization module is used to set up a compensator in the helmet, input the final control instruction into the compensator, and the compensator compensates for the final control instruction to obtain the optimal control instruction.

[0059] The update module is used to collect the target prediction command and the actual position of the target after the control ends, calculate the prediction error, and feed the prediction error back to the prediction model for optimization training.

[0060] The data transformation module includes a transformation matrix calculation unit and a data transformation unit;

[0061] The transformation matrix calculation unit is used to calculate the transformation matrix from sensor coordinate system S to carrier coordinate system B and the transformation matrix from sensor coordinate system B to carrier coordinate system W respectively, and combine them to obtain the transformation matrix from sensor coordinate system S to carrier coordinate system W.

[0062] The data transformation unit is used to transform the original target data into real coordinates in the world coordinate system using a transformation matrix.

[0063] The instruction generation module includes a feedforward control unit, a feedback control unit, and a fusion unit;

[0064] The feedforward control unit is used to perform differential processing on the predicted target command to obtain the expected acceleration and velocity when the helmet is driven, and to generate the feedforward control quantity using the expected acceleration and velocity.

[0065] The feedback control unit is used to calculate the error between the target and the helmet, and to calculate the feedback control quantity using the angle error.

[0066] The fusion unit is used to fuse the feedforward control quantity and the feedback control quantity to obtain the final control command.

[0067] Compared with the prior art, the beneficial effects of the present invention are:

[0068] 1. This invention achieves end-to-end error control from the data layer to the positioning layer to the control layer to the execution layer, solving the three core accuracy problems of traditional systems: "noise interference, delay lag, and mechanical error". The tracking error can be reduced to the limit of the system hardware.

[0069] 2. This invention constructs a complete closed loop of "acquisition → transformation → prediction → control → feedback optimization". The model can continuously evolve with the change of the target's motion mode. It is not only suitable for tracking targets at a constant speed, but also can accurately cope with complex dynamic scenarios such as high-speed changes of direction and sudden acceleration. Its versatility and adaptability far exceed those of traditional systems with fixed parameters. Attached Figure Description

[0070] Figure 1 This is a module distribution diagram of a high-precision drive and control system for a servo helmet based on artificial intelligence, according to the present invention.

[0071] Figure 2 This is a schematic diagram illustrating the steps of a high-precision drive and control method for a servo helmet based on artificial intelligence, as described in this invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example: Figures 1-2 As shown, the present invention provides a technical solution.

[0074] A high-precision drive and control method for a servo helmet based on artificial intelligence, the method comprising the following steps:

[0075] S100. When the helmet captures a target, it uses a target tracking sensor to collect the target's raw data and uses a body state sensor installed in the helmet to collect the helmet's own motion state data.

[0076] The specific steps for collecting the helmet's own motion data using the body state sensors installed in the helmet are as follows:

[0077] S101. Collect the target's raw data Z_t[k]=(r, θ, d, v_m) using a target tracking sensor, where r represents the distance between the helmet and the target, θ represents the target's azimuth angle, d represents the target's pitch angle, v represents the target's velocity, including velocities in three directions; k represents the k-th time moment; collect the helmet's own motion state data S_h[k]=(α, β, γ, ω_α, ω_β, ω_γ, (x, y, z), a_x, a_y, a_z), α, β and γ represent the current roll angle, pitch angle, and yaw angle of the helmet, respectively; ω_α, ω_β, and ω_γ represent the angular velocities of the current roll angle, pitch angle, and yaw angle of the helmet, respectively; a_x, a_y, and a_z represent the linear velocities of the helmet itself in three-dimensional coordinates, respectively; and (x, y, z) represent the three-dimensional position coordinates. The original target data is data in the sensor coordinate system, and the helmet's own motion state data is data in the carrier coordinate system. The relative positions between the sensor coordinate system and the carrier coordinate system remain fixed.

[0078] Simultaneously collecting the target's raw data and the helmet's own motion state data avoids tracking deviations caused by insufficient data dimensions and provides complete input for subsequent calculations; clarifying that the target data is based on the sensor coordinate system and the helmet state data is based on the carrier coordinate system, and that the relative positions of the two are fixed, provides a stable premise for subsequent coordinate transformation and reduces errors caused by coordinate system confusion.

[0079] S200. The Kalman filter algorithm is used to optimize the motion state data of the helmet itself to obtain the optimal state estimate. The optimal state estimate is used to generate a transformation matrix from the sensor coordinate system to the stable world coordinate system. The transformation matrix is ​​used to process the original target data to obtain the target's true data.

[0080] The specific steps for processing the original target data using a transformation matrix to obtain the true target data are as follows:

[0081] S201. Convert the roll, pitch, and yaw angles from the helmet's motion data into quaternion representations q = [q0, q1, q2, q3], where q0 represents the scalar part and q1, q2, and q3 represent the vector part. Extract the position coordinates to construct a position vector p = [x, y, z]. Extract the linear velocity coordinates to construct a velocity vector v = [a_x, a_y, a_z]. Extract the angular velocities of the helmet's current roll, pitch, and yaw angles to construct an angular velocity vector ω = [ω_α, ω_β, ω_γ]. Construct the helmet state vector X_ka = [q, p, ω, v]. T The helmet state vector is input into the Kalman filter algorithm for filtering optimization to obtain the optimal state estimate.

[0082] S202. Regarding the sensor coordinate system S, the carrier coordinate system B, and the world coordinate system W, the sensor coordinate system has its origin at the sensor itself, and the carrier coordinate system has its origin at the helmet itself. The three-dimensional coordinates of the origin of the sensor coordinate system in the carrier coordinate system are p_S. B =[x_S B ,y_S B ,z_S B ] T The angle data of the sensor origin in the carrier coordinate system is collected and converted into quaternions, and then the quaternions are converted into a rotation matrix R_S. B The transformation matrix from the sensor coordinate system S to the carrier coordinate system B is constructed using the following formula:

[0083] ;

[0084] In the formula, T_S B This represents the transformation matrix from the sensor coordinate system S to the carrier coordinate system B. The transformation matrix is ​​a 4×4 matrix, and the three-dimensional coordinates p_S of the origin of the sensor coordinate system in the carrier coordinate system are... B It is a 3×1 matrix, 03 T Let represent a 1×3 zero vector, where 1 represents a scalar;

[0085] Convert the quaternions in the optimal state estimation into a rotation matrix R_B W The 3×1 carrier position coordinates extracted from the optimal state estimate are p_B WThe transformation matrix from the sensor coordinate system B to the carrier coordinate system W is constructed using the following formula:

[0086] ;

[0087] In the formula, T_B w The transformation matrix from sensor coordinate system B to carrier coordinate system W is represented by the transformation matrix T_S. Using these two transformation matrices, the transformation matrix from sensor coordinate system S to carrier coordinate system W is obtained. W =T_B w ×T_S B The transformation matrices are all 4×4 matrices.

[0088] S203. Calculate the homogeneous coordinates of the target in the sensor coordinate system using the target's original data. The formula is:

[0089] ;

[0090] In the formula, P_t S This represents the homogeneous coordinates of the target in the sensor coordinate system. The true coordinates of the target in the world coordinate system are calculated using the transformation matrix from the sensor coordinate system S to the carrier coordinate system W. The formula is:

[0091] ;

[0092] In the formula, P_t W The true coordinates of the target in the world coordinate system are given. The true velocity of the target in the world coordinate system is calculated using the following formula:

[0093] ;

[0094] In the formula, V_t W The target's true velocity in the world coordinate system is represented by a 3×1 velocity matrix with three dimensions; v_m represents the target's position coordinates in the original data, which is also a 3×1 matrix; ω_B W v_B represents the helmet angular velocity vector extracted from the optimal state estimation, and is a 3×1 matrix; W This represents the helmet velocity vector extracted from the optimal state estimation, which is a 3×1 matrix;

[0095] By combining the actual coordinates and actual velocity, we obtain the actual data of the target in the world coordinate system.

[0096] Kalman filtering is used to optimize the helmet motion data, filtering out sensor measurement noise and obtaining the optimal state estimate, thus improving data reliability. A three-level transformation matrix is ​​constructed from sensor to carrier to world coordinate system to convert the original target data in the sensor coordinate system into real data in the world coordinate system, eliminating the influence of the helmet's own motion on target positioning and achieving "decoupling" and accurate calculation of target position and velocity.

[0097] S300: Based on time series calculation, obtain the real data of the target's change over time in history, input the real historical data into the neural network to train the prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target command.

[0098] The specific steps for calculating the predicted target instruction using real data are as follows:

[0099] S301. Based on time series calculation, the real data of the target changing over time in history is obtained. The real historical data is input into the neural network to train a prediction model. The prediction time is set to Δt. The prediction model is used to predict the real data after Δt in the future as X_p[k+Δt]. The prediction time = helmet system information processing time + execution delay time.

[0100] S302. Perform an inverse transformation on the transformation matrix from sensor coordinate system B to carrier coordinate system W to obtain the transformation matrix T_W from sensor coordinate system W to carrier coordinate system B. B Extract the predicted true coordinates from the predicted true data, add scalar 1s to form a 4×1 matrix, and then use the transformation matrix T_W B Transform the predicted true coordinates to coordinates p_p in the carrier coordinate system B. B = (x_p, y_p, z_p), using coordinates p_p B Calculate the azimuth and pitch angles of the target being tracked by the helmet-mounted system, where the formula for calculating the azimuth angle is:

[0101] ;

[0102] In the formula, θ_c represents the azimuth angle of the target being tracked by the helmet-mounted system, atan2 represents the two-parameter arctangent function, y_p represents the y-axis coordinate of the target in the vehicle coordinate system, and x_p represents the x-axis coordinate of the target in the vehicle coordinate system; the pitch angle calculation formula is: In the formula, d_c represents the pitch angle of the target being tracked by the helmet drive, and z_p represents the z-axis coordinate of the target in the carrier coordinate system.

[0103] The azimuth and pitch angles of the target being tracked by the helmet-driven system are combined to form the target prediction command (θ_c, φ_c)[k+Δt].

[0104] A neural network prediction model is trained based on historical real data from time series to accurately predict the target state in the future Δt, fundamentally solving the "lag tracking" problem caused by delays in traditional servo systems. The predicted target data in the world coordinate system is inversely transformed to the carrier coordinate system to calculate the azimuth and pitch angles required for helmet actuation, ensuring that the prediction results are directly adapted to the helmet control logic and reducing command conversion losses.

[0105] S400: The helmet calculates the feedforward control quantity based on the predicted target command, and calculates the feedback control quantity based on the original target data and the helmet's own motion state data. The feedforward control quantity and the feedback control quantity are then fused to obtain the final control command.

[0106] The specific steps for fusing the feedforward control input and the feedback control input to obtain the final control command are as follows:

[0107] S401. Differentiate the predicted target command to obtain the desired acceleration and velocity during helmet-driven operation. Use the desired acceleration and velocity to generate the feedforward control quantity, as shown in the formula: In the formula, U_ff represents the feedforward control quantity, K_a represents the acceleration proportional gain, K_v represents the velocity proportional gain, a_c represents the desired acceleration, and v_c represents the desired velocity.

[0108] S402. Extract the current target angle (θ_m, d_m)[k] and the actual helmet angle (θ_h, d_h)[k] in the carrier coordinate system, and calculate the error between the target and the helmet using the following formula: In the formula, e[k] represents the current angle error between the target and the helmet; the feedback control quantity is calculated using the angle error, and the formula is:

[0109] ;

[0110] In the formula, U_fb represents the feedback control quantity, and K_p, Ki, and K_d represent the angle error, error integral, and error differential coefficients, respectively.

[0111] The feedforward control quantity and the feedback control quantity are fused to obtain the final control command, and the formula is: U=U_ff+U_fb, where U represents the final control command.

[0112] Based on the expected velocity and acceleration predicted by the target command, a feedforward control quantity is generated to drive the helmet to move towards the target's future position in advance, significantly improving the system's response speed to dynamic targets and avoiding the lag of "passive tracking". By calculating the angular error between the current target and the helmet, a feedback control quantity is generated in combination with a PID algorithm to correct the prediction deviation in the feedforward control in real time, achieving a dual guarantee of "prediction + correction". Feedforward control solves the response speed problem, and feedback control solves the accuracy deviation problem. The combination of the two makes the control command both "fast" and "accurate", balancing dynamic response and steady-state accuracy.

[0113] S500: A compensator is installed in the helmet. The final control command is input into the compensator, and the compensator compensates for the final control command to obtain the optimal control command.

[0114] The specific steps by which the compensator compensates for the final control command to obtain the optimal control command are as follows:

[0115] S501. The compensator is a nonlinear compensator. The nonlinear compensator is used to compensate for friction and tooth groove in the final control command and output the optimal control command.

[0116] By using a nonlinear compensator to specifically address mechanical nonlinear errors such as friction and cogging effect in the helmet drive mechanism, the final control command is corrected to the optimal command with "no loss", ensuring that the control signal is accurately converted into mechanical motion.

[0117] S600: The helmet rotates to track the target according to the optimal control command. After the control ends, the helmet collects the target prediction command and the actual position of the target, calculates the prediction error, and feeds the prediction error back to the prediction model for optimization training.

[0118] The specific steps for feeding the prediction error back into the prediction model for optimization training are as follows:

[0119] S601. Use the best control command to control the helmet drive to track the target. After the control is completed, the helmet collects the target prediction command and the actual position of the target. Calculate the difference between the predicted target command and the actual position of the target as the prediction error. Input the prediction error into the prediction model for optimization training.

[0120] The deviation between the predicted target command and the actual position is fed back to the neural network model, continuously optimizing the model parameters so that the prediction accuracy improves with the number of uses, adapts to the motion characteristics of different targets, and achieves "self-learning" evolution.

[0121] A high-precision drive and control system for a servo helmet based on artificial intelligence, comprising a data acquisition module, a data transformation module, a prediction module, an instruction generation module, an instruction optimization module, and an update module;

[0122] The data acquisition module is used to collect the target's raw data using the target tracking sensor and to collect the helmet's own motion state data using the body state sensor installed in the helmet when the helmet captures the target.

[0123] The data transformation module is used to optimize the helmet's own motion state data using the Kalman filter algorithm to obtain the optimal state estimate, generate a transformation matrix from the sensor coordinate system to the stable world coordinate system using the optimal state estimate, and process the original target data using the transformation matrix to obtain the target's true data.

[0124] The prediction module is used to calculate the real data of the target's changes over time in history based on time series calculations, input the real historical data into a neural network to train a prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target instructions.

[0125] The instruction generation module is used for the helmet to calculate the feedforward control quantity based on the predicted target instruction, calculate the feedback control quantity based on the original target data and the helmet's own motion state data, and fuse the feedforward control quantity and the feedback control quantity to obtain the final control instruction.

[0126] The instruction optimization module is used to set up a compensator in the helmet, input the final control instruction into the compensator, and the compensator compensates for the final control instruction to obtain the optimal control instruction.

[0127] The update module is used to collect the target prediction command and the actual position of the target after the control ends, calculate the prediction error, and feed the prediction error back to the prediction model for optimization training.

[0128] The data transformation module includes a transformation matrix calculation unit and a data transformation unit;

[0129] The transformation matrix calculation unit is used to calculate the transformation matrix from sensor coordinate system S to carrier coordinate system B and the transformation matrix from sensor coordinate system B to carrier coordinate system W respectively, and combine them to obtain the transformation matrix from sensor coordinate system S to carrier coordinate system W.

[0130] The data transformation unit is used to transform the original target data into real coordinates in the world coordinate system using a transformation matrix.

[0131] The instruction generation module includes a feedforward control unit, a feedback control unit, and a fusion unit;

[0132] The feedforward control unit is used to perform differential processing on the predicted target command to obtain the expected acceleration and velocity when the helmet is driven, and to generate the feedforward control quantity using the expected acceleration and velocity.

[0133] The feedback control unit is used to calculate the error between the target and the helmet, and to calculate the feedback control quantity using the angle error.

[0134] The fusion unit is used to fuse the feedforward control quantity and the feedback control quantity to obtain the final control command.

[0135] Example: Taking the ground control servo helmet for UAVs as the application scenario, when the helmet captures the UAV target, the target tracking sensor collects the raw data Z_t[0]=(r=50m, θ=30°, d=15°, v_m=[2m / s,0,0]) (three-dimensional velocity along the positive x-axis) at k=0; the IMU collects the helmet's own motion state data S_h[0]=(α=0°, β=0°, γ=0°, ω_α=0rad / s, ω_β=0rad / s, ω_γ=0rad / s, (x=0,y=0,z=1.5m), a_x=0m / s², a_y=0m / s², a_z=0m / s²).

[0136] The attitude angles in S_h[0] are converted into quaternions q=[1,0,0,0], and the helmet state vector X_ka=[q,p,ω,v]^T is constructed. After inputting into the Kalman filter, the IMU noise is filtered to obtain the optimal state estimate.

[0137] The transformation matrix is ​​calculated and then processed through T_S. W The world coordinate system true coordinates P_t are obtained after transformation. W ≈[41.11,21.65,14.44,1] T The actual speed V_t W ≈[2,0,0] T ;

[0138] Extract historical real data (11 sets in total) from k=-10 to k=0, input them into the LSTM neural network to train the prediction model, and set Δt=0.1s;

[0139] The model predicts the actual data at time k+0.1s:

[0140] X_p[0.1]=(P_t W '≈[41.31,21.65,14.44,1] T V_t W '≈[2,0,0] T );

[0141] Calculate the target prediction instruction: θ_c=atan2(21.65,41.21)≈27.8°, d_c≈15°, that is, (27.8°,15°)[0.1].

[0142] Differentiating the predicted target command yields the desired velocity v_c = 22° / s and the desired acceleration a_c = 0. Taking K_a = 0.5 and K_v = 0.8, the feedforward control quantity U_ff = 0.5 × 0 + 0.8 × 22 = 17.6.

[0143] Extract the current target angle (θ_m=30°, d_m=15°)[0], the actual helmet angle (θ_h=0°, d_h=0°)[0], the error e[0]=(30°,15°), take K_p=1.2, Ki=0.1, K_d=0.05, and calculate the feedback control quantity U_fb=1.2×(30,15)+0.1×∫(30,15)dt+0.05×(0,0)≈(36,18). The final control command obtained by fusion is U=17.6+(36,18)≈(53.6,35.6).

[0144] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A high-precision drive and control method for a servo helmet based on artificial intelligence, characterized in that: The method includes the following steps: S100. When the helmet captures a target, it uses a target tracking sensor to collect the target's raw data and uses a body state sensor installed in the helmet to collect the helmet's own motion state data. S200. The Kalman filter algorithm is used to optimize the motion state data of the helmet itself to obtain the optimal state estimate. The optimal state estimate is used to generate a transformation matrix from the sensor coordinate system to the stable world coordinate system. The transformation matrix is ​​used to process the original target data to obtain the target's true data. The specific steps for collecting the helmet's own motion data using the body state sensors installed in the helmet are as follows: S101. Collect the target's raw data Z_t[k]=(r, θ, d, v_m) using a target tracking sensor, where r represents the distance between the helmet and the target, θ represents the target's azimuth angle, d represents the target's pitch angle, v represents the target's velocity, including velocities in three directions; k represents the k-th time moment; collect the helmet's own motion state data S_h[k]=(α, β, γ, ω_α, ω_β, ω_γ, (x, y, z), a_x, a_y, a_z), α, β and γ represent the current roll, pitch, and yaw angles of the helmet, respectively; ω_α, ω_β, and ω_γ represent the angular velocities of the current roll, pitch, and yaw angles of the helmet, respectively; a_x, a_y, and a_z represent the linear velocities of the helmet itself in three-dimensional coordinates; and (x, y, z) represent the three-dimensional position coordinates. The target's original data is data in the sensor coordinate system, while the helmet's motion state data is data in the carrier coordinate system. The relative positions between the sensor coordinate system and the carrier coordinate system remain constant. S300: Based on time series calculation, obtain the real data of the target's change over time in history, input the real historical data into the neural network to train the prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target command. S400: The helmet calculates the feedforward control quantity based on the predicted target command, and calculates the feedback control quantity based on the original target data and the helmet's own motion state data. The feedforward control quantity and the feedback control quantity are then fused to obtain the final control command. S500: A compensator is installed in the helmet. The final control command is input into the compensator, and the compensator compensates for the final control command to obtain the optimal control command. S600: The helmet rotates to track the target according to the optimal control command. After the control ends, the helmet collects the target prediction command and the actual position of the target, calculates the prediction error, and feeds the prediction error back to the prediction model for optimization training.

2. The high-precision drive and control method for a servo helmet based on artificial intelligence according to claim 1, characterized in that: The specific steps in S200 to process the original target data using a transformation matrix to obtain the true target data are as follows: S201. Convert the roll, pitch, and yaw angles from the helmet's motion data into quaternion representations q = [q0, q1, q2, q3], where q0 represents the scalar part and q1, q2, and q3 represent the vector part. Extract the position coordinates to construct a position vector p = [x, y, z]. Extract the linear velocity coordinates to construct a velocity vector v = [a_x, a_y, a_z]. Extract the angular velocities of the helmet's current roll, pitch, and yaw angles to construct an angular velocity vector ω = [ω_α, ω_β, ω_γ]. Construct the helmet state vector X_ka = [q, p, ω, v]. T ; The helmet state vector is input into the Kalman filter algorithm for filtering optimization to obtain the optimal state estimate; S202. Regarding the sensor coordinate system S, the carrier coordinate system B, and the world coordinate system W, the sensor coordinate system has its origin at the sensor itself, and the carrier coordinate system has its origin at the helmet itself. The three-dimensional coordinates of the origin of the sensor coordinate system in the carrier coordinate system are p_S. B =[x_S B ,y_S B ,z_S B ] T The angle data of the sensor origin in the carrier coordinate system is collected and converted into quaternions, and then the quaternions are converted into a rotation matrix R_S. B The transformation matrix from the sensor coordinate system S to the carrier coordinate system B is constructed using the following formula: ； In the formula, T_S B This represents the transformation matrix from the sensor coordinate system S to the carrier coordinate system B. The transformation matrix is ​​a 4×4 matrix, and the three-dimensional coordinates p_S of the origin of the sensor coordinate system in the carrier coordinate system are... B It is a 3×1 matrix, 03 T Let represent a 1×3 zero vector, where 1 represents a scalar; Convert the quaternions in the optimal state estimation into a rotation matrix R_B W The 3×1 carrier position coordinates extracted from the optimal state estimate are p_B W The transformation matrix from the sensor coordinate system B to the carrier coordinate system W is constructed using the following formula: ； In the formula, T_B w The transformation matrix from sensor coordinate system B to carrier coordinate system W is represented by the transformation matrix T_S. Using these two transformation matrices, the transformation matrix from sensor coordinate system S to carrier coordinate system W is obtained. W =T_B w ×T_S B ; All transformation matrices are 4×4 matrices; S203. Calculate the homogeneous coordinates of the target in the sensor coordinate system using the target's original data. The formula is: ； In the formula, P_t S This represents the homogeneous coordinates of the target in the sensor coordinate system. The true coordinates of the target in the world coordinate system are calculated using the transformation matrix from the sensor coordinate system S to the carrier coordinate system W. The formula is: ； In the formula, P_t W The true coordinates of the target in the world coordinate system are given. The true velocity of the target in the world coordinate system is calculated using the following formula: ； In the formula, V_t W The target's true velocity in the world coordinate system is represented by a 3×1 velocity matrix with three dimensions; v_m represents the target's position coordinates in the original data, which is also a 3×1 matrix; ω_B W v_B represents the helmet angular velocity vector extracted from the optimal state estimation, and is a 3×1 matrix; W This represents the helmet velocity vector extracted from the optimal state estimation, which is a 3×1 matrix; By combining the actual coordinates and actual velocity, we obtain the actual data of the target in the world coordinate system.

3. The high-precision drive and control method for a servo helmet based on artificial intelligence according to claim 2, characterized in that: The specific steps in S300 to calculate the predicted target instruction using real data are as follows: S301. Based on time series calculation, the real data of the target changing over time in history is obtained. The real historical data is input into the neural network to train a prediction model. The prediction time is set to Δt. The prediction model is used to predict the real data after Δt in the future as X_p[k+Δt]. The prediction time = helmet system information processing time + execution delay time. S302. Perform an inverse transformation on the transformation matrix from sensor coordinate system B to carrier coordinate system W to obtain the transformation matrix T_W from sensor coordinate system W to carrier coordinate system B. B Extract the predicted true coordinates from the predicted true data, add scalar 1s to form a 4×1 matrix, and then use the transformation matrix T_W B Transform the predicted true coordinates to coordinates p_p in the carrier coordinate system B. B = (x_p, y_p, z_p), using coordinates p_p B Calculate the azimuth and pitch angles of the target being tracked by the helmet-mounted system, where the formula for calculating the azimuth angle is: ； In the formula, θ_c represents the azimuth angle of the target being tracked by the helmet-mounted system, atan2 represents the two-parameter arctangent function, y_p represents the y-axis coordinate of the target in the vehicle coordinate system, and x_p represents the x-axis coordinate of the target in the vehicle coordinate system; the pitch angle calculation formula is: In the formula, d_c represents the pitch angle of the target being tracked by the helmet drive, and z_p represents the z-axis coordinate of the target in the carrier coordinate system. The azimuth and pitch angles of the target being tracked by the helmet-driven system are combined to form the target prediction command (θ_c, φ_c)[k+Δt].

4. The high-precision drive and control method for a servo helmet based on artificial intelligence according to claim 3, characterized in that: The specific steps in S400 to fuse the feedforward control quantity and the feedback control quantity to obtain the final control command are as follows: S401. Differentiate the predicted target command to obtain the desired acceleration and velocity during helmet-driven operation. Use the desired acceleration and velocity to generate the feedforward control quantity, as shown in the formula: In the formula, U_ff represents the feedforward control quantity, K_a represents the acceleration proportional gain, K_v represents the velocity proportional gain, a_c represents the desired acceleration, and v_c represents the desired velocity. S402. Extract the current target angle (θ_m, d_m)[k] and the actual helmet angle (θ_h, d_h)[k] in the carrier coordinate system, and calculate the error between the target and the helmet using the following formula: In the formula, e[k] represents the current angle error between the target and the helmet; the feedback control quantity is calculated using the angle error, and the formula is: ； In the formula, U_fb represents the feedback control quantity, and K_p, Ki, and K_d represent the angle error, error integral, and error differential coefficients, respectively. The feedforward control quantity and the feedback control quantity are fused to obtain the final control command, and the formula is: U=U_ff+U_fb, where U represents the final control command.

5. The high-precision drive and control method for a servo helmet based on artificial intelligence according to claim 4, characterized in that: The specific steps in S500 for the compensator to compensate for the final control command to obtain the optimal control command are as follows: S501. The compensator is a nonlinear compensator. The nonlinear compensator is used to compensate for friction and tooth groove in the final control command and output the optimal control command.

6. The high-precision drive and control method for a servo helmet based on artificial intelligence according to claim 5, characterized in that: The specific steps in S600 for feeding the prediction error back to the prediction model for optimization training are as follows: S601. Use the best control command to control the helmet drive to track the target. After the control is completed, the helmet collects the target prediction command and the actual position of the target. Calculate the difference between the predicted target command and the actual position of the target as the prediction error. Input the prediction error into the prediction model for optimization training.

7. A high-precision drive and control system for a servo helmet based on artificial intelligence, applying the high-precision drive and control method for a servo helmet based on artificial intelligence as described in any one of claims 1-6, characterized in that: The high-precision drive and control system for the servo helmet includes a data acquisition module, a data transformation module, a prediction module, a command generation module, a command optimization module, and an update module; The data acquisition module is used to collect the target's raw data using the target tracking sensor and to collect the helmet's own motion state data using the body state sensor installed in the helmet when the helmet captures the target. The data transformation module is used to optimize the helmet's own motion state data using the Kalman filter algorithm to obtain the optimal state estimate, generate a transformation matrix from the sensor coordinate system to the stable world coordinate system using the optimal state estimate, and process the original target data using the transformation matrix to obtain the target's true data. The prediction module is used to calculate the real data of the target's changes over time in history based on time series calculations, input the real historical data into a neural network to train a prediction model, use the prediction model to predict the real data of the target's future time, and use the real data to calculate the predicted target instructions. The instruction generation module is used for the helmet to calculate the feedforward control quantity based on the predicted target instruction, calculate the feedback control quantity based on the original target data and the helmet's own motion state data, and fuse the feedforward control quantity and the feedback control quantity to obtain the final control instruction. The instruction optimization module is used to set up a compensator in the helmet, input the final control instruction into the compensator, and the compensator compensates for the final control instruction to obtain the optimal control instruction. The update module is used to collect the target prediction command and the actual position of the target after the control ends, calculate the prediction error, and feed the prediction error back to the prediction model for optimization training.

8. The high-precision drive and control system for a servo helmet based on artificial intelligence according to claim 7, characterized in that: The data transformation module includes a transformation matrix calculation unit and a data transformation unit; The transformation matrix calculation unit is used to calculate the transformation matrix from sensor coordinate system S to carrier coordinate system B and the transformation matrix from sensor coordinate system B to carrier coordinate system W respectively, and combine them to obtain the transformation matrix from sensor coordinate system S to carrier coordinate system W. The data transformation unit is used to transform the original target data into real coordinates in the world coordinate system using a transformation matrix.

9. The high-precision drive and control system for a servo helmet based on artificial intelligence according to claim 7, characterized in that: The instruction generation module includes a feedforward control unit, a feedback control unit, and a fusion unit; The feedforward control unit is used to perform differential processing on the predicted target command to obtain the expected acceleration and velocity when the helmet is driven, and to generate the feedforward control quantity using the expected acceleration and velocity. The feedback control unit is used to calculate the error between the target and the helmet, and to calculate the feedback control quantity using the angle error. The fusion unit is used to fuse the feedforward control quantity and the feedback control quantity to obtain the final control command.

Images