UGV active disturbance rejection control based on modeling compensation and extended disturbance observation

Abstract

The invention discloses a UGV active disturbance rejection control method based on modeling compensation and extended disturbance observation, and belongs to the technical field of unmanned ground vehicle trajectory tracking control. An active disturbance rejection control strategy for the UGV is provided by improving the estimation precision of an observer, and the dynamic disturbance rejection performance of the UGV is improved in a targeted manner by suppressing interference in a trajectory tracking process; firstly, part of information of a transverse control model is actively reserved in a state matrix so as to compensate estimation deviation of an extended state observer to a UGV state; then, the free variable used for estimating the high-order disturbance state is integrated into the high-order variable of the ESO, so that the estimation precision of the ESO on the total disturbance is further improved; according to a feedback linearization theory, an estimated value of UGV total disturbance is added into a control law, and accurate suppression of complex disturbance is achieved; a simpler and more convenient parameter configuration method is designed for key parameters of the controller.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned ground vehicle trajectory tracking control, and in particular relates to a UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation. Background Art

[0002] Unmanned Ground Vehicle (UGV) has attracted extensive attention from academia and industry due to its flexibility and ease of design. With the continuous development of sensor and computing technology, UGV has been widely used in transportation, industrial production and other fields. The practical application of UGV relies on unmanned driving technology, and its main purpose is to complete path planning and trajectory tracking through perception, decision-making and control. UGV control is generally divided into longitudinal control and lateral control. Due to longitudinal dynamics and controllable acceleration, longitudinal control is usually easier to design. However, due to the uncertainty of vehicle models, complex internal characteristics and sensitivity to external interference, the design of lateral controller becomes a challenge. For the lateral control of UGV, the existing control strategy does not compensate for the total disturbance observation deviation caused by observation order mismatch. Based on this, it is necessary to invent a method to enable UGV to maintain its tracking and anti-interference performance in a complex external operating environment. Summary of the invention

[0003] Aiming at the problem that the accurate tracking and dynamic disturbance rejection of the UGV control system cannot be guaranteed in a multi-level, highly coupled complex operating environment, the present invention provides a UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation. In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0004] The present invention proposes a UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation, which can simultaneously optimize the tracking and robustness of the UGV in the trajectory tracking process. In view of the uncertainty disturbance existing in the typical self-disturbance rejection control process, the model information of the controlled object is actively retained to compensate for the modeling deviation type disturbance. On the basis of retaining the model of the controlled object, the disturbance observation order is expanded to a high order to form an improved self-disturbance rejection control strategy (ADRC-MOEA) with optimized estimation accuracy. The stability of the system is analyzed using the Lienard-Chipart theory. In addition, a simpler parameter setting method is designed to improve the wide application of this control strategy.

[0005] A UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation, which is characterized by the following steps:

[0006] 1. Description of the model

[0007] Construct the Ackerman geometry model of the UGV system and decouple and linearize it to form a three-degree-of-freedom dynamic model, as shown in the attached figure. Figure 1 As shown, it can be expressed as

[0008]

[0009] Where d y (t) and ω(t) represent the lateral offset error, heading angle and the vehicle's rotation around its center of gravity, respectively. The yaw velocity, v represents The forward speed at f (t) is the steering angle of the front wheels of the vehicle. f and l r Respectively Distance between front and rear axles, C f and C r is the cornering stiffness of the front and rear wheels, I z is the inertia torque.

[0010] Second, the design of active disturbance rejection control strategy based on modeling compensation and extended disturbance observation

[0011] Section 2.1, Controller Structure Design

[0012] The ESO is designed to actively retain the known model information, and its state space expression is designed as

[0013]

[0014] In the formula and They are and the reconstructed variable matrix of x(t); and are the system matrix and control vector after retaining the model information, C is the output vector, L=[L 1 L 2 L 3 L 4 ] T represents the observer gain matrix, u(t) is the system control quantity;

[0015] Then introduce the two-order state variable and The internal model deviation and external disturbance of the control system are jointly compensated to accurately reconstruct the total disturbance state:

[0016]

[0017] In the formula and They represent the low-order and high-order components of the disturbance observation channel respectively, and the gain matrix L is expanded to L 5 ;

[0018] Section 2.2, Control Law Design

[0019] The purpose of this invention is to propose an auto-disturbance rejection control method based on modeling compensation and extended disturbance observation to deal with the impact of high-order disturbances in complex operating environments, so as to improve the trajectory tracking accuracy of UGV; under the premise of appropriate configuration of the observer gain matrix L, the proposed strategy can track the total disturbance in UGV in real time, so the UGV system trajectory tracking controller is designed to convert the total disturbance observation into Actively added to the control law, it can be expressed as:

[0020]

[0021] where k p and k d is the controller gain, b 0 is the gain of the known disturbance; y r (t) represents the input reference trajectory, is its derivative, the above formula corresponds to the control law module; Figure 2 The structural diagram of the designed controller is given;

[0022] Section 2.3, Parameter Setting Rule Design

[0023] The bandwidth method is used to adjust the parameters of the designed controller, and the closed-loop poles of the cascaded ESO are configured within the observer bandwidth ω. o The parameter configuration rule is written as:

[0024]

[0025] Where I is the unit matrix; expand and solve the above formula to get:

[0026]

[0027] Place the closed-loop pole of the controller at the controller bandwidth ω c At , the controller gain is solved as:

[0028]

[0029] Where ξ is the damping ratio of the control system, which is usually set to ξ=1. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Attached Figure 1 is a controlled object model of the present invention;

[0031] Attached Figure 2 It is a schematic diagram of the structure of the present invention;

[0032] Attached Figure 3 It is the frequency domain characteristic curve of input noise under different bandwidths;

[0033] Attached Figure 4 It is the frequency domain characteristic curve of the output noise under different bandwidths;

[0034] Attached Figure 5 It is the system equivalent structure model of the present invention;

[0035] Attached Figure 6 Comparison of the tracking process of error signals of different orders (left: w(t) = K, middle: w(t) = Kt, right: w(t) = Kt 2 ));

[0036] Attached Figure 7 Comparison of logarithmic amplitude and phase curves output by CLESO and ESO-MOEA;

[0037] Attached Figure 8 The simulation results of lateral control under double-lane-shifting trajectory (upper left: lateral displacement y(t), upper right: yaw angle Bottom left: Lateral tracking error e y (t), bottom right: yaw angle tracking error

[0038] Attached Fig. 9 The simulation results of lateral control under circular trajectory (upper left: UGV path without disturbance, upper right: lateral tracking error e without disturbance) y (t), lower left: e under white noise disturbance y (t), lower right: e under intermittent disturbance y (t));

[0039] Attached Fig.10 This is a schematic diagram of the TCar model robot platform;

[0040] Attached Fig.11 The experimental results of lateral control under single-line trajectory (upper figure: UGV path, lower figure: lateral deviation e y (t)). DETAILED DESCRIPTION

[0041] A UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation, the specific implementation steps include:

[0042] First, theoretical analysis

[0043] 1.1 Analysis of ADRC-MOEA’s noise disturbance suppression performance

[0044] In the total disturbance observation process, the internal disturbance is mainly manifested as the noise N at the observer input. r (s) and the noise N at the output side c (s), which affects the front wheel steering angle δ f The output feedback has a certain impact. ESO-MOEA output side equivalent noise N r (s) The transfer function satisfies:

[0045]

[0046] Take the observer bandwidth ω o =10ω′ o (ω′ o =1~4), and put it into (8) to draw the frequency domain characteristic curve as shown in the attached figure Figure 3 As shown in the figure, as the pole ω o As the system moves away from the imaginary axis, the response speed of the system increases, but the high-frequency gain on the output side also increases. Therefore, by selecting a suitable bandwidth, the system's sensitivity to high-frequency noise can be reduced while ensuring the system's response speed.

[0047] Similarly, the noise interference N on the output side of the observer can be obtained: c The transfer function of (s) is:

[0048]

[0049] Set the same bandwidth ω o Change rule, take b 0 =10Substitute into N c (s) in the transfer function, the output noise frequency domain characteristic curve changes as shown in the attached figure: Figure 4 Compared with the frequency domain characteristic curve of the input side noise under different bandwidths, the change of the output side observer bandwidth can improve the phase lag of the tracking signal without affecting the system's ability to resist high-frequency noise brought by the high-frequency gain. Therefore, the improved ESO can effectively suppress the output disturbance N c (s).

[0050] Section 1.2 Numerical analysis of noise immunity

[0051] Substituting equations (6) and (7) into the final control law (4) of the system, the control system satisfies the following equation:

[0052]

[0053] The model information of the controlled object is integrated into the feedforward channel transfer function G 1 (s) and the feedback channel transfer function H(s), the system equivalent structure model can be obtained as shown in the attached figure Figure 5 shown.

[0054] According to the attached Figure 5 The equivalent structural model given can be written as:

[0055]

[0056] Where c 0 ~c 8 It can be calculated by (10).

[0057] From (11), we can get the transfer function G from the total disturbance to the system output Y(s): f (s) and the transfer function G from the reference signal R(s) to the output terminal Y(s) c1 (s) are:

[0058]

[0059] Substituting into (10), the equivalent transfer function of the control system under external disturbance can be calculated as:

[0060]

[0061] Considering different external disturbance excitation functions w(t) input, the lateral deviation of the vehicle under the traditional CLESO and ESO-ADRC observations with expanded disturbance observation order are compared. The corresponding deviation expressions are shown in Table 1.

[0062] Table 1. Vehicle offset expressions under different disturbance functions

[0063]

[0064] All the equations in Table 1 satisfy The variable q is the power of the polynomial, and the variable p varies from 1 to 6, which respectively represent the corresponding relationships in Table 1. i Combining the observer bandwidth and controller bandwidth, the error C 1 ~C 4 Respectively expressed as

[0065]

[0066] Combining Table 1 with (14), we can plot the signal observation effects of CLESO and ESO-MOEA under different external disturbances. Figure 6As shown in the left and middle figures, when the external disturbance excitation function is a step function or a ramp function, the ADRC-MOEA controller with an expanded high-order observation dimension can track the disturbance efficiently, that is, the vehicle's driving trajectory can be consistent with the expected trajectory after a period of time, while the traditional ADRC control system will produce an uneliminated static error C when the external disturbance w(t) is a ramp function. 1 , which makes the error unable to converge to zero, causing the vehicle trajectory to continuously deviate from the desired trajectory.

[0067] When the external disturbance signal is an acceleration function, combined with the attached Figure 6 The right figure shows that both controllers have certain errors in the process of tracking the disturbance signal. However, since ADRC-MOEA has a second-order disturbance observation channel, it can be adjusted by changing the controller bandwidth ω c and observer bandwidth ω o The value of static error C 4 The impact on disturbance tracking makes the driving trajectory consistent with the expected trajectory. However, CLESO only has a first-order error observation dimension, and the disturbance compensation order is too low. CLADRC not only cannot 2 t+C 3 Effective compensation is performed, and as time t continues to increase, the controller will gradually lose the ability to track the error, and the vehicle will gradually deviate from the track. The above comparative analysis shows that ADRC-MOEA with high-order disturbance compensation dimension has stronger anti-disturbance ability and is more suitable for vehicle control.

[0068] Section 1.3: Analysis of the total disturbance observation process

[0069] All non-ideal components are defined as the total disturbance function, so the actual total disturbance of the system can be written as

[0070]

[0071] CLESO estimates the total disturbance It can be equivalently expressed by the input and output side control quantities as follows

[0072]

[0073] Combining equations (15) and (16), the transfer function G between the actual disturbance value and the disturbance estimate of the controlled object can be obtained: 0 (s)

[0074]

[0075] Disturbance estimation transfer function G 0(s) represents the influence of the observation gain matrix L on the process of the disturbance estimate of the extended observer gradually converging to the actual value. By changing the parameter configuration of the matrix L, the transfer function G can be changed. 0 (s), which further affects the convergence speed and effect of the observer.

[0076] For the purpose of simplifying calculation, the retained model information is regarded as part of the known disturbance, so the actual total disturbance of the improved ADRC-MOEA system can be written as:

[0077]

[0078] Similarly, the total disturbance estimation transfer function G′ can be obtained: 0 (s) is:

[0079]

[0080] Substituting the parameters in equations (6) and (10) and combining them with equation (19), we can obtain the logarithmic frequency characteristic curve of the controller output side as shown in the following figure: Figure 7 As shown in the figure, the improved ESO system uses the second-order state and The disturbance is observed, and a fixed zero and a free pole are added to the transfer function. Satisfactory dynamic performance can be obtained by designing the position of the newly added pole, and the properly placed zero can improve the convergence performance of ESO-MOEA. Both ensure that the controller meets the tracking requirements and suppresses signal noise.

[0081] Current bandwidth ω o When =10, it can be read from the figure that the phase margin of the traditional CLESO is 180° and the amplitude margin is 12dB, while the phase margin of the improved ESO-MOEA is 44° and the amplitude margin is 5.09dB, which improves the dynamic performance of the observer, has stronger adaptability to parameter changes and has a good transition process.

[0082] Example 1

[0083] Section 2.1 Trajectory Tracking Performance

[0084] Considering the application scenarios of UGV, a simulation was conducted in the double lane change condition to verify the feasibility of lateral control. r (t) vs. yaw angle As shown below:

[0085]

[0086] The solution p(t) = 0.096(vt-27.19)-1.2, q(t) = 0.1093(vt-56.46)-1.2. The generated reference trajectory is shown in the attached figure. Figure 8 As shown in the figure above.

[0087] UGV vehicle weight m = 1480kg, body length l = l f +l r =1.05+1.63=2.86m, the cornering stiffness of the front and rear wheels C f , C r They are 67500N / rad and 47500N / rad respectively, and the vehicle inertia torque I z Take 2350kgm 2 , the forward speed v is 20m / s. According to the observation accuracy of the system With observation time t d The performance requirements of , there are inequality constraints:

[0088]

[0089] In the formula, ε sup =10 4 is the perturbation set The supremum of x 1 (0) is the lateral trajectory deviation x 1 (t) is the initial state, is the corresponding ESO observation initial value. By setting the expected index requirements, the observer bandwidth ω can be solved o The value range is 98.528≤ω o ≤118.762, take ω o =110rad / s.

[0090] For comparison, the effectiveness of the PID controller and the CLADRC controller under double lane shifting conditions is also studied. To ensure the fairness of the comparison, the CLADRC controller parameters and the ADRC-MOEA are selected with the same bandwidth method, and the poles of both are placed at ω o = 110, and set the three parameters of CLADRC to [3ω o , 3ω o 2 ,ω o 3 ] = [330, 3.3 × 10 4 , 1.1×10 6 ], the three parameters of the PID controller are tuned to [3, 4, 0.05] by using the critical proportion method combined with the trial and error method.

[0091] In order to better reflect the lateral control effect under different control strategies, the lateral displacement y(t) and yaw angle are selected The tracking results are analyzed. Figure 8 As shown in Figure 2, all three controllers can track y(t) and y(t) while ensuring the convergence of the tracking error. In addition, the PID controller and CLADRC controller both oscillate when tracking the two signals, while the ADRC-MOEA control strategy relies on the role of high-order disturbance observation compensation to achieve the correction of y(t) and Fast and smooth tracking.

[0092] From the attached Figure 8 From the upper left figure, it can be seen that from the beginning of the yaw angle change to the final stabilization, the transition time required by the PID and CLADRC controllers is 4.86s and 4.75s respectively, while the ADRC-MOEA controller only needs 3.18s. Figure 8 As can be seen from the bottom figure, since the ADRC-MOEA strategy retains the model information of the controlled object, its maximum tracking error and root mean square tracking error are smaller than those of PID and CLADRC. Although the deviations of the PID and CLADRC controllers are within the acceptable range, they will affect the trajectory tracking performance in practical applications. Therefore, the ADRC-MOEA controller has better overshoot suppression ability and higher control accuracy for large curvature turns and rapid lane changes.

[0093] Section 2.2, Anti-interference performance

[0094] From the previous section, we know that the yaw angle Rapid changes in may cause the controller to oscillate slightly when tracking. A uniformly changing circular trajectory is used to test the anti-disturbance performance of the controller:

[0095]

[0096] The trajectory radius is R cir =1.5m, vehicle yaw speed ω cir =1rad / s. In addition, since the backstepping controller has good adaptability to trajectory tracking, the backstepping controller is added to the controller comparison experiment, and its four control parameters are all selected as 1.65.

[0097] The tracking results are as attached Fig. 9As shown in the figure above, the trajectories under the three controllers have a certain degree of deviation at the beginning. The backstepping controller fluctuates due to its low response speed and has a large lateral tracking error, while the two ADRC controllers fluctuate slightly due to their fast response. CLADRC has the longest transition time of 5.56s and the maximum offset of the expected trajectory is 0.15m. The backstepping controller significantly shortens the transition time, but the maximum offset increases to 0.17m. Compared with the backstepping controller, ADRC-MOEA shortens the transition time and reduces the maximum offset.

[0098] In addition, there are many undesirable disturbances in the process of UGV trajectory tracking, such as the measurement noise of the internal instruments of the vehicle system and the external disturbances during driving. Therefore, we consider adding white noise interference and intermittent disturbance to simulate the above two situations to test the anti-disturbance performance of the control system. Fig. 9 In the lower left figure, white noise interference will affect the transient performance of the controller. The recovery time of ADRC-MOEA increases by 0.42s, but compared with CLADRC and backstepping control strategies, the dynamic response speed is still improved by 62% and 15%.

[0099] When a large intermittent disturbance is added to the UGV system, the recovery time increases significantly. The intermittent disturbance with an amplitude of 0.3m enters at t=1s and exits at t=3s. The experimental results are shown in the attached figure. Fig. 9 As shown in the lower right figure. Since there is only a first-order error observation order, the CLADRC controller requires the longest recovery time of 5.9s. The backstepping control controller shortens the recovery time to 3.8s, and the ADRC-MOEA further shortens the recovery time to 2.7s. Compared with the CLADRC and backstepping controllers, the ADRC-MOEA has a faster response speed and active suppression capability under the action of white noise and intermittent disturbances, thereby reducing the recovery time of lateral trajectory tracking.

[0100] Section 2.3, Physical Experiment

[0101] In order to further verify the control strategy proposed in this paper, a vehicle-type mobile robot platform (attached Fig.10 ) were tested experimentally. The vehicle motion process is driven by a brushless motor, and encoders and six-axis inertial measurement units are used to measure the speed and record the position of TCar in real time. The upper layer of the vehicle control system is the NVIDIA Jetson Nano computing platform, and the lower layer uses STM32 to receive real-time data from vehicle sensors and upper-level control instructions. In this experiment, the control algorithm was written and executed in the Simulink module on the workstation, and the workstation communicated with the vehicle via WiFi.

[0102] The body weight of TCar is m = 3.5kg, the body length is l = 0.38m, and C f , C r They are 1700N / rad, 1340N / rad, I z Take 2.37kgm 2 The single lane change trajectory is selected, and the longitudinal speed v is fixed at 0.4m / s. The starting point of the vehicle is (0m, 0m, 0rad), during which it undergoes a lateral lane change with an offset distance of 2m, and returns to the initial lane at t=7.5s.

[0103] The experiment selected CLADRC and PID methods for comparison. The trajectory tracking results are shown in the attached figure. Fig.11 As shown in Figure 2, both PID and CLADRC controllers oscillate within an acceptable range during tracking, while ADRC-MOEA has the highest accuracy during trajectory tracking. Compared with the CLADRC method, the ADRC-MOEA strategy reduces the maximum offset by 40.3%.

[0104] 3. Conclusion

[0105] The present invention proposes a UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation. First, by reconstructing the UGV model, the modeling deviation and internal uncertainty interference are successfully compensated, thereby reducing the observation burden of the extended state observer. Then, by expanding the disturbance observation order, the total disturbance is accurately estimated and suppressed. On the basis of the linearization of the self-disturbance rejection controller, the UGV controller is designed in combination with disturbance observation, which improves the trajectory tracking effect of the UGV in a strong uncertain environment. Based on the analysis of stability, total disturbance observation capability and anti-disturbance performance, the effectiveness of the ADRC-MOEA strategy proposed in the present invention is verified. Afterwards, multiple scenario tests were used to verify that the ADRC-MOEA strategy is superior to the comparative method in UGV trajectory tracking.

[0106] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in the field. Although the illustrative specific embodiments of the present invention are described above to facilitate the understanding of the present invention by the technical personnel in the field, it should be clear that the present invention is not limited to the scope of the specific embodiments. For the ordinary technical personnel in the field, as long as various changes are within the spirit and scope of the present invention defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A UGV self-disturbance rejection control method based on modeling compensation and extended disturbance observation, characterized in that: The steps include:

1. Description of the model The Ackerman geometric model of the UGV system is constructed and decoupled and linearized to form a three-degree-of-freedom dynamic model, as shown in Figure 1, which can be expressed as Where d y (t) and ω(t) represent the lateral offset error, heading angle and the vehicle's rotation around its center of gravity, respectively. The yaw velocity, v represents The forward speed at f (t) is the steering angle of the front wheels of the vehicle. f and l r Respectively Distance between front and rear axles, C f and C r is the cornering stiffness of the front and rear wheels, I z is the inertia torque. Second, the design of active disturbance rejection control strategy based on modeling compensation and extended disturbance observation Section 2.1, Controller Structure Design The ESO is designed to actively retain the known model information, and its state space expression is designed as In the formula and They are and the reconstructed variable matrix of x(t); and are the system matrix and control vector after retaining the model information, C is the output vector, L = [L1 L2 L3 L4] T represents the observer gain matrix, u(t) is the system control quantity; Then introduce the two-order state variables and The internal model deviation and external disturbance of the control system are jointly compensated to accurately reconstruct the total disturbance state: In the formula and They represent the low-order and high-order components of the disturbance observation channel respectively, and the gain matrix L is expanded to L5; Section 2.2, Control Law Design The purpose of this invention is to propose an auto-disturbance rejection control method based on modeling compensation and extended disturbance observation to deal with the impact of high-order disturbances in complex operating environments, so as to improve the trajectory tracking accuracy of UGV; under the premise of appropriate configuration of the observer gain matrix L, the proposed strategy can track the total disturbance in UGV in real time, so the UGV system trajectory tracking controller is designed to convert the total disturbance observation into Actively added to the control law, it can be expressed as: where k p and k d is the controller gain, b0 is the gain of the known disturbance; y r (t) represents the input reference trajectory, As its derivative, the above formula corresponds to the control law module; Figure 2 shows the structure diagram of the designed controller; Section 2.3, Parameter Setting Rule Design The bandwidth method is used to adjust the parameters of the designed controller, and the closed-loop poles of the cascaded ESO are configured within the observer bandwidth ω. o The parameter configuration rule is written as: Where I is the unit matrix; expand and solve the above formula to get: Place the closed-loop pole of the controller at the controller bandwidth ω c At , the controller gain is solved as: Where ξ is the damping ratio of the control system, which is usually set to ξ=1.

Images