Unicycle bending steering control method and system based on dynamic zero point and MLP prediction

Abstract

The application discloses a unicycle pressure bending steering control method and system based on a dynamic zero point and MLP prediction, and comprises the following steps: S1, setting a dynamic zero point update period; S2, collecting eight posture parameters of the unicycle in real time; S3, calculating a basic dynamic zero point; S4, calculating a roll error; S5, inputting the eight posture parameters, the basic dynamic zero point and the roll error into an MLP to obtain a dynamic zero point compensation amount; S6, summing the dynamic zero point compensation amount after weighting and the basic dynamic zero point to obtain a dynamic zero point; and S7, steering control: when the running speed is in a high-speed range, the unicycle enters a high-speed pressure bending mode, and the final dynamic zero point is equal to the dynamic zero point of S6; when the running speed is in a low-speed range, the unicycle enters a low-speed safe steering mode, and the dynamic zero point of S6 is reduced to form a final dynamic zero point. The application can control the steady-state error of the final dynamic zero point to be within 0.5°-1.0°, and the dynamic zero point output delay is not greater than 10 ms.

Description

Technical Field

[0001] This invention relates to the field of cornering and steering control of unicycles, and in particular to a method and system for cornering and steering control of unicycles based on dynamic zero point and MLP prediction. Background Technology

[0002] Unicycle self-balancing vehicles have broad application prospects in intelligent racing, inspection robots, and confined indoor environments due to their compact structure, high mobility, and small turning radius. However, compared to two-wheeled or multi-wheeled mobile platforms, unicycles are highly dependent on attitude stability control during movement, and their system dynamics exhibit significant strong nonlinearity, high coupling, and time-varying characteristics.

[0003] The existing steering control methods for unicycles mainly include: (1) using the differential speed of dual momentum wheels or multiple actuators to generate yaw torque to achieve steering; and (2) using the tilt angle set manually in low-speed scenarios to achieve passive steering. Although the above steering control methods can achieve steering control to a certain extent, they generally have problems such as complex structure, large number of actuators, increased system mass and power consumption, and difficulty in adjusting control parameters. Especially during high-speed travel or rapid steering, they are prone to attitude oscillation or even instability.

[0004] To simplify the system structure, unicycle designs employing a single momentum wheel have emerged in recent years. These designs typically alter the system's equilibrium state, causing controlled tilting of the vehicle body, and then utilizing the gravitational component and the lateral force of the tires to achieve a steering method similar to a motorcycle's "cornering." However, this type of steering is highly sensitive to the system's equilibrium zero point.

[0005] In actual operation, the dynamic equilibrium zero point of a unicycle will drift in real time with factors such as travel speed, steering requirements, changes in wheel angular momentum, and ground disturbances. If a fixed zero point or a linear compensation method based on a simplified model is still used, it is often difficult to accurately describe the real system state, resulting in lag in steering response, excessive tilting, or unstable control.

[0006] Therefore, there is a need for a control system and method that can accurately predict and adjust the dynamic zero point of the system in real time on a single momentum wheel unicycle platform, so as to achieve safe, stable and highly responsive bending and steering control. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a unicycle cornering and steering control method and system based on dynamic zero point and MLP prediction. This unicycle cornering and steering control method and system based on dynamic zero point and MLP prediction can control the steady-state error of the final dynamic zero point relative to the ideal dynamic zero point within 0.5° to 1.0°; at the same time, the dynamic zero point output delay is no more than 10 ms.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A method for controlling the cornering and steering of a unicycle based on dynamic zero point and MLP prediction includes the following steps.

[0010] S1. Set the dynamic zero-point update cycle.

[0011] S2. During each dynamic zero-point update cycle, eight attitude parameters of the unicycle are collected in real time, namely: roll angle. Roll rate Pitch angle Pitch angular velocity speed of travel yaw rate angular velocity of the moving wheels and momentum wheel angular velocity .

[0012] S3, based on the travel speed collected in S2 and yaw rate The fundamental dynamic zero point is calculated. .

[0013] S4. Roll angle collected from S2 The final dynamic zero point obtained from the previous dynamic zero point update cycle The roll error was calculated. .

[0014] S5. Combine the 8 attitude parameters acquired in S2 with the basic dynamic zero point obtained in S3. The roll error obtained from S4 All of these are used as input variables and fed into the MLP to predict the dynamic zero-point compensation amount. .

[0015] S6, adjust the dynamic zero-point compensation amount of S5. Weighted and the basic dynamic zero point Summing yields the dynamic zeros. .

[0016] S7. Based on travel speed and dynamic zero point Perform the following steering control:

[0017] A. When the speed of travel When within the set high-speed range, the unicycle enters high-speed cornering mode; at this time, the final dynamic zero point... .

[0018] B. When the speed of travel When within the set low-speed range, the unicycle enters a low-speed safe steering mode; at this time, the final dynamic zero point... ;in, A scaling factor less than 1.

[0019] In S3, a safety tilt angle upper limit is set. For the basic dynamic zero point The specific expression for amplitude limiting is as follows:

[0020] ;

[0021] In the formula, g is the gravitational acceleration, and sat(·) is the amplitude limiting function.

[0022] In S3, the basic dynamic zero point Apply the following rate-of-change constraint:

[0023] ;

[0024] In the formula, T s This refers to the duration of the dynamic zero-point update cycle; This is the base dynamic zero point for the previous dynamic zero point update cycle k-1; The maximum permissible rate of change of the basic dynamic zero point.

[0025] If the base dynamic zero point of the current dynamic zero point update cycle meets the rate of change constraint, then the base dynamic zero point of the current dynamic zero point update cycle is used directly; otherwise, it is rolled back. As the basis for the current dynamic zero-point update cycle, the dynamic zero point... .

[0026] In S5, if any input variable is missing or abnormal, or if the MLP output duration exceeds half the dynamic zero-point update cycle duration, the dynamic zero-point compensation amount is directly adjusted. It is zero.

[0027] In S6, dynamic zero point The calculation formula is:

[0028] ;

[0029] In the formula, λ is the compensation weight coefficient, 0≤λ≤1, and its value is related to the travel speed. Related; when traveling speed The compensation weight coefficient when the vehicle is within the set high-speed range shall not be less than the travel speed. The value of the compensation weight coefficient when the speed is within the set low speed range.

[0030] In S6, the dynamic zero point of the calculation is... The following formula is used to perform a continuous periodic consistency check:

[0031] ;

[0032] In the formula, , and These are the dynamic zeros of the previous two dynamic zero update cycles k-2, the previous dynamic zero update cycle k-1, and the current dynamic zero update cycle k, respectively.

[0033] when When the continuous cycle consistency check condition is met, the dynamic zero-point compensation amount in S5 is set to... If the result is zero, recalculate and update. ;otherwise It remains unchanged.

[0034] In S7, scaling factor The calculation formula is:

[0035] ;

[0036] In the formula, v th The reference speed threshold for transitioning from low-speed safe steering mode to high-speed cornering mode.

[0037] In S7, the high-speed range is set to [v th1,min , v th1,max ], set the low speed range to [v th2,min , v th2,max ]; where v th1,min v th1,max v th2,min and v th2,max These are the minimum value in the high-speed range, the maximum value in the high-speed range, the minimum value in the low-speed range, and the maximum value in the low-speed range, respectively, and v th1,min ≥ v th2,max .

[0038] In S5, the MLP is trained with training samples before use; the training samples include simulation samples, real vehicle operation samples and abnormal enhancement samples; the abnormal enhancement samples include dynamic zero-point jitter samples, dynamic zero-point lag samples and dynamic zero-point mutation samples.

[0039] A unicycle cornering and steering control system based on dynamic zero point and MLP prediction includes a controller installed inside the unicycle body; the controller has a built-in memory that stores the aforementioned unicycle cornering and steering control method based on dynamic zero point and MLP prediction.

[0040] The present invention has the following beneficial effects:

[0041] 1. The steady-state error of the final dynamic zero point of the present invention relative to the ideal dynamic zero point can be controlled within 0.5° to 1.0°.

[0042] 2. The output delay of the dynamic zero point of this invention is no more than 10 ms.

[0043] 3. The dynamic zero-point change within adjacent control cycles of this invention meets the preset change rate limit, and the maximum allowable change rate is no greater than [missing value]. .

[0044] 4. The final dynamic zero point of this invention can meet the high-frequency jitter suppression condition, that is, there is no rapid alternation between positive and negative values ​​within the continuous sampling period, thus avoiding dynamic zero points. High-frequency disturbances are directly injected into the roll closed loop, thereby suppressing high-frequency attitude oscillations, bending irregularities, and frequent actuator impacts. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the unicycle structure based on dynamic zero point and MLP prediction of the present invention.

[0046] Figure 2 This is a flowchart of the unicycle cornering and steering control method based on dynamic zero point and MLP prediction of the present invention.

[0047] Figure 3 This is a schematic diagram of the structure of the MLP of the present invention.

[0048] Among them are: 10. Propulsion wheel; 20. Momentum wheel. Detailed Implementation

[0049] The present invention will now be described in further detail with reference to the accompanying drawings and specific preferred embodiments.

[0050] In the description of this invention, it should be understood that the terms "left side," "right side," "upper part," "lower part," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. "First," "second," etc., do not indicate the importance of the components, and therefore should not be construed as a limitation of this invention. The specific dimensions used in this embodiment are only for illustrating the technical solution and do not limit the scope of protection of this invention.

[0051] like Figure 1 As shown, a unicycle based on dynamic zero point and MLP prediction includes a travel wheel 10 and a momentum wheel 20.

[0052] The aforementioned travel wheel, also referred to as the W-wheel, is used to contact the ground and provide forward driving force. The drive shaft of the travel wheel is preferably connected to two motors located at the bottom of the unicycle body via belt drive; the two motors are used to drive the unicycle forward and backward respectively. During reversal, by switching motors and applying preload torque, the torque transition of the travel wheel from forward to reverse is made more continuous, dead zone is reduced, forward and backward rotation response is accelerated, and pitch balance ring correction is more timely.

[0053] The aforementioned momentum wheel is mounted on the upper part of the vehicle body, and its axis of rotation is preferably tilted at 45° relative to the vertical direction. The output torque of the momentum wheel is decomposed in the roll direction and the longitudinal direction, forming roll component and longitudinal component. Among them, the longitudinal component provides part of the forward driving force during high-speed movement, which reduces the driving pressure on the motor that drives the traveling wheels forward, reduces the risk of peak current and temperature rise, and improves the posture stability of the unicycle in high-speed cornering mode.

[0054] The unicycle in this invention has only one momentum wheel, so it does not rely on the differential speed of two momentum wheels to achieve steering. Instead, it forms a stable roll angle through dynamic zero-point adjustment. And based on the speed of travel Constructing the driving trajectory on the curve. Final dynamic zero point. The real-time calculation accuracy requirements are extremely high: if the final dynamic zero point Vibration can cause the roll ring to oscillate at high frequencies; if there is a final dynamic zero-point lag, the vehicle will exhibit a trajectory deviation of "going straight first and then suddenly turning".

[0055] like Figure 2 As shown, a unicycle cornering and steering control method based on dynamic zero point and MLP prediction includes the following steps.

[0056] S1, Set the duration T of the dynamic zero-point update cycle. s In this embodiment, a fixed-period update mechanism is adopted, preferably T. s =10ms, thus ensuring that the dynamic zero-point output delay is no more than 10ms.

[0057] S2. During each dynamic zero-point update cycle, eight attitude parameters of the unicycle are collected in real time, namely: roll angle. Roll rate Pitch angle Pitch angular velocity speed of travel yaw rate angular velocity of the moving wheels and momentum wheel angular velocity .

[0058] S3, based on travel speed and yaw rate The fundamental dynamic zero point is calculated. The preferred method includes the following steps.

[0059] S3-1, Calculating the Fundamental Dynamic Zero Point base value Specifically:

[0060] ;

[0061] In the formula, g is the acceleration due to gravity.

[0062] S3-2, Set the upper limit of the safe tilt angle In this embodiment, it is preferred that... The value ranges from 10° to 18°, representing the basic value for the fundamental dynamic zero point. After applying the amplitude limiter, the following results are obtained: Specifically:

[0063] ;

[0064] In the formula, sat(·) is the amplitude limiting function.

[0065] This invention limits the basic dynamic zero point within a safe range to prevent the reference tilt angle from becoming too large under conditions of high curvature or high speed.

[0066] S3-3, Basic Dynamic Zero Point Apply the following rate-of-change constraint:

[0067] ;

[0068] In the formula, The base dynamic zero point is the previous dynamic zero point update cycle k-1, where k is the index value of the current dynamic zero point update cycle.

[0069] The maximum permissible rate of change of the basic dynamic zero point is typically taken as a value. .

[0070] If the base dynamic zero point of the current dynamic zero point update cycle meets the rate of change constraint, then the base dynamic zero point of the current dynamic zero point update cycle is used directly; otherwise, it is rolled back. As the basis for the current dynamic zero-point update cycle, the dynamic zero point... Setting the rate of change constraint can prevent the basic dynamic zero point from rapidly transitioning at the entrance of a curve or during a sudden change in steering, thus stabilizing and smoothly transitioning it, and ensuring the smooth transition of the subsequent final dynamic zero point. Steady-state error control relative to the ideal dynamic zero point is within Within.

[0071] S4. Roll angle collected from S2 The final dynamic zero point obtained from the previous dynamic zero point update cycle The roll error was calculated. Specifically: Initially, it can be set to It is zero.

[0072] S5. Combine the 8 attitude parameters acquired in S2 with the basic dynamic zero point obtained in S3. The roll error obtained from S4 All variables are used as input variables. After normalization, they together form the input vector X, which is input to the MLP to predict the dynamic zero-point compensation amount. .

[0073] The above input vector X can be written as:

[0074]

[0075] Furthermore, the input vector X can also incorporate historical state variables from the previous few sampling periods to enhance its ability to characterize dynamic zero-point lag, transient drift, and short-term jitter.

[0076] Since the basic dynamic zero point is mainly derived from the analytical mapping of travel speed and yaw rate, although it can quickly provide a bending reference, its ability to characterize model simplification errors, external disturbances, actuator dynamic hysteresis, and multivariable nonlinear coupling factors remains limited. Therefore, this invention introduces MLP to predictively compensate for the basic dynamic zero point, thereby reducing the steady-state deviation of the final dynamic zero point relative to the ideal dynamic zero point and suppressing high-frequency fluctuations in the compensation output, thus improving the accuracy and smoothness of the dynamic zero point reference.

[0077] The MLP mentioned above is a multilayer perceptron, such as Figure 3 As shown, a lightweight feedforward structure with 10-dimensional input variables, two hidden layers, and a 1-dimensional output is preferred. The number of neurons in the hidden layers is 32 and 16, respectively, and the output is a dynamic zero-point compensation quantity. .

[0078] Before use, the MLP is pre-trained using training samples, which include simulation samples, real vehicle operation samples, and abnormal enhancement samples. The abnormal enhancement samples include dynamic zero-point jitter samples, dynamic zero-point hysteresis samples, and dynamic zero-point mutation samples. This invention, by introducing abnormal enhancement samples, improves the MLP's ability to learn the dynamic zero-point compensation rules under complex operating conditions, thereby enhancing the accuracy and stability of the final dynamic zero-point adjustment.

[0079] The aforementioned dynamic zero-point jitter samples are preferably obtained at a constant speed with a small curvature (preferably curvature no greater than 0.2m). -1The high-frequency perturbation (preferably with an amplitude of 0.2° to 1.0°), sensor noise, or lateral impact are superimposed under turning conditions, which can be used to cover dynamic zero-point jitter conditions.

[0080] The aforementioned dynamic zero-lag type samples are preferably obtained by recording trajectory deviations and attitude responses under conditions such as step turn, S-curve switching, or sudden speed change, and can be used to cover dynamic zero-lag conditions.

[0081] The aforementioned dynamic zero-point mutation type samples are preferably obtained by simulating short-term sensor distortion, changes in road surface adhesion, or load shifts, and can be used to cover dynamic zero-point mutation conditions.

[0082] To make the final dynamic zero point closer to the ideal dynamic zero point, the training phase will supervise the learning of labels.

[0083] Defined as the difference between the ideal dynamic zero and the fundamental dynamic zero, i.e.:

[0084] ;

[0085] in, It can be obtained through experimental calibration, error minimization, time advance compensation, or closed-loop correction.

[0086] In this invention, the MLP learns "how much compensation is needed above the basic dynamic zero," rather than directly learning the final dynamic zero itself. Therefore, it can effectively reduce the steady-state error of the final dynamic zero relative to the ideal dynamic zero, and preferably maintain its control accuracy at a level that is within the range of the ideal dynamic zero. Within the range.

[0087] Furthermore, in the training phase of this invention, the mean squared error loss function and the Adam optimizer are preferably used, and the learning rate is preferably set to [value missing]. The batch size is preferably set to 64.

[0088] Furthermore, if any input variable is missing or abnormal, or if the MLP output duration exceeds half the dynamic zero-point update cycle duration, the dynamic zero-point compensation amount is directly adjusted. The value is zero, without waiting for the compensation result to return, thereby further ensuring that the dynamic zero-point output delay is no more than 10ms.

[0089] S6, adjust the dynamic zero-point compensation amount of S5. Weighted and the basic dynamic zero point Summing yields the dynamic zeros. The specific expression is:

[0090] ;

[0091] In the formula, λ is the compensation weight coefficient, 0≤λ≤1, and its value is related to the travel speed. Related; when traveling speed The compensation weight coefficient when the vehicle is within the set high-speed range shall not be less than the travel speed. The compensation weight coefficient is set to a value that reduces the dominant role of prediction compensation under low-speed conditions when the speed is within a set low-speed range.

[0092] In this embodiment, the preferred value of the compensation weight coefficient λ is:

[0093] A. When the speed of travel When within the set high-speed range, λ is preferably 0.3 to 1.

[0094] B. When the speed of travel When the speed is within the set low speed range, λ is preferably 0 to 0.3.

[0095] Furthermore, this invention provides for the calculation of dynamic zeros. The preferred method for continuous periodic consistency checks is as follows:

[0096] ;

[0097] In the formula, , and These are the dynamic zeros of the previous two dynamic zero update cycles k-2, the previous dynamic zero update cycle k-1, and the current dynamic zero update cycle k, respectively.

[0098] when When the continuous cycle consistency check condition is met, it indicates that the dynamic zero point has reversed its direction of change within an adjacent control cycle. At this time, the dynamic zero point compensation amount in S5 is directly set. If the result is zero, recalculate and update. To make the dynamic zero point Degenerate to the fundamental dynamic zero point recalculated in the current period; otherwise It remains unchanged.

[0099] This invention, through the aforementioned continuous periodic consistency check, can prevent the oscillating component in the compensation amount from being directly transmitted to the roll closed loop, thus preventing dynamic zero points. During continuous control, persistent high-frequency oscillations occur, thereby suppressing high-frequency attitude oscillations, uneven bending, and frequent actuator impacts.

[0100] S7. Based on travel speed and dynamic zero point The following steering control will be performed.

[0101] A. When the speed of travel Within the set high speed range [v th1,min , v th1,max At this point, the unicycle enters high-speed cornering mode; at this time, the final dynamic zero point... Among them, v th1,min and v th1,max These are the minimum and maximum values ​​within the high-speed range, respectively; in this embodiment, v is preferred. th1,min =1.0 m / s, v th1,max = v max m / s, v max This represents the maximum speed at which the unicycle can run, and is a set value.

[0102] B. When the speed of travel Within the set low speed range [v th2,min , v th2,max When [v] is engaged, the unicycle enters a low-speed, safe steering mode. th2,min and v th2,max These are the minimum and maximum values ​​in the low-speed range, respectively, and v th2,max ≤v th1,min In this embodiment, v is preferred. th2,min =0 m / s, v th2,max =1.0 m / s.

[0103] At low speeds, the lateral support required for cornering is significantly reduced due to the lower vehicle speed. If the dynamic zero-point output method used at high speeds is applied in this condition, it can easily lead to excessive body roll, over-adjustment of posture, and uneven steering. Therefore, in the low-speed safe steering mode, the dynamic zero point is weakened by reducing its amplitude, lowering the compensation ratio, and strengthening posture constraints, allowing the unicycle to achieve steering adjustments more smoothly in the low-speed range.

[0104] Therefore, in low-speed safe steering mode, the final dynamic zero point for:

[0105] ;

[0106] in:

[0107] ;

[0108] in, A scaling factor less than 1.

[0109] v th As a reference speed threshold for the transition from low-speed safe steering mode to high-speed cornering mode, v is preferred in this embodiment. th = v th2,max =v th1,min =1.0 m / s.

[0110] When the unicycle speed decreases This decreases accordingly, causing the dynamic zero-point output amplitude to decrease synchronously, automatically weakening the intensity of the dynamic zero-point action, and preventing the unicycle from forming excessive lean angles during start-up, low-speed maneuvering, or micro-steering; as the unicycle speed gradually increases and approaches the mode switching threshold, Gradually increasing the size allows for a smooth transition to high-speed bending mode.

[0111] A unicycle cornering and steering control system based on dynamic zero point and MLP prediction includes a controller installed inside the unicycle body; the controller has a built-in memory that stores the aforementioned unicycle cornering and steering control method based on dynamic zero point and MLP prediction.

[0112] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for unicycle cornering and steering control based on dynamic zero point and MLP prediction, characterized in that: include: S1. Set the dynamic zero-point update cycle; S2. During each dynamic zero-point update cycle, eight attitude parameters of the unicycle are collected in real time, namely: roll angle. Roll rate Pitch angle Pitch angular velocity speed of travel yaw rate angular velocity of the moving wheels and momentum wheel angular velocity ; S3, based on the travel speed collected in S2 and yaw rate The fundamental dynamic zero point is calculated. ; By setting an upper limit for the safe tilt angle For the basic dynamic zero point The specific expression for amplitude limiting is as follows: ； In the formula, g is the acceleration due to gravity, and sat(·) is the limiting function; For the basic dynamic zero point Apply the following rate-of-change constraint: ； In the formula, T s This refers to the duration of the dynamic zero-point update cycle; This is the base dynamic zero point for the previous dynamic zero point update cycle k-1; The maximum permissible rate of change of the basic dynamic zero point; If the base dynamic zero point of the current dynamic zero point update cycle satisfies the rate of change constraint, then the base dynamic zero point of the current dynamic zero point update cycle is used directly. Otherwise, roll back to the previous method. As the basis for the current dynamic zero-point update cycle, the dynamic zero point... ; S4. Roll angle collected from S2 The final dynamic zero point obtained from the previous dynamic zero point update cycle The roll error was calculated. ; S5. Combine the 8 attitude parameters acquired in S2 with the basic dynamic zero point obtained in S3. The roll error obtained from S4 All of these are used as input variables and fed into the MLP to predict the dynamic zero-point compensation amount. ; S6, adjust the dynamic zero-point compensation amount of S5. Weighted and the basic dynamic zero point Summing yields the dynamic zeros. The calculation formula is as follows: ； In the formula, λ is the compensation weight coefficient, 0≤λ≤1, and its value is related to the travel speed. Related; when traveling speed The compensation weight coefficient when the vehicle is within the set high-speed range shall not be less than the travel speed. The value of the compensation weight coefficient when the speed is within the set low speed range; For the dynamic zero point of the calculation The following formula is used to perform a continuous periodic consistency check: ； In the formula, , and These are the dynamic zeros of the previous two dynamic zero update cycles k-2, the previous dynamic zero update cycle k-1, and the current dynamic zero update cycle k, respectively. when When the continuous cycle consistency check condition is met, the dynamic zero-point compensation amount in S5 is set to... If the result is zero, recalculate and update. ;otherwise Remain unchanged; S7. Based on travel speed and dynamic zero point Perform the following steering control: A. When the speed of travel When within the set high-speed range, the unicycle enters high-speed cornering mode; at this time, the final dynamic zero point... ; B. When the speed of travel When within the set low-speed range, the unicycle enters a low-speed safe steering mode; at this time, the final dynamic zero point... ;in, A scaling factor less than 1.

2. The unicycle cornering and steering control method based on dynamic zero point and MLP prediction according to claim 1, characterized in that: In S5, if any input variable is missing or abnormal, or if the MLP output duration exceeds half the dynamic zero-point update cycle duration, the dynamic zero-point compensation amount is directly adjusted. It is zero.

3. The unicycle cornering and steering control method based on dynamic zero point and MLP prediction according to claim 1, characterized in that: In S7, scaling factor The calculation formula is: ； In the formula, v th The reference speed threshold for transitioning from low-speed safe steering mode to high-speed cornering mode.

4. The unicycle cornering and steering control method based on dynamic zero point and MLP prediction according to claim 1, characterized in that: In S7, the high-speed range is set to [v th1,min , v th1,max ], set the low speed range to [v th2,min , v th2,max ]; Among them, v th1,min v th1,max v th2,min and v th2,max These represent the minimum value in the high-speed range, the maximum value in the high-speed range, the minimum value in the low-speed range, and the maximum value in the low-speed range, respectively, and v th1,min ≥ v th2,max .

5. The unicycle cornering and steering control method based on dynamic zero point and MLP prediction according to claim 1, characterized in that: In S5, the MLP is trained with training samples before use; the training samples include simulation samples, real vehicle operation samples and abnormal enhancement samples; the abnormal enhancement samples include dynamic zero-point jitter samples, dynamic zero-point lag samples and dynamic zero-point mutation samples.

6. A unicycle cornering and steering control system based on dynamic zero point and MLP prediction, characterized in that: The device includes a controller installed inside the unicycle body; the controller has a built-in memory that stores the unicycle bending and steering control method based on dynamic zero point and MLP prediction as described in any one of claims 1-5.

Images