A port unmanned container truck finite time trajectory tracking control method and system

Abstract

The application discloses a port unmanned truck finite time trajectory tracking control method and system, relates to the fields of port unmanned transportation equipment, intelligent vehicle chassis control, random switching nonlinear system control and finite time trajectory tracking control, and discloses the following technical scheme. The method comprises the following steps: collecting vehicle positioning, speed, heading, wheel speed slip rate, load, road adhesion estimation value, actuator health state and positioning quality; combining confidence smoothing and mode residence to identify random working condition modes; and establishing a mode-related nonlinear vehicle model. The application is suitable for the finite time trajectory tracking control of the port unmanned truck under random working conditions such as low adhesion, heavy load and actuator derating.

Description

Technical Field

[0001] This invention belongs to the technical fields of port unmanned transportation equipment, intelligent vehicle chassis control, random switching nonlinear system control, and finite-time trajectory tracking control, specifically relating to a finite-time trajectory tracking control method and system for port unmanned container trucks. Background Technology

[0002] With the development of automated terminals and smart ports, unmanned container trucks in ports need to continuously perform transportation tasks between quay cranes, yards, gates, inspection areas, and loading / unloading waiting areas. Port unmanned container trucks typically obtain vehicle status through positioning, inertial measurement, wheel speed acquisition, load identification, road surface adhesion estimation, and actuator health monitoring, and output control commands to the drive, braking, and steering actuators through the chassis domain controller.

[0003] In actual port operations, unmanned trucks are characterized by a high proportion of heavy loads, frequent low-speed, large-angle turning conditions, frequent changes in road adhesion, significant actuator thermal decay, and strict safety boundaries. When a vehicle switches from unloaded to heavy load, from dry road surfaces to slippery roads, or from normal actuator states to drive or brake de-rated states, the control effectiveness, damping parameters, and disturbance input channels in vehicle dynamics may all change. If trajectory tracking control is still performed based on single operating condition parameters, problems such as excessive control input, increased lateral deviation of the vehicle, breach of tire adhesion constraints, or the chassis actuators failing to achieve the equivalent control input of the upper layer can easily occur.

[0004] Existing unmanned vehicle trajectory tracking methods mostly use lateral error, heading error, Euclidean distance, fixed quadratic index, or fixed-condition controller to describe tracking performance. While these methods can achieve trajectory tracking under certain fixed conditions, they struggle to differentiate the different risks associated with the same geometric error under conditions of low adhesion, heavy load, or actuator derating. For example, a 0.2-meter lateral deviation under normal adhesion and light load conditions has different impacts on the vehicle's safety boundary than a 0.2-meter lateral deviation under wet and slippery heavy load conditions; fixed Euclidean error cannot reflect this difference.

[0005] Existing random switching control methods mostly remain at the level of offline stability analysis, lacking an engineering closed loop that integrates with vehicle pattern recognition, controller authentication, constraint control allocation, and safety backoff triggering. Even if the theoretical control law satisfies a certain stability condition, it still cannot be directly applied to the control of unmanned container truck chassis in ports if the upper-level equivalent control input cannot be realized under driving, braking, steering, and tire adhesion constraints.

[0006] Furthermore, the controllers and metrics obtained through offline certification typically require lookup calls via time grids, state grids, or operating condition grids in actual chassis domain controllers. Without explanations regarding table interpolation margins, mode dwell, debouncing switching, and input unrealizable determinations, a disconnect can easily arise between theoretical certification results and online execution conditions. Therefore, a technical solution is needed that can seamlessly integrate random operating condition identification, online Riemann path energy certification, finite-time generator dissipative certification, controller-metric recertification, constraint control allocation, and safety backoff triggering. Summary of the Invention

[0007] Technical problems to be solved

[0008] The technical problem this invention aims to solve is that traditional trajectory tracking methods based on Euclidean error or fixed quadratic indices are difficult to distinguish between different risks corresponding to the same geometric error under low adhesion, heavy load, and actuator derating conditions; existing random switching control methods lack error certificates that can be monitored online by the port unmanned truck chassis domain controller and triggered for safe back-off; and existing upper-level trajectory tracking control easily ignores drive, braking, steering, and tire adhesion constraints, resulting in the inability to realize equivalent control inputs.

[0009] Technical solution

[0010] To achieve the above objectives, this invention provides a finite-time trajectory tracking and control method for unmanned container trucks in ports. This method is applied to unmanned container trucks in ports, including a vehicle positioning unit, an inertial measurement unit, a wheel speed acquisition unit, a load identification unit, a road surface adhesion estimation unit, an actuator health monitoring unit, a chassis domain controller, a drive actuator, a braking actuator, and a steering actuator.

[0011] The method uses mode-dependent Riemann path energy as the online trajectory error certificate, finite-time generator dissipation criterion as the authentication condition for the controller-metric pair, and path energy exceeding limits, state exceeding the authentication region, insufficient authentication margin, or unrealizable equivalent control input as safety backoff triggers. The method includes state acquisition and pattern recognition, mode-dependent vehicle modeling, reference state and feedforward input generation, Riemann metric and path energy calculation, trajectory tracking controller construction, finite-time generator dissipation authentication, phased controller-metric re-authentication, constraint control allocation, and path energy monitoring and safety backoff control.

[0012] Specifically, the chassis domain controller collects vehicle positioning, speed, heading, wheel slip ratio, load, road adhesion estimate, actuator health status, and positioning quality. It then identifies the current stochastic operating condition mode based on threshold rules, confidence rules, and mode dwell rules. Next, under this current stochastic operating condition mode, the planar motion process of the unmanned container truck in the port is established as a mode-dependent nonlinear vehicle model. Based on the load estimate, road adhesion estimate, and actuator health status under the current mode, the vehicle drift term, control effectiveness matrix, and disturbance input matrix are determined.

[0013] After obtaining the current stochastic operating condition and vehicle model, the reference trajectory generation module generates a synchronous reference vehicle state based on the port area operation path, speed limit constraints, curvature constraints, and acceleration constraints, and generates reference feedforward control inputs based on the vehicle dynamics compensation relationship under the current stochastic operating condition. The Riemann path energy calculation module constructs a mode-dependent Riemann contraction metric for each stochastic operating condition and calculates the straight-line path energy and path energy verification margin based on the straight-line virtual path between the reference vehicle state and the actual vehicle state.

[0014] Within a preset local state-time authentication region, the controller-metric authentication module authenticates the controller-metric pair consisting of the trajectory tracking controller and the Riemann contraction metric using a finite-time generator dissipation criterion that simultaneously includes a Markov jump metric, a disturbance supply rate metric, and a controlled output metric. Upon successful authentication, a control gain table, a metric table, a safety threshold table, and an authentication margin table are generated for online lookup by the chassis domain controller.

[0015] The constraint control allocation module converts the certified equivalent control input into driving force, braking force, and steering angle commands, and determines whether the equivalent control input is feasible under driving constraints, braking constraints, steering constraints, steering rate constraints, and tire adhesion constraints. The safety backoff module triggers speed degradation, control rescheduling, input limiting, or safe stopping when path energy exceeds limits, certification margin is insufficient, vehicle status exceeds the certification area, low adhesion abrupt change is detected, heavy load anomaly occurs, actuator derating occurs, positioning anomaly occurs, communication anomaly occurs, or the equivalent control input is unavailable.

[0016] This invention also provides a finite-time trajectory tracking and control system for unmanned container trucks in ports, including a pattern recognition module, a reference trajectory generation module, a Riemann path energy calculation module, a controller-metric authentication module, an online control module, a constraint control allocation module, and a safety backoff module. Each module can be implemented as a software functional unit within the chassis domain controller, or it can be implemented collaboratively by the chassis domain controller and the onboard computing platform.

[0017] The present invention also provides an unmanned container truck for ports, comprising a vehicle body, a vehicle positioning unit, an inertial measurement unit, a wheel speed acquisition unit, a load identification unit, a road surface adhesion estimation unit, an actuator health monitoring unit, a chassis domain controller, a drive execution unit, a braking execution unit, and a steering execution unit; the chassis domain controller is configured with the aforementioned finite-time trajectory tracking and control system for unmanned container trucks based on Riemann contraction metric, or the chassis domain controller is configured to execute the aforementioned method.

[0018] Beneficial effects

[0019] Compared with the prior art, the present invention provides a method and system for finite-time trajectory tracking and control of unmanned container trucks in ports, which has the following beneficial effects:

[0020] 1. This invention converts the error risk of unmanned container trucks in ports under low adhesion, heavy load and actuator derating conditions into path energy certificates that can be monitored online through mode-correlated Riemann path energy. Compared with fixed Euclidean error, it can more accurately reflect the risk differences under different random operating conditions.

[0021] 2. This invention achieves finite-time authentication of controller-metric pairs by using a finite-time generator dissipation criterion containing a Markov jump metric term, a disturbance supply rate term, and a controlled output term. Furthermore, it enables offline authentication results to be naturally transferred to online lookup table execution through authentication margin and interpolation preservation conditions.

[0022] 3. This invention reduces the difficulty of joint solution of controller and metric by implementing a phased controller-metric implementation strategy; it converts the equivalent control input into real drive, braking and steering commands by a constraint control distributor with slack, and uses the lack of feasible solution or slack out of bound as the basis for safe backoff.

[0023] 4. This invention improves the reliability of trajectory tracking and operational safety of unmanned container trucks in complex port environments by triggering graded safety backoff based on path energy exceeding limits, state exceeding the authentication area, insufficient authentication margin, and unrealizable input. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the architecture of the port unmanned container truck limited-time trajectory tracking control system of the present invention.

[0025] Figure 2 This is a flowchart of the finite-time trajectory tracking control method of the present invention.

[0026] Figure 3 This is a schematic diagram of the random operating conditions and mode transition relationships of the present invention.

[0027] Figure 4 This is a schematic diagram of the reference trajectory, actual trajectory, and virtual straight path of the present invention.

[0028] Figure 5 This is a schematic diagram of the time-limited authentication and phased re-authentication process of the present invention.

[0029] Figure 6 This is a schematic diagram of the constraint control allocation and safety rollback process of the present invention.

[0030] Figure 7 This is a schematic diagram of the pattern recognition, anti-shake, and dwell determination process of the present invention.

[0031] Figure 8 This is a schematic diagram illustrating the relationship between offline data table generation and online table lookup in this invention.

[0032] Figure 9 These are position tracking error response diagrams for three controller implementation forms in the algorithm verification semi-numerical embodiment of the present invention. Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the embodiments described are for illustrative purposes and should not be construed as limiting the scope of protection of the present invention. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0034] Example 1: Main System and Overall Control Flow

[0035] The purpose of this embodiment is to provide a verifiable trajectory tracking control method and system based on Riemann contraction metric in the case of random working condition switching, external disturbances, actuator constraints and limited time operation windows of unmanned container trucks in ports. This enables the vehicle to maintain bounded trajectory error within a preset local state-time verification area and to perform safe back-off when the path energy exceeds the limit or the equivalent control input is unavailable.

[0036] Reference Appendix Figure 1 The unmanned container truck of the present invention includes a vehicle positioning unit, an inertial measurement unit, a wheel speed acquisition unit, a load identification unit, a road surface adhesion estimation unit, an actuator health monitoring unit, a chassis domain controller, a drive actuator unit, a braking actuator unit, and a steering actuator unit. The vehicle positioning unit can employ one or more of satellite positioning, inertial combined positioning, roadside assisted positioning, or visual positioning; the inertial measurement unit outputs vehicle acceleration, angular velocity, and heading information; the wheel speed acquisition unit outputs the wheel speed of each wheel and calculates the wheel slip ratio; the load identification unit estimates the vehicle load based on suspension height, axle load sensor data, or loading operation information; the road surface adhesion estimation unit estimates road surface adhesion based on wheel speed, acceleration, braking pressure, or drive torque; and the actuator health monitoring unit monitors drive motor temperature, braking pressure, steering actuator status, and actuator derating information.

[0037] The chassis domain controller is configured with a pattern recognition module, a reference trajectory generation module, a Riemann path energy calculation module, a controller-metric authentication module, an online control module, a constraint control allocation module, and a safety backoff module. The pattern recognition module receives vehicle sensor information and outputs the current random operating condition mode; the reference trajectory generation module generates a reference vehicle state and a reference feedforward control input; the Riemann path energy calculation module calculates the straight-line path energy and authentication margin; the controller-metric authentication module generates a control gain table, a Riemann metric table, and a safety threshold table during offline or quasi-online phases; the online control module looks up the tables to generate equivalent control inputs; the constraint control allocation module converts the equivalent control inputs into driving force, braking force, and steering angle; and the safety backoff module triggers a backoff action based on energy, state, actuator, and communication status.

[0038] Reference Appendix Figure 2 The method of this invention is generally divided into an offline authentication phase and an online execution phase. The offline authentication phase mainly completes mode-dependent vehicle model calibration, controller candidate gain generation, Riemann metric reconstruction, finite-time generator dissipation criterion verification, and safety threshold table generation. The online execution phase mainly completes state acquisition, pattern recognition, reference trajectory generation, control gain lookup, equivalent control input generation, constraint control allocation, path energy monitoring, and safety backoff. The advantage of this approach is that it eliminates the need to repeatedly solve matrix inequalities or high-dimensional nonlinear optimization problems during online operation. The chassis domain controller only needs to perform table lookup, interpolation, path energy integration, and low-dimensional constraint allocation, making the computational load more suitable for the vehicle-mounted controller.

[0039] Example 2: Random Operating Condition Pattern Recognition, Mode Transfer, and Anti-shake Processing

[0040] In port scenarios, the same unmanned container truck may exhibit significant differences at different stages of operation. For example, when driving towards a quay crane unloaded, the vehicle responds quickly; when driving towards a storage yard fully loaded, the braking distance increases; rainy weather or watered sections reduce road surface adhesion; prolonged low-speed heavy-load operation may lead to derating of the drive motor or braking system. To avoid using a single model to describe all operating conditions, this embodiment divides the vehicle's operating state into multiple random operating condition modes.

[0041] The set of random operating conditions is denoted as:

[0042] ;

[0043] in, It is a set of random operating conditions. This represents the number of random operating condition modes. In this embodiment, it can be taken as... Mode 1 is the normal adhesion light load mode, Mode 2 is the low adhesion wet skid mode, Mode 3 is the heavy load low speed mode, Mode 4 is the actuator efficiency reduction mode, and Mode 5 is the lateral disturbance enhancement mode. The number and names of the above modes can be adjusted according to the specific port area vehicles and road conditions.

[0044] To facilitate engineering implementation, the pattern recognition module first generates the evidence quantity for each pattern based on threshold rules, and then obtains the pattern confidence based on confidence and smoothing rules. For the... In this model, the confidence update can be represented as:

[0045] ;

[0046] Where t represents the time variable, and i and j represent the random operating condition indexes. For the first The confidence level of this pattern at the current moment. To control the sampling period, For the first The amount of evidence for each pattern This is the confidence decay coefficient. This indicates that the value is limited to between 0 and 1. The purpose of this formula is to avoid a single sensor fluctuation directly causing a mode switch; its technical effect is to make the pattern recognition results smoother and reduce the frequent switching of control gain and measurement.

[0047] In low-adhesion wet skid pattern recognition, when the road surface adhesion estimate Less than the low adhesion threshold And wheel speed slip ratio Greater than the slip threshold At that time, the amount of evidence for the low-adhesion wet slip mode will be increased. Set to a larger value. This is an estimate of road surface adhesion. For low adhesion threshold, Wheel speed slip ratio, This is the slip threshold. By using both the road adhesion estimate and the wheel speed slip ratio simultaneously, the probability of misjudgment due to error from a single sensor can be reduced.

[0048] In heavy-load low-speed mode recognition, when the load estimate is... Greater than the overload threshold Furthermore, when the vehicle speed is below the low-speed threshold, the confidence level of the heavy-load low-speed mode is increased. This is the estimated load value. This is the heavy load threshold. The heavy load low speed mode corresponds to greater rolling resistance and less available acceleration, therefore subsequent reference trajectory generation and control allocation need to be more conservative.

[0049] In actuator performance degradation mode recognition, the confidence level of the actuator performance degradation mode is increased when the drive motor temperature exceeds a temperature threshold, the braking pressure is lower than the preset response requirement, or the steering actuator feedback delay exceeds a preset threshold. This mode does not represent a single actuator failure, but rather an overall decrease in the vehicle chassis's actuator capability. By incorporating actuator performance degradation into random operating condition modes, the trajectory tracking controller can promptly reduce the intensity of control commands when actuators are derated.

[0050] To avoid controller jitter caused by frequent mode switching, this embodiment employs a mode dwell rule. Let the current mode be... Candidate mode is Only when the confidence level of a candidate pattern continuously exceeds the preset difference in confidence level of the current pattern... and continue to reach the required length of stay. Only when the current random operating mode is switched to the candidate mode will the current random operating mode be switched. The confidence difference threshold. This refers to dwell time. This process reduces repeated switching near low-adhesion boundaries, light-heavy-load boundaries, and actuator derating boundaries.

[0051] Reference Appendix Figure 3 The different stochastic operating conditions are described by the mode transition rate matrix. The mode transition rate matrix is ​​as follows:

[0052] ;

[0053] in, The mode transition rate matrix, For vehicles by the This mode is directed towards the first The instantaneous transition rate of each mode transition. To facilitate time-limited authentication, this embodiment also sets an upper bound matrix for the mode transition rate:

[0054] ;

[0055] in, This is the upper bound matrix for mode transition rates. for The upper bound of the mode transition rate obtained through online estimation satisfies:

[0056] ;

[0057] The technical significance of this relation lies in the fact that the finite-time generator dissipation criterion can be used... Conservative certification is implemented so that even if the online mode transfer rate changes with the port environment, as long as it does not exceed the upper limit, the certified controller-metric pair will not be compromised.

[0058] Example 3: Generation of Mode-Related Vehicle Models and Reference Feedforward

[0059] In step S2, if a planar equivalent chassis model is used, the vehicle state is defined as follows:

[0060] ;

[0061] in, This represents the current vehicle status. and For the vehicle's planar position, and Let the vehicle's planar velocity be... This represents a vector composed of columns. The equivalent control input is defined as:

[0062] ;

[0063] in, This is the equivalent longitudinal and lateral control input. This is the longitudinal equivalent control component. This refers to the lateral equivalent control component. The equivalent control input here is not the drive motor torque, braking pressure, or steering angle, but rather the equivalent acceleration control quantity that the upper-level trajectory tracking controller wants the chassis to generate. This is then converted into actual actuator commands by the constraint control distributor.

[0064] Current random operating mode The vehicle dynamics below are written as:

[0065] ;

[0066] in, Let be the derivative of the vehicle state with respect to time. For the first Nonlinear drift term in this mode For the first Control effectiveness matrix under various modes For the first The perturbation input matrix under this mode, This is an external disturbance. It can represent disturbances such as lateral wind disturbance, slope disturbance, tire slip disturbance, positioning error injection, or actuator delay.

[0067] Under the four-dimensional state model, the nonlinear drift term and the control effectiveness matrix can be taken as follows:

[0068] ;

[0069] ;

[0070] in, For the first Damping coefficient under this mode For the first The control effectiveness coefficient under this mode For the velocity norm, To prevent the velocity norm from being a singular positive constant. Damping coefficient. Determined by load, rolling resistance, road conditions, and speed range; control effectiveness coefficient It is determined by the estimated road surface adhesion, load, and actuator health status. When road surface adhesion decreases, load increases, or the actuator is derated... Reduce; when rolling resistance, gradient resistance, or low-speed heavy-load resistance increases. Increase.

[0071] The model described above is not intended to precisely describe all tire details, but rather to provide a certifiable, online-calculated pattern-dependent model for finite-time trajectory tracking control. For vehicles requiring higher precision, the vehicle state can be extended to:

[0072] ;

[0073] in, For the vehicle's heading angle, This represents the yaw rate. The extended state model can still utilize the pattern recognition, Riemann path energy calculation, and finite-time authentication framework of this invention.

[0074] In step S3, the reference vehicle state is generated jointly by the port area operation path, speed limit constraints, curvature constraints, and acceleration constraints. For the four-dimensional plane state, the reference vehicle state is:

[0075] ;

[0076] in, For reference vehicle condition, and For reference position, and The reference speed is used as a reference. The reference trajectory generation module adjusts the reference speed and curvature according to the current mode. For example, in low-adhesion wet skid mode, the maximum reference speed and maximum lateral acceleration are reduced; in heavy-load low-speed mode, the rate of change of acceleration is limited; and in actuator efficiency degradation mode, the amplitude of the feedforward control input is limited.

[0077] To match the reference state with the vehicle dynamics of the current mode, the reference feedforward control input is calculated according to the mode-synchronized reference dynamics:

[0078] ;

[0079] in, For reference vehicle state derivative, For the first The reference feedforward control input for this mode. For the four-dimensional plane state, the reference feedforward control input can be written as:

[0080] ;

[0081] in, As the reference velocity vector, and This is the reference acceleration corresponding to the reference path in the port area at the current moment. Here is the reference velocity norm. The motivation for introducing this formula is that if only the geometric path is used without considering the damping and control effectiveness in the current mode, the feedforward input may be too small or too large; through mode synchronization reference dynamics compensation, the reference state can be made to naturally satisfy the vehicle model in the current mode under undisturbed conditions.

[0082] when Below the control effectiveness threshold This indicates that the chassis's responsiveness to upper-level control inputs is insufficient. In this case, the reference trajectory generation module does not directly use the original reference speed, but instead performs reference speed degradation or curvature reduction, and regenerates the reference vehicle state and reference feedforward input. This avoids generating feedforward control inputs that are impossible to achieve under actuator derating or low adhesion conditions.

[0083] Example 4: Riemann Path Energy and Online Error Certificate

[0084] Reference Appendix Figure 4 In order to convert the error between the actual vehicle state and the reference vehicle state into an error certificate that can be monitored online, this embodiment constructs a straight virtual path between the two:

[0085] ;

[0086] in, This represents the state along a straight virtual path. For path parameters. When hour, Equal to the reference vehicle status; when hour, This is equal to the actual vehicle state. Let the tracking error be:

[0087] ;

[0088] in, This represents the tracking error between the actual vehicle state and the reference vehicle state. The straight-path energy is defined as:

[0089] ;

[0090] in, The energy of the straight path at the current moment. For the current random operating mode The corresponding Riemann contraction metric This indicates transpose. The technical scenario for this formula is that unmanned container trucks in ports face different risks for the same geometric error under different operating conditions. The purpose of introducing the formula is to use mode correlation metrics to weight the error direction and magnitude. Its technical effect is to transform the tracking error into an online error certificate that can change with operating conditions.

[0091] In actual deployment, the chassis domain controller does not need to calculate continuous integrals, but approximates them through numerical integration:

[0092] ;

[0093] in, The number of integral sampling points and , For the first One integral sampling point, For the first The integral weights are as follows. For chassis domain controllers with limited computing resources, a 3-point integral can be used first; under high-risk conditions such as low adhesion, heavy load, or actuator derating, a 5-point or 7-point integral can be used to improve the accuracy of path energy calculation.

[0094] The Riemann contraction metric in this embodiment can take the following form:

[0095] ;

[0096] in, For the first Riemann contraction metric under this model As a model-dependent time-varying benchmark metric, This is a correction factor for the position error measurement. For positional error, For actual location, For reference position, It is a positive semi-definite matrix that is weighted only for the position error components. This represents the Euclidean L2 norm. Since the safety risks of unmanned container trucks in ports are mainly related to position and speed deviations, the greater the position deviation, the more the chassis domain controller needs to increase the risk weight in the corresponding error direction. Therefore, this metric is adopted.

[0097] To ensure that path energy can be used as a valid certificate, the Riemann metric must satisfy positive definite upper and lower bounds:

[0098] ;

[0099] in, This is the positive definite lower bound of the Riemannian metric. This is the positive definite upper bound of the Riemannian metric. Let be the identity matrix. The technical effect of this relation is: when When the error is small, the actual tracking error is also limited; when When approaching the upper limit, the system can take a rollback action in advance.

[0100] Path energy certification margin is defined as:

[0101] ;

[0102] in, For path energy certification margin, For a finite-time path energy boundary. When When the value is less than the warning threshold but still greater than zero, the safety rollback module performs a first-level rollback, such as reducing the reference speed; when When the value is less than zero, the path energy is determined to be out of bounds, and control rescheduling, input limit, or safe stop are executed.

[0103] Example 5: Tracking Controller and Finite-Time Generator Dissipative Authentication

[0104] In step S5, the mode-dependent trajectory tracking controller consists of a reference feedforward control input and a feedback correction input:

[0105] ;

[0106] in, For equivalent control input, For the first Reference feedforward control input in this mode For the first Feedback correction input in this mode. The feedback correction input satisfies:

[0107] ;

[0108] The technical function of this condition is that when the actual vehicle state is consistent with the reference vehicle state, the controller does not introduce additional feedback correction, thus avoiding unnecessary shaking of the vehicle near the reference trajectory.

[0109] In one implementation, the feedback correction input takes the form of a state-time affine gain:

[0110] ;

[0111] in, For the first Time-varying reference gain in this mode For the first In the second mode and the third Affine gain related to each error component For tracking error The One portion, Let represent the vehicle state dimension. The technical scenario for this formula is that the online computing power of the chassis domain controller is limited, but the vehicle error state changes. The reason for introducing this formula is to enable the feedback gain to change with the error state and time without excessively increasing the online computing load. Its technical effect is that it has better tracking and disturbance response capabilities compared to constant gain.

[0112] In step S6, the controller-metric authentication module performs finite-time generator dissipation criterion verification within a preset local state-time authentication region. The preset local state-time authentication region can be:

[0113] ;

[0114] in, For preset local state-time authentication region, For the length of the finite-time control interval, For positional error, For speed error, As the position error boundary, Let v(t) be the velocity error boundary, where v(t) represents the actual velocity vector composed of the vehicle's planar velocity components; the superscript 2 indicates the Cartesian product of the two-dimensional components.

[0115] The closed-loop vector field is defined as:

[0116] ;

[0117] in, For the first The closed-loop vector field under this mode. The closed-loop Jacobian matrix is ​​defined as:

[0118] ;

[0119] in, For the first The closed-loop Jacobian matrix under each mode. The controlled output Jacobian matrix is ​​defined as:

[0120] ;

[0121] in, For the first The controlled output Jacobian matrix in this mode It is a controlled output function.

[0122] When the Riemann contraction metric is related to the vehicle state, the total derivative of the metric along the closed-loop vector field is:

[0123] ;

[0124] in, for The total derivative along the closed-loop vector field, Vehicle status The One portion, The first closed-loop vector field Each component.

[0125] The finite-time generator dissipation criterion includes the following matrix inequalities:

[0126] ;

[0127] in, For the first Closed-loop dissipation term in this mode, This is a metric generator term used for metric generation caused by conservative substitution mode transitions. The perturbation input matrix is... The level of disturbance attenuation. Let be the identity matrix, and the lower triangular block be the transpose of the upper triangular block. The technical scenario for the above matrix inequality is the operation of unmanned container trucks in ports under random operating condition switching and external disturbances; the purpose of introducing the formula is to simultaneously limit the growth of closed-loop error, the metric changes caused by mode jumps, and the impact of external disturbances; its technical effect is to provide verifiable conditions for the energy bound of finite-time paths.

[0128] in, satisfy:

[0129] ;

[0130] For mode transition conservative terms, This is the upper bound of the mode transition rate. For the first The Riemann contraction metric under these modes. This relationship enables certification results to cover the switching between different stochastic operating conditions.

[0131] The closed-loop dissipation term can be written as:

[0132] ;

[0133] in, , For the first Energy growth rate parameters under this model. Used to incorporate controlled output constraints into the dissipation criterion, avoiding the neglect of safety-related outputs while only controlling state errors. X represents any matrix whose dimension matches the corresponding matrix operation.

[0134] The energy bound of a finite-time path satisfies:

[0135] ;

[0136] in, For the initial path energy boundary, For a finite-time path energy boundary, This is the upper bound parameter for the energy growth rate. This represents the upper bound of the perturbation energy. This formula indicates that, as long as the initial path energy and perturbation energy are limited to the certified range, the path energy will not exceed [a certain value] within a finite time interval. In the exponential term, e is the natural constant (the base of the exponential function), which is different from the aforementioned tracking error e(t).

[0137] Example 6: Offline Authentication, Online Table Lookup, and Interpolation Preservation

[0138] Reference Appendix Figure 5 and attached Figure 8 The phased controller-metric implementation strategy comprises three phases. The first phase generates candidate controllers under preset metrics. Candidate controllers can include constant gain, time-varying gain, and state-time affine gain. The second phase reconstructs a mode-dependent Riemannian metric compatible with the selected controller after fixing it. The third phase recertifies the fixed controller and the reconstructed metric based on positive definiteness, dissipation, interpolation preservation, and path energy bounds.

[0139] Interpolation preservation is to ensure that offline authentication results can be securely used for online table lookups. Specifically, offline authentication not only verifies the semi-negative definite dissipation matrix at the grid nodes, but also sets authentication margins between adjacent time grids. Let the th... The time grid points are , No. The time grid points are The control gain obtained by interpolation within this interval is:

[0140] ;

[0141] in, The interpolated first Various modes control gain, and Control the gain for adjacent grid points. These are the interpolation coefficients. The authentication module reserves margin for the interpolated controller-metric pair. When the authentication margin corresponding to the interpolation point is lower than the online runtime... The chassis domain controller does not use this interpolation controller, but instead switches to a conservative gain table or triggers speed degradation.

[0142] After successful offline authentication, the controller-metric authentication module outputs the following tables: a control gain table, storing control gains under different modes, times, and error partitions; a metric table, storing Riemann contraction metric parameters under different modes and times; a safety threshold table, storing finite-time path energy limits, warning thresholds, authentication zone boundaries, and relaxation thresholds; and an authentication margin table, recording the minimum authentication margin for each grid point or partition. During online operation, the chassis domain controller indexes these tables according to the current mode, current time, and tracking error, avoiding repeated complex authentication processes at the vehicle end.

[0143] If re-authentication fails, the system does not directly output the control table. Instead, it takes action based on the cause of the failure: if positive definiteness fails, the Riemann metric parameters are adjusted or the authentication region is narrowed; if dissipative failure fails, the reference velocity is reduced, damping compensation is increased, or the controller parameters are searched again; if the path energy boundary fails, the initial energy boundary, disturbance energy boundary, or finite-time control interval is reset. Through this process, a closed loop is formed between the offline authentication results and the online execution conditions, and the technical solution disclosed in the manual no longer remains at the level of abstract algorithms.

[0144] Example 7: Constraint Control Assignment and Input Unrealizable Determination

[0145] Reference Appendix Figure 6 In step S8, the upper-level trajectory tracking controller outputs the equivalent control input. The chassis's actual actuators can receive and execute driving force, braking force, and steering angle. Therefore, a constraint control distributor is needed to... Convert to chassis execution commands.

[0146] The actuator input vector is defined as:

[0147] ;

[0148] in, For the chassis actuator input vector, As the driving force, For braking force, Let be the front wheel steering angle. The mapping from the vehicle chassis actuator input to the equivalent control input is denoted as . The constraint control allocation optimization problem with slack is as follows:

[0149] ;

[0150] in, For optimal actuator input, For relaxation amount, For the equivalent input tracking error weight matrix, For the executor instruction smoothing weight matrix, For smoothing weighting coefficients, This represents the relaxation penalty coefficient. The reason for introducing relaxation is that when approaching the adhesion limit, actuator derating, or steering speed limitations, the constraint problem may not be fully satisfied. The relaxation amount can help the system determine the "degree of deviation from the upper control intention". In the formula, the norm with subscript W_u or W_s represents the weighted 2-norm squared defined by the corresponding weight matrix.

[0151] The optimization problem satisfies the following constraints:

[0152] ;

[0153] ;

[0154] ;

[0155] in, As the upper limit of driving force, As the upper limit of braking force, This is the upper limit of the front wheel steering angle. This is the upper limit of the turning rate. and These are the longitudinal and lateral forces acting on the tire, respectively. This is an estimate of road surface adhesion. The normal load is denoted by . This adhesion constraint is used to prevent the control distribution result from exceeding the tire's available adhesion capacity. In the equation, the first derivative of the steering angle in the constraint equation represents the rate of change of the front wheel steering angle.

[0156] When the above optimization problem has a feasible solution and the relaxation factor satisfies When the constraint control distributor outputs driving force, braking force, and front wheel steering angle; when the optimization problem has no feasible solution or When this occurs, it is determined that the equivalent control input is not feasible. This is the slack threshold. The result of this determination is directly input into the safety rollback module to prevent the vehicle from continuing to operate under control commands that the chassis cannot fulfill.

[0157] Example 8: Safe rollback, false alarm handling, and handling of abnormal scenarios

[0158] In step S9, the safety backoff module monitors online the straight path energy, path energy authentication margin, vehicle state constraints, actuator constraints, positioning quality, communication status, and the feasibility of equivalent control inputs. When the monitored quantities meet the trigger conditions, the system does not immediately take a single fixed action, but instead performs a graded backoff based on the degree of risk.

[0159] Level 1 rollback is a downgrade to the reference speed. Triggering conditions include: The following conditions are considered low-adhesion modes: below the warning threshold but still greater than zero; vehicle status approaching the certification zone boundary; low-adhesion mode confidence increasing but not yet stably switching. The first-level rollback logic involves reducing the reference speed, reducing the reference acceleration, and limiting the rate of curvature change. Its technical function is to reduce the energy growth trend of the vehicle in advance while it is still within the certification range.

[0160] Level 2 rollback involves control rescheduling and input limiting. Triggering conditions include: path energy approaching or briefly exceeding the finite-time path energy limit, mode switching occurring, control slack approaching the threshold, and actuator experiencing slight derating. The processing logic of Level 2 rollback is to switch to the conservative control gain table of the corresponding stochastic operating mode, while simultaneously limiting the rate of change of driving force, braking force, and steering angle. Its technical purpose is to prevent further error expansion and maintain vehicle controllability.

[0161] Level 3 reversal is for emergency braking or safe stopping. Triggering conditions include: positioning anomaly, communication anomaly, braking anomaly, steering anomaly, obstacle intrusion into the safety envelope, infeasible constraint control allocation, or relaxation exceeding a threshold. The processing logic of Level 3 reversal is to execute emergency braking, safe stopping, or maintain a minimum risk state according to the port area safety strategy. Its technical function is to combine theoretical trajectory tracking control with the safety operation requirements of unmanned container trucks in the port.

[0162] To reduce the impact of false alarms, this embodiment also includes anomaly confirmation logic. For sensor-related anomalies such as sudden changes in low adhesion, heavy load anomalies, and decreased positioning quality, the safety backoff module can employ a strategy of continuous multi-sampling cycle confirmation, sensor cross-validation, and confidence maintenance. For example, when the positioning quality deteriorates in a single instance but inertial measurement and wheel speed estimation can still maintain a short-term state estimate, the system preferentially enters level one or level two backoff instead of immediately stopping safely; when the positioning quality continuously falls below a threshold and the communication status is abnormal, the system enters level three backoff. This ensures safety while reducing unnecessary shutdowns caused by short-term sensor fluctuations.

[0163] Example 9: Data Table Storage, Device Implementation, and Data Security

[0164] In this embodiment, the control gain table, Riemann metric table, safety threshold table, and certification margin table can be stored in the non-volatile memory of the chassis domain controller. Each table contains a mode number, time grid, error partition, parameter value, and version number. After the vehicle is powered on, the chassis domain controller first checks the table version number and integrity check value. Only after confirming that the table matches the current vehicle parameter version, tire specifications, actuator parameters, and port operation strategy is it allowed to enter the automatic trajectory tracking mode.

[0165] To ensure real-time table access, the online control module can cache commonly used control gain and metric tables for the current mode in high-speed storage. During mode switching, the system preloads the conservative gain table for the candidate mode to avoid lookup delays during the mode transition. For online interpolation results, the system also checks the certification margin table. If the interpolation position is outside the certification coverage or the certification margin is insufficient, the control input is not output; instead, speed degradation or a conservative controller is triggered.

[0166] Regarding data security, this invention does not require uploading the vehicle's original positioning trajectory or load data to the cloud to complete chassis trajectory tracking control. The offline authentication table can be distributed to the chassis domain controller by the fleet management system or maintenance equipment. The table distribution can include a version number, checksum, and applicable vehicle parameter range. The chassis domain controller only needs to perform pattern recognition, table lookup control, path energy calculation, and constraint control allocation locally, thereby reducing the impact of communication anomalies on chassis safety control.

[0167] In terms of device implementation, the chassis domain controller may include a processor, a memory, an input interface, an output interface, and a vehicle communication interface. The processor is used to execute pattern recognition, reference trajectory generation, path energy calculation, online control, constraint control allocation, and safety rollback procedures. The memory stores control gain tables, Riemann metric tables, safety threshold tables, certification margin tables, and program instructions. The input interface receives positioning, inertial measurement, wheel speed, load, adhesion estimation, and actuator health status. The output interface sends instructions to the drive, braking, and steering actuators. The vehicle communication interface exchanges task paths and operating status with the fleet management system or port dispatching system.

[0168] Example 10: Semi-numerical Example for Algorithm Verification

[0169] In a semi-numerical embodiment for algorithm verification, a two-dimensional equivalent motion model is used to verify the feasibility of offline authentication, phased controller-metric re-authentication processes, and online path energy monitoring, constraint control allocation, and safety backoff processes. The finite-time control interval is taken as... Control the sampling period Monte Carlo run count In the formula, s after the numerical value is the time unit in seconds.

[0170] Preset local state - time authentication area:

[0171] ;

[0172] in, For the number of Monte Carlo runs. The candidate initial error satisfies During online operation As a finite-time path energy boundary condition.

[0173] During the offline authentication phase, the reconstructed pattern-related Riemannian metric and dissipation matrix are... Sampling verification was performed within the model. In the example, multiple sampling points were used for verification of each mode. The reconstructed Riemannian metric remained positive definite, the dissipation matrix remained semi-negative definite, and no authentication violations were observed. This result demonstrates that the controller-metric pair can form a callable finite-time path energy certificate within a preset local state-time authentication region.

[0174] In the controller comparison embodiments, three types of controller implementations are compared: constant gain, time-varying gain, and state-time affine gain. Constant gain is simple to implement and suitable for chassis domain controllers with limited computing power; time-varying gain can vary over a finite time interval; and state-time affine gain can further adjust the control strength based on the error state. All three types of controllers are allowed to be used online only after passing the aforementioned finite-time certification.

[0175] Reference Appendix Figure 9 The horizontal axis represents time in seconds, and the vertical axis represents position tracking error. The figure shows the position tracking error responses under nominal and disturbed operating conditions for three controller implementations: constant gain, time-varying gain, and state-time affine gain. It can be seen that all three controller implementations can reduce the position tracking error over time and maintain it within a small range at the end of a finite time interval; among them, the state-time affine gain exhibits better overall performance in terms of end-point error and disturbance response.

[0176] The above verification is used to illustrate the relative performance differences of different controller implementations under finite-time path energy constraints, and does not limit the present invention to using the specific simulation model, specific parameters, or specific values ​​described above. During the online execution phase, the chassis domain controller obtains the equivalent control input by looking up a table based on the current random operating mode, current time, and tracking error, and calculates the straight-line path energy in real time. When the straight-line path energy does not exceed the finite-time path energy limit and the constraint control allocator has a feasible solution, the vehicle continues to perform trajectory tracking; when the straight-line path energy exceeds the limit, the vehicle state exceeds the certification area, or the constraint control allocator has no feasible solution, the vehicle performs speed degradation, control rescheduling, input limiting, or safe stopping.

[0177] In summary, this invention does not simply apply Riemann contraction metric to ordinary vehicle tracking control. Instead, it combines the low adhesion, heavy load, actuator derating, mode switching, and safe backoff requirements of unmanned container trucks in ports with a closed-loop technical solution that can be executed by the chassis domain controller. This solution integrates pattern recognition, finite-time authentication, online path energy certificate, constraint control allocation, and tiered safe backoff. This solution can improve the reliability and safety of trajectory tracking for unmanned container trucks in complex port environments.

Claims

1. A finite-time trajectory tracking and control method for unmanned container trucks in ports, characterized in that, The Riemann path energy related to the random operating mode is used as the online trajectory error certificate, the finite-time generator dissipation criterion is used as the offline authentication condition for the controller-metric pair, and path energy exceeding the limit, insufficient authentication margin, state exceeding the authentication area, or equivalent control input being unrealizable are used as safety backoff triggers. The process includes the following steps: S1. Collect vehicle status, load, road surface adhesion and actuator health information of unmanned container trucks in the port, and identify the current random working condition mode; S2. Establish a nonlinear vehicle model corresponding to the current stochastic operating condition, and determine the vehicle drift term, control effectiveness matrix, and disturbance input matrix; S3. Generate reference vehicle status and reference feedforward control input that are synchronized with the current random working condition mode based on the port area operation path and vehicle constraints. S4. Construct mode-related Riemann contraction metric, and calculate straight path energy and its certification margin based on the straight virtual path between the reference vehicle state and the actual vehicle state; S5. Construct a mode-dependent trajectory tracking controller that includes a reference feedforward control input and a feedback correction input; S6. Within a preset local state-time authentication region, the controller-metric pair consisting of the trajectory tracking controller and the Riemann contraction metric is authenticated using the finite-time generator dissipation criterion. S7. Generate a control gain table, a measurement table, and a safety threshold table based on the certification results. When the chassis domain controller is running online, it obtains the equivalent control input based on the current random operating mode, the current time, and the tracking error. S8. The equivalent control input is converted into driving force, braking force and steering angle commands through the constraint control distributor, and it is determined whether the equivalent control input is feasible. S9. When the path energy exceeds the limit, the authentication margin is insufficient, the vehicle status exceeds the authentication area, or the equivalent control input is not available, trigger speed degradation, control rescheduling, input limiting, or safe stopping.

2. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The random operating condition modes include at least two of the following: normal adhesion light load mode, low adhesion wet skid mode, heavy load low speed mode, actuator performance degradation mode, and lateral disturbance enhancement mode. When the road surface adhesion estimate is lower than the low adhesion threshold and the wheel speed slip ratio is higher than the slip threshold, the confidence level of the low adhesion wet skid mode is increased. When the load estimate is higher than the heavy load threshold, the confidence level of the heavy load low speed mode is increased. When the drive motor temperature, braking pressure, or steering actuator status meets the derating condition, the confidence level of the actuator performance degradation mode is increased. The confidence levels of each mode are normalized and smoothed, and after the confidence level of a candidate mode continuously exceeds the preset difference of the confidence level of the current mode and the duration reaches the dwell time, the candidate mode is taken as the current random operating condition mode.

3. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The vehicle state includes the vehicle's planar position and vehicle's planar velocity, or further includes the vehicle's heading angle and yaw rate; the control effectiveness coefficient in the mode-dependent nonlinear vehicle model is jointly determined by the road adhesion estimate, load estimate, and actuator health state, and the damping coefficient is jointly determined by the load, rolling resistance, road state, and speed range.

4. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The reference feedforward control input is calculated by the mode synchronization reference dynamics; when the control effectiveness corresponding to the random operating mode is lower than the control effectiveness threshold, the reference trajectory generation module reduces the reference speed or reduces the curvature command and then regenerates the reference vehicle state and the reference feedforward control input.

5. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The straight path energy is obtained by numerically integrating the mode-dependent Riemann contraction metric on a straight virtual path between the reference vehicle state and the actual vehicle state; the mode-dependent Riemann contraction metric includes a mode-dependent time-varying reference metric and a position error metric correction term, and the position error metric correction term only weights the position error component.

6. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The feedback correction input adopts a state-time affine gain form, which is generated by an offline finite-time certification process and stored in the form of a time grid table. During online operation, the chassis domain controller interpolates based on the current random operating mode, the current time, and the tracking error. When the certification margin of the interpolation point is lower than the lower limit of the certification margin, it switches to a conservative gain table or triggers speed degradation.

7. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The finite-time generator dissipation criterion includes a matrix term for measuring generator terms caused by conservative alternative mode transitions, a disturbance supply rate term for limiting the influence of external disturbances, and a controlled output term for constraining controlled outputs. The controller-metric pair, after authentication, generates a control gain table, a metric table, a security threshold table, and an authentication margin table.

8. The finite-time trajectory tracking and control method for unmanned container trucks in ports according to claim 1, characterized in that, The constraint control distributor converts the equivalent control input into driving force, braking force, and steering angle through an optimization problem with relaxation. The optimization problem simultaneously satisfies driving force limit, braking force limit, steering angle limit, steering rate limit, and tire adhesion constraint. When the optimization problem has no feasible solution or the relaxation exceeds the relaxation threshold, the equivalent control input is determined to be unrealizable.

9. A finite-time trajectory tracking control system for unmanned container trucks in ports, characterized in that, The system includes a pattern recognition module, a reference trajectory generation module, a Riemann path energy calculation module, a controller-metric authentication module, an online control module, a constraint control allocation module, and a safety backoff module. The pattern recognition module outputs the current random operating condition pattern based on vehicle positioning, speed, heading, wheel slip ratio, load, road surface adhesion estimate, actuator health status, and positioning quality. The reference trajectory generation module generates a reference vehicle state and a reference feedforward control input. The Riemann path energy calculation module calculates the straight path energy and path energy authentication margin. The controller-metric authentication module is used to perform finite-time generator dissipation criterion verification, controller-metric re-authentication, and generate control gain table, metric table and security threshold table; The online control module is used to generate equivalent control inputs by looking up a table based on the current random operating mode, current time, and tracking error; the constraint control allocation module is used to convert the equivalent control inputs into driving force, braking force, and steering angle commands; the safety back-off module is used to trigger speed degradation, control rescheduling, input limiting, or safe stopping when the energy exceeds the limit on the straight path, the vehicle state exceeds the preset local state-time authentication area, the authentication margin is insufficient, or the equivalent control input is unrealizable; the online control module is configured to execute the finite-time trajectory tracking control method for unmanned container trucks in ports as described in any one of claims 1 to 8.

10. A port unmanned container truck, characterized in that, The system includes a vehicle body, a vehicle positioning unit, an inertial measurement unit, a wheel speed acquisition unit, a load identification unit, a road surface adhesion estimation unit, an actuator health monitoring unit, a chassis domain controller, a drive execution unit, a braking execution unit, and a steering execution unit; the chassis domain controller is configured with a finite-time trajectory tracking control system for unmanned container trucks in ports as described in claim 9, or the chassis domain controller is configured to execute a finite-time trajectory tracking control method for unmanned container trucks in ports as described in any one of claims 1 to 8.

Images