Unmanned longitudinal control method of concrete truck

Abstract

The invention relates to the technical field of unmanned longitudinal control, and particularly discloses an unmanned longitudinal control method for a concrete transport vehicle, which comprises the following steps: acquiring vehicle running state data information of the concrete transport vehicle in real time; the vehicle running state data information is input into a pre-trained longitudinal speed control strategy network, and the pre-trained longitudinal speed control strategy network obtains a function attenuation rate after an auxiliary sliding mode controller is designed according to a vehicle longitudinal dynamic model; constructing a mixed integer model prediction controller based on the auxiliary sliding mode controller to optimize the driving torque and the braking torque, and solving the mixed integer model prediction controller according to a discrete and continuous mixed action space near-end optimization strategy; and sending the driving control instruction to an execution mechanism of the concrete truck. According to the unmanned longitudinal control method for the concrete transport vehicle, the robustness, the real-time performance and the control precision of automatic driving of the concrete transport vehicle can be improved.

Description

Technical Field

[0001] This invention relates to the field of unmanned driving longitudinal control technology, and in particular to an unmanned driving longitudinal control method for a concrete transport vehicle. Background Technology

[0002] Longitudinal motion control of concrete transport vehicles is a fundamental and crucial task in autonomous driving systems. It aims to drive the vehicle to precisely track a preset speed through coordinated control of engine throttle opening and braking pressure. For autonomous concrete transport vehicles operating in complex environments such as mines and construction sites, longitudinal control presents significant challenges. First, concrete transport vehicles exhibit significant dynamic nonlinearity, and their engine drive and braking systems show marked response lag. Second, the dynamic changes in concrete load during transportation (such as mass fluctuations during empty, full-load, and unloading processes) lead to highly uncertainties in the vehicle's inertia parameters. Third, the vehicle is subjected to continuous external disturbances due to the coupled effects of road gradient fluctuations, rolling resistance, and nonlinear air resistance.

[0003] Current control schemes can be categorized into two types: one is based on analytical control laws (such as PID control and sliding mode control), and the other is based on optimal control (such as model predictive control).

[0004] Methods based on analytical control laws often use explicit expressions as control variables, which may include feedback terms, feedforward terms, or more complex algebraic terms. While some existing analytical control methods based on analytical control laws can provide explicit control laws, ensuring real-time control and exhibiting good control performance under specific conditions, they still have certain limitations: 1) Since vehicles are generally not allowed to perform simultaneous driving and braking operations, control schemes based on analytical control laws all require additional definitions of drive / braking switching logic to determine whether to perform drive torque control or braking torque control. However, in concrete transportation scenarios, under different road slopes and different load capacities, a fixed switching threshold may not be conducive to reasonable mode switching. Developing adaptive switching thresholds is cumbersome in engineering applications. 2) Driving and braking often require two separate control schemes with different control parameters. These parameters directly affect control accuracy and response speed, requiring separate parameter optimization in practical engineering applications, which is relatively cumbersome. 3) Both the drive torque and braking torque of concrete transport vehicles have physical limits. The aforementioned control strategies do not consider input saturation when analyzing the stability of the closed-loop system, and mostly adopt a direct truncation post-processing method. This cannot theoretically guarantee that the truncated control input meets the closed-loop stability, and may cause adverse phenomena such as control oscillation and instability.

[0005] Optimal control-based methods often model the drive and braking control problem as a single optimization problem, simultaneously solving for the optimal drive and braking control quantities. Integer decision variables are introduced to determine whether the current mode is drive or braking. While existing optimal control-based schemes can achieve self-optimization of drive and braking torque and improve control efficiency compared to analytical control laws, they still have the following limitations: 1) The concrete transportation scenario involves harsh environments with real-time changes in road gradient and vehicle load, and potential model mismatches in vehicle longitudinal dynamics. Current optimal control strategies often neglect the stability of the closed-loop system under uncertain disturbances, lacking robustness guarantees and making them unsuitable for the harsh environment and complex conditions of concrete transportation. 2) Feasibility proof of the optimization problem often relies on the feasibility at the initial moment, which in turn depends on the setting of the prediction time domain. As the prediction time domain increases, the computational burden also increases, making it difficult to meet the real-time requirements of engineering applications. 3) Some studies simplify the problem through methods such as simplified dynamic modeling or linearization, but the decrease in model accuracy inevitably reduces control accuracy.

[0006] Therefore, how to solve the problems of poor control robustness, limited feasibility, and difficulty in balancing real-time performance and control accuracy caused by uncertainties in the current autonomous driving of concrete transport vehicles has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] This invention provides a longitudinal control method for unmanned concrete transport vehicles, which solves the problems of poor control robustness, limited feasibility, and difficulty in balancing real-time performance and control accuracy caused by uncertainties in the autonomous driving of concrete transport vehicles in related technologies.

[0008] As one aspect of the present invention, an unmanned longitudinal control method for a concrete transport vehicle is provided, comprising:

[0009] Real-time acquisition of vehicle operating status data of concrete transport trucks, including at least vehicle speed, driving torque, braking torque and tilt sensor data;

[0010] The vehicle operating status data is input into a pre-trained longitudinal speed control strategy network to obtain drive control commands. The pre-trained longitudinal speed control strategy network is a function decay rate used for stability constraints after designing an auxiliary sliding mode controller based on the vehicle's longitudinal dynamics model. A hybrid integer model predictive controller is constructed based on the auxiliary sliding mode controller to optimize drive torque and braking torque. The hybrid integer model predictive controller is obtained by solving the hybrid integer model predictive controller according to a near-end optimization strategy for discrete and continuous hybrid motion space.

[0011] The drive control command is sent to the actuator of the concrete truck to realize unmanned longitudinal control of the concrete truck.

[0012] The pre-trained longitudinal speed control strategy network is updated based on the vehicle control response results of the concrete transport vehicle and the drive control commands.

[0013] Furthermore, the pre-trained longitudinal speed control strategy network is obtained by designing an auxiliary sliding mode controller based on the vehicle's longitudinal dynamics model, using the function decay rate for stability constraints. A hybrid integer model predictive controller is constructed based on the auxiliary sliding mode controller to optimize driving torque and braking torque. This hybrid integer model predictive controller is obtained by solving the network according to a near-end optimization strategy for a mixed discrete and continuous motion space, including:

[0014] Based on the longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels of the concrete transport vehicle, a dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level is constructed.

[0015] Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, a sliding mode controller under the driving condition and a sliding mode controller under the braking condition are constructed respectively, and the function decay rate for stability constraint under the driving condition and the function decay rate for stability constraint under the braking condition are obtained respectively.

[0016] A hybrid integer model predictive controller is constructed based on the sliding mode controller under driving conditions and the sliding mode controller under braking conditions;

[0017] The mixed integer model predictive controller is trained offline using the proximal policy optimization algorithm in the offline continuous mixed action space to obtain a pre-trained longitudinal velocity control policy network.

[0018] Furthermore, based on the overall longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels of the concrete transport vehicle, a dynamic model of the overall longitudinal motion and actuator level of the concrete transport vehicle is constructed, including:

[0019] Construct a longitudinal dynamic model of the entire concrete transport vehicle;

[0020] Construct rolling dynamics models for the front and rear wheels of a concrete transport vehicle, respectively.

[0021] The relationship between engine speed and speed of concrete transport vehicles is determined based on the transportation scenario of the concrete transport vehicles;

[0022] Determine the hysteresis coefficient under driving conditions and the hysteresis coefficient under braking conditions;

[0023] Based on the longitudinal dynamics model of the concrete transport vehicle, the rolling dynamics model of the front wheels, the rolling dynamics model of the rear wheels, the relationship between the vehicle's engine speed and velocity, the hysteresis coefficient under driving conditions, and the hysteresis coefficient under braking conditions, the dynamics model of the longitudinal motion of the concrete transport vehicle and the dynamics model at the actuator level are obtained.

[0024] Furthermore, based on the dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level, a sliding mode controller is constructed for both the driving and braking conditions. The function decay rate for stability constraints under the driving and braking conditions is obtained respectively, including:

[0025] Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the sliding mode variables under the driving and braking conditions are determined respectively.

[0026] The sliding mode control law under the driving condition is determined based on the sliding mode variables under the driving condition, and the sliding mode control law under the braking condition is determined based on the sliding mode variables under the braking condition.

[0027] The function decay rate for stability constraints under the driving condition is obtained based on the sliding mode control law under the driving condition, and the function decay rate for stability constraints under the braking condition is obtained based on the sliding mode control law under the braking condition.

[0028] Furthermore, based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the sliding mode variables under driving and braking conditions are determined respectively, including:

[0029] Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the adaptive update law under the driving condition and the adaptive update law under the braking condition are determined respectively. The expression of the adaptive update law under the driving condition is as follows:

[0030] ,

[0031] in, This represents the adaptive update law under the driving condition. This represents the anti-saturation variable of the control input under driving conditions. This represents the update coefficient under the driving condition. This indicates the transmission ratio from the engine to the wheels. These represent the dynamic coefficients in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This represents the driving hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates an engine torque control command. This represents the control input saturation function;

[0032] The expression for the adaptive update law under the braking condition is:

[0033] ,

[0034] in, This represents the adaptive update law under braking conditions. This represents the anti-saturation variable of the control input under braking conditions. This represents the update coefficient under braking conditions. This represents the braking hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates a torque control command for the braking system;

[0035] Based on the adaptive update laws under driving and braking conditions, sliding surfaces under driving and braking conditions are designed respectively, and sliding variables under driving and braking conditions are obtained. The sliding variables under driving conditions are expressed as follows:

[0036] ,

[0037] in, Represents the sliding mode variable under driving conditions. Indicates the sliding surface coefficient. Indicates control error. The first derivative represents the control error;

[0038] The sliding mode variable under the braking condition is expressed as follows:

[0039] ,

[0040] in, This represents the sliding mode variable under braking conditions.

[0041] Furthermore, determining the sliding mode control law under the driving condition based on the sliding mode variables under the driving condition, and determining the sliding mode control law under the braking condition based on the sliding mode variables under the braking condition, includes:

[0042] The sliding mode reaching law under the driving condition is determined based on the sliding mode variables under the driving condition. The expression of the sliding mode reaching law under the driving condition is as follows:

[0043] ,

[0044] in, This represents the sliding mode reaching law under driving conditions. Indicates the parameters of the sliding surface. Represents a symbolic function. This represents the upper bound of the change in the first derivative of the system's uncertainty.

[0045] The sliding mode control law under the driving condition is determined based on the sliding mode reaching law under the driving condition. The expression of the sliding mode control law under the driving condition is as follows:

[0046] ,

[0047] in, This indicates an engine torque control command. This indicates the engine's actual torque. The second derivative of the desired velocity, The first derivative representing the coefficient of drag on a vehicle;

[0048] The sliding mode reaching law under braking conditions is determined based on the sliding mode variables under braking conditions. The expression for the sliding mode reaching law under braking conditions is as follows:

[0049] ,

[0050] in, This represents the sliding mode approach law under braking conditions;

[0051] Based on the sliding mode approach law under the braking condition, the sliding mode control law under the braking condition is determined, and the expression of the sliding mode control law under the braking condition is as follows:

[0052] ,

[0053] in, This indicates a torque control command for the braking system. This indicates the total braking torque.

[0054] Furthermore, a hybrid integer model predictive controller is constructed based on the sliding mode controller under driving conditions and the sliding mode controller under braking conditions, including:

[0055] The cost function of the mixed integer model predictive controller is determined based on the control objective of the concrete transport vehicle;

[0056] The constraint objectives of the mixed integer model predictive controller are determined based on vehicle longitudinal dynamics constraints, drive system hysteresis constraints, braking system hysteresis constraints, function decay rate constraints, and control input saturation constraints.

[0057] Furthermore, the mixed-integer model predictive controller is trained offline using an offline continuous mixed action space proximal policy optimization algorithm to obtain a pre-trained longitudinal velocity control policy network, including:

[0058] Based on the speed error and its derivative, the anti-saturation state variables of the drive system and the braking system, the decision mode of the previous moment and the current road slope, a hybrid action space vector is constructed, and the actions of the hybrid action space vector are defined to include discrete actions and continuous actions.

[0059] Construct a policy network and a value network, wherein the policy network includes discrete policy heads and continuous policy heads;

[0060] The probability distribution of the policy network is modified according to the action masking mechanism based on function decay constraints so that the sampled torque command is located in the stable region.

[0061] Determine the reward function;

[0062] The policy network and value network are trained based on the reward function to construct a hybrid action space loss function and obtain a pre-trained longitudinal velocity control policy network.

[0063] Furthermore, the policy network and value network are trained according to the reward function to construct a hybrid action space loss function, including:

[0064] Initialize the policy network and value network, and determine the physical parameters of the concrete transport vehicle and the slip surface coefficient;

[0065] Perform trajectory sampling and embed stability constraints;

[0066] Calculate the advantage value and objective value at each moment based on the generalized advantage estimation;

[0067] Calculate the probability ratio of the mixed space and construct the loss function of the mixed action space, and update the policy network parameters and value network parameters simultaneously through backpropagation;

[0068] Repeat the above steps until the policy network converges.

[0069] Further, trajectory sampling and stability constraint embedding are performed, including:

[0070] Obtain the hybrid action space vector at the current moment;

[0071] Determine the stability boundary of the function decay constraint;

[0072] Determine the output pattern of the policy network based on the action mask;

[0073] The current action is applied to the vehicle model, and the hybrid action space vector for the next moment is updated according to the dynamic equations, and the current reward is calculated according to the reward function.

[0074] The unmanned longitudinal control method for concrete transport vehicles provided by this invention acquires real-time vehicle operating status data of the concrete transport vehicle and inputs this data into a pre-trained longitudinal speed control strategy network to obtain drive control commands. Based on these commands, unmanned longitudinal control of the concrete transport vehicle is achieved, and the pre-trained longitudinal speed control strategy network is updated based on the vehicle control response results. This unmanned longitudinal control method for concrete transport vehicles utilizes a pre-trained longitudinal speed control strategy network to construct a vehicle longitudinal dynamics model considering rear-wheel drive characteristics and actuator dynamic response. It designs an auxiliary sliding mode controller with anti-saturation variables and establishes a mixed-integer model predictive controller. The method employs a hybrid action space proximal strategy optimization algorithm to perform deep neural network offline approximation of the MPC solution process. Furthermore, it introduces an action masking mechanism based on a decay criterion during training and inference, achieving millisecond-level real-time optimal decision-making for drive / braking modes and control torque while meeting stringent vehicle stability constraints. The unmanned longitudinal control method for this concrete transport vehicle does not require additional definition of drive-brake switching logic. It automatically optimizes the drive-brake mode, drive torque, and brake torque based entirely on actual operating conditions. Simultaneously, drive and braking are considered in a single optimization problem, eliminating the need for iterative optimization of auxiliary sliding mode controller parameters. The auxiliary controller is only used to ensure stability, while tracking performance is guaranteed by the optimization problem. Furthermore, the auxiliary sliding mode controller considers uncertainties and ensures the robustness of the MPC based on function decay rate constraints. This invention solves the MPC optimization problem through discrete continuous mixed integer reinforcement learning, which significantly improves computational real-time performance compared to directly using a solver, ensuring the normal operation of unmanned concrete transport. Attached Figure Description

[0075] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof.

[0076] Figure 1 A flowchart of the unmanned longitudinal control method for a concrete transport vehicle provided by the present invention.

[0077] Figure 2 A flowchart illustrating the process of acquiring the pre-trained longitudinal velocity control strategy network provided by this invention.

[0078] Figure 3 A schematic diagram of the unmanned longitudinal control system architecture for a concrete transport vehicle provided by the present invention. Detailed Implementation

[0079] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0080] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0081] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0082] This embodiment provides a method for unmanned longitudinal control of a concrete transport vehicle. Figure 1 A flowchart of an unmanned longitudinal control method for a concrete transport vehicle according to an embodiment of the present invention is shown below. Figure 1 As shown, it includes:

[0083] S100. Real-time acquisition of vehicle operating status data information of concrete transport trucks, wherein the vehicle operating status data information includes at least vehicle speed, driving torque, braking torque and tilt angle sensor data.

[0084] In this embodiment of the invention, the on-board controller of the concrete transport truck communicates via the CAN bus at a certain control cycle (e.g., Real-time vehicle speed data collection Actual engine torque Braking system pressure (converted to braking torque) and tilt sensor data .

[0085] S200. The vehicle operating status data information is input into a pre-trained longitudinal speed control strategy network to obtain drive control commands. The pre-trained longitudinal speed control strategy network is a function decay rate obtained for stability constraints after designing an auxiliary sliding mode controller based on the vehicle longitudinal dynamics model. A hybrid integer model predictive controller is constructed based on the auxiliary sliding mode controller to optimize the drive torque and braking torque. The hybrid integer model predictive controller is obtained by solving the hybrid integer model predictive controller according to the near-end optimization strategy of discrete and continuous hybrid motion space.

[0086] In this embodiment of the invention, the vehicle operating status data, i.e., the current status, is... Input a pre-trained longitudinal velocity control policy network, and it directly outputs the Logits of discrete actions and the mean of continuous actions. After correcting the discrete and continuous actions, the corrected drive and braking mode variables are obtained. Continuous actions .

[0087] S300: Send the drive control command to the actuator of the concrete truck to realize unmanned longitudinal control of the concrete truck.

[0088] In this embodiment of the invention, the final control command after mask verification will be... , Converted to specific throttle opening Or braking pressure signal It is sent to the underlying execution mechanism.

[0089] S400: Update the pre-trained longitudinal speed control strategy network based on the vehicle control response results of the concrete transport vehicle and the drive control command.

[0090] In this embodiment of the invention, the pre-trained longitudinal speed control strategy network is updated at the next moment based on the actual executed control command and the vehicle response, forming a closed-loop control.

[0091] Therefore, the unmanned longitudinal control method for concrete transport vehicles provided by this invention acquires real-time vehicle operating status data of the concrete transport vehicle and inputs this data into a pre-trained longitudinal speed control strategy network to obtain drive control commands. Based on these commands, unmanned longitudinal control of the concrete transport vehicle is achieved, and the pre-trained longitudinal speed control strategy network is updated based on the vehicle control response results. This unmanned longitudinal control method for concrete transport vehicles utilizes a pre-trained longitudinal speed control strategy network to construct a vehicle longitudinal dynamics model considering rear-wheel drive characteristics and actuator dynamic response. It designs an auxiliary sliding mode controller with anti-saturation variables and establishes a mixed-integer model predictive controller (MPC) based on this model. The method employs a deep neural network offline approximation of the MPC solution process using a hybrid action space proximal policy optimization (PPO) algorithm and introduces an action masking mechanism based on attenuation criteria during training and inference. This achieves millisecond-level real-time optimal decision-making for drive / braking modes and control torque while meeting stringent vehicle stability constraints. The unmanned longitudinal control method for this concrete transport vehicle does not require additional definition of drive-brake switching logic. It automatically optimizes the drive-brake mode, drive torque, and brake torque based entirely on actual operating conditions. Simultaneously, drive and braking are considered in a single optimization problem, eliminating the need for iterative optimization of auxiliary sliding mode controller parameters. The auxiliary controller is only used to ensure stability, while tracking performance is guaranteed by the optimization problem. Furthermore, the auxiliary sliding mode controller considers uncertainties and ensures the robustness of the MPC based on function decay rate constraints. This invention solves the MPC optimization problem through discrete continuous mixed integer reinforcement learning, which significantly improves computational real-time performance compared to directly using a solver, ensuring the normal operation of unmanned concrete transport.

[0092] In this embodiment of the invention, the pre-trained longitudinal speed control strategy network is obtained by designing an auxiliary sliding mode controller based on the vehicle's longitudinal dynamics model, using a function decay rate for stability constraints. A hybrid integer model predictive controller is then constructed based on this auxiliary sliding mode controller to optimize driving torque and braking torque. The hybrid integer model predictive controller is then solved using a near-end optimization strategy for a mixed discrete and continuous motion space. Figure 2 As shown, it includes:

[0093] S210. Based on the longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels of the concrete transport vehicle, construct the longitudinal motion dynamics model of the concrete transport vehicle and the actuator level.

[0094] Specifically, based on the overall longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels of the concrete transport vehicle, a dynamic model of the overall longitudinal motion of the concrete transport vehicle and the actuator level is constructed, including:

[0095] (1) Construct a longitudinal dynamic model of the entire concrete transport vehicle;

[0096] Specifically, the expression for the longitudinal dynamics model of the concrete transport vehicle is as follows: (1)

[0097] in, Indicates the mass of concrete transport vehicles. , These represent the tangential reaction forces exerted on the front and rear wheels by the ground, respectively. Represents the air density constant. Indicates the air drag coefficient. Indicates the vehicle's frontal area. Indicates the ground rolling resistance coefficient. Represents the gravitational acceleration constant. Indicates the slope angle.

[0098] (2) Construct the front wheel rolling dynamics model and the rear wheel rolling dynamics model of the concrete transport vehicle respectively;

[0099] In this embodiment of the invention, concrete transport vehicles are mostly rear-wheel drive vehicles, with the front wheels generally having no driving function. A rolling dynamics model of the vehicle's front and rear wheels is established:

[0100] (2)

[0101] (3)

[0102] in, , These represent the moments of inertia of the front and rear wheels, respectively. , These represent the rotational angular velocities of the front and rear wheels, respectively. Indicates the driving torque at the rear wheel end. , and These represent the distance from the center of gravity to the front axle, the distance from the center of gravity to the rear axle, and the vehicle wheelbase, respectively. For the effective wheel radius, , These are the wheel-end braking torques for the front and rear wheels, respectively.

[0103] (3) Determine the relationship between engine speed and speed of the concrete transport vehicle based on the transportation scenario of the concrete transport vehicle;

[0104] In this embodiment of the invention, under a typical low-speed concrete transportation scenario, the relationship between engine speed and speed is as follows:

[0105] (4)

[0106] in, This indicates the transmission ratio from the engine to the wheels. This indicates the engine speed.

[0107] (4) Determine the hysteresis coefficient under driving conditions and the hysteresis coefficient under braking conditions;

[0108] (5) Based on the longitudinal dynamics model of the whole vehicle, the rolling dynamics model of the front wheel, the rolling dynamics model of the rear wheel, the relationship between the engine speed and speed of the vehicle, the hysteresis coefficient under the driving condition and the hysteresis coefficient under the braking condition, the dynamics model of the longitudinal motion of the whole vehicle and the actuator level of the concrete transport vehicle are obtained.

[0109] In this embodiment of the invention, the rotational dynamics equation of the vehicle engine is:

[0110] (5)

[0111] in, This represents the engine's moment of inertia. This indicates the engine's actual torque.

[0112] Combining the above equations, we can obtain a dynamic model that encompasses both the longitudinal motion of the entire vehicle and the actuator level:

[0113] (6)

[0114] Among them, coefficient , Indicates the total braking torque. Describing lumped uncertainty, its first derivative satisfies the bounded condition. , The upper bound of the derivative change represents uncertainty. The vehicle drag coefficient is defined as... Its first derivative can be approximately expressed as .

[0115] In addition, the hysteresis of the drive and braking system controls should also be considered:

[0116] (7)

[0117] (8)

[0118] in, , These represent the hysteresis coefficients of the drive system and the braking system, respectively. This indicates an engine torque control command. This indicates a torque control command for the braking system. This represents the engine's steady-state torque characteristic function. This indicates the throttle opening. Indicates braking pressure. This represents the braking pressure control coefficient. The desired control command is then obtained. and Then, it can be calculated in reverse. and .

[0119] It should be understood that the dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level in the embodiments of the present invention includes not only the above formula (6) but also formula (7) and formula (8) that take into account hysteresis factors.

[0120] S220. Based on the dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level, construct a sliding mode controller under the driving condition and a sliding mode controller under the braking condition, respectively, and obtain the function decay rate for stability constraint under the driving condition and the function decay rate for stability constraint under the braking condition.

[0121] In this embodiment of the invention, the design of a sliding mode controller is realized based on the dynamic model constructed above, thereby obtaining the stability constraints of the whole vehicle by introducing function decay as a mandatory constraint.

[0122] Specifically, based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, a sliding mode controller is constructed for both the driving and braking conditions. The function decay rate for stability constraints under the driving and braking conditions is obtained respectively, including:

[0123] (1) Determine the sliding mode variables under the driving condition and the braking condition respectively based on the dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level.

[0124] In this embodiment of the invention, adaptive update laws are obtained based on predefined anti-control input saturation variables under driving conditions and anti-control input saturation variables under braking conditions.

[0125] Specifically, based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the sliding mode variables under driving and braking conditions are determined respectively, including:

[0126] 11) Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, determine the adaptive update law under the driving condition and the adaptive update law under the braking condition, respectively. The expression for the adaptive update law under the driving condition is as follows:

[0127] (9)

[0128] in, This represents the adaptive update law under the driving condition. This represents the anti-saturation variable of the control input under driving conditions. This represents the update coefficient under the driving condition. This indicates the transmission ratio from the engine to the wheels. These represent the dynamic coefficients in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This represents the driving hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates an engine torque control command. This represents the control input saturation function;

[0129] The expression for the adaptive update law under the braking condition is:

[0130] (10)

[0131] in, This represents the adaptive update law under braking conditions. This represents the anti-saturation variable of the control input under braking conditions. This represents the update coefficient under braking conditions. This represents the braking hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates a torque control command for the braking system;

[0132] 12) Based on the adaptive update law under the driving condition and the adaptive update law under the braking condition, design the sliding surface under the driving condition and the sliding surface under the braking condition respectively, and obtain the sliding variables under the driving condition and the sliding variables under the braking condition, wherein the sliding variables under the driving condition are expressed as:

[0133] (11)

[0134] in, Represents the sliding mode variable under driving conditions. Indicates the sliding surface coefficient. Indicates control error. The first derivative represents the control error;

[0135] The sliding mode variable under the braking condition is expressed as follows:

[0136] (12)

[0137] in, This represents the sliding mode variable under braking conditions.

[0138] It should be noted that, in the embodiments of the present invention, Indicates control error, and , Indicates the desired speed.

[0139] (2) Determine the sliding mode control law under the driving condition based on the sliding mode variables under the driving condition, and determine the sliding mode control law under the braking condition based on the sliding mode variables under the braking condition.

[0140] In this embodiment of the invention, the respective sliding mode control law is determined according to different working conditions.

[0141] Specifically, determining the sliding mode control law under driving conditions based on the sliding mode variables under driving conditions, and determining the sliding mode control law under braking conditions based on the sliding mode variables under braking conditions, includes:

[0142] 21) Determine the sliding mode reaching law under the driving conditions based on the sliding mode variables under the driving conditions. The expression for the sliding mode reaching law under the driving conditions is as follows:

[0143] (13)

[0144] in, This represents the sliding mode reaching law under driving conditions. Indicates the parameters of the sliding surface. Represents a symbolic function. This represents the upper bound of the change in the first derivative of the system's uncertainty.

[0145] 22) Determine the sliding mode control law under the driving condition based on the sliding mode approach law under the driving condition. The expression of the sliding mode control law under the driving condition is as follows:

[0146] (14)

[0147] in, This indicates an engine torque control command. This indicates the engine's actual torque. The second derivative of the desired velocity, The first derivative representing the coefficient of drag on a vehicle;

[0148] 23) Determine the sliding mode reaching law under braking conditions based on the sliding mode variables under braking conditions. The expression for the sliding mode reaching law under braking conditions is as follows:

[0149] (17)

[0150] in, This represents the sliding mode approach law under braking conditions;

[0151] 24) Determine the sliding mode control law under the braking condition based on the sliding mode approach law under the braking condition. The expression of the sliding mode control law under the braking condition is as follows:

[0152] (18)

[0153] in, This indicates a torque control command for the braking system. This indicates the total braking torque.

[0154] (3) Obtain the function decay rate for stability constraint under the driving condition based on the sliding mode control law under the driving condition, and obtain the function decay rate for stability constraint under the braking condition based on the sliding mode control law under the braking condition.

[0155] It should be understood that, under driving conditions, the Lyapunov candidate function is defined as follows:

[0156] (15)

[0157] The Lyapunov function decay rate of the closed-loop system under the auxiliary control law is:

[0158] (16)

[0159] Therefore, under driving conditions, the closed-loop system state can gradually reach the sliding surface in a finite amount of time.

[0160] Similarly, under braking conditions, the Lyapunov candidate function is defined as follows:

[0161] (19)

[0162] The Lyapunov function decay rate of the closed-loop system under the auxiliary control law is:

[0163] (20)

[0164] Therefore, under braking conditions, the closed-loop system state can gradually reach the sliding surface in a finite amount of time.

[0165] S230. Construct a hybrid integer model predictive controller based on the sliding mode controller under driving conditions and the sliding mode controller under braking conditions;

[0166] In this embodiment of the invention, for concrete transport vehicles, it is necessary to ensure the accuracy of speed tracking, the smoothness of control, and the smoothness of drive and brake switching.

[0167] Specifically, a hybrid integer model predictive controller is constructed based on the sliding mode controller under driving conditions and the sliding mode controller under braking conditions, including:

[0168] (1) Determine the cost function of the mixed integer model predictive controller based on the control objective of the concrete transport vehicle;

[0169] In this embodiment of the invention, the cost function of the constructed hybrid integer model prediction controller is as follows:

[0170] (21a)

[0171] in, Indicates continuous control quantity. This represents the prediction time domain of MPC. Indicates the sampling period, and the subscript "" in all variable symbols "All of these indicate that the variable is in the first" The value at time, for example Indicates speed In the The value at time, The first term in the cost function is the speed tracking accuracy cost, which penalizes the difference between the vehicle speed and the reference speed in the future time domain; the second term is the ride comfort cost, which penalizes the speed change. The third item is the cost of driving and braking switching, which penalizes the frequency of driving and braking switching in the future time domain. , and These are the weighting coefficients for these three costs; the more emphasis is placed on a certain cost, the larger its weight value should be.

[0172] (2) Determine the constraint objectives of the mixed integer model predictive controller based on the vehicle longitudinal dynamics constraints, drive system hysteresis constraints, braking system hysteresis constraints, function decay rate constraints, and control input saturation constraints.

[0173] In this embodiment of the invention, the constraint equations for the mixed-integer model prediction controller are as follows:

[0174] (21b)

[0175] It should be understood that in the above constraint equations, the first term represents the longitudinal dynamics constraint of the vehicle, which introduces a binary decision variable. The first term ensures that driving and braking do not trigger simultaneously; the second term represents the drive system hysteresis constraint, indicating engine torque response hysteresis; the third term represents the braking system hysteresis constraint, indicating braking torque response hysteresis; the fourth term represents the Lyapunov function decay rate constraint, where... This constraint requires that the decay rate of the Lyapunov function of the closed-loop system under the control variable obtained by MPC is always no slower than the decay rate of the auxiliary sliding mode control law, thus ensuring the stability of the MPC optimization problem; the fifth term represents the control input saturation constraint, which limits the control variable solved by MPC to a minimum value. and maximum value between.

[0176] S240. The mixed integer model predictive controller is trained offline according to the proximal policy optimization algorithm of the offline continuous mixed action space to obtain the pre-trained longitudinal velocity control policy network.

[0177] It should be noted that, since the above optimization problem is an NP-hard mixed-integer nonconvex, nonlinear programming problem, in order to ensure the real-time performance of control during the solution of the above optimization problem, this embodiment of the invention uses a proximal policy optimization algorithm based on a discrete-continuous hybrid action space to train the controller offline in a simulation environment. This method uses a deep neural network to approximate the solution of MPC, which can effectively solve the problem of poor real-time performance in online solutions of mixed-integer nonconvex, nonlinear programming problems.

[0178] Specifically, the mixed-integer model predictive controller is trained offline using an offline continuous mixed action space proximal policy optimization algorithm to obtain a pre-trained longitudinal velocity control policy network, including:

[0179] (1) Construct a hybrid action space vector based on the speed error and its derivative, the anti-saturation state variables of the drive system, the anti-saturation state variables of the braking system, the decision mode of the previous moment and the current road slope, and define the actions of the hybrid action space vector as including discrete actions and continuous actions.

[0180] In this embodiment of the invention, a state space vector is constructed. .in: Represents the velocity error and its derivative; These represent the anti-saturation state variables of the drive and braking systems, respectively. The decision mode of the previous moment (0 represents braking, 1 represents driving) is used to calculate the switching cost; This indicates the current road gradient.

[0181] Define hybrid action space vector action It consists of two parts:

[0182] 1) Discrete Actions Decision-driven model ( ) or braking mode ( );

[0183] 2) Continuous actions Normalized torque output.

[0184] Mapping of actual control variables:

[0185] 1) If Then the driving torque control command Braking torque ;

[0186] 2) If Then the braking torque control command driving torque .

[0187] (2) Construct a policy network and a value network, wherein the policy network includes discrete policy heads and continuous policy heads;

[0188] In this embodiment of the invention, a dual-head policy network is established. ,include:

[0189] 1) Discrete Strategy Head: Outputs the non-normalized logarithmic probabilities (i.e., Logits) of the driving and braking modes, denoted as... .

[0190] 2) Continuous Policy Header: Outputs Gaussian distribution parameters (mean) of continuous actions. and standard deviation ).

[0191] Value Network With policy network It shares a state vector and also requires an action vector as input, ultimately outputting a policy value assessment.

[0192] (3) Correct the probability distribution of the policy network according to the action masking mechanism based on function decay constraint so that the sampled torque command is located in the stable region;

[0193] In this embodiment of the invention, in order to ensure that the output action strictly satisfies the Lyapunov stability constraint, an action masking mechanism is introduced before sampling the action. This mechanism will forcibly correct the probability distribution of the policy network.

[0194] Based on the Lyapunov decay constraint formula derived above, substituting the current state... The solution satisfies The feasible region of control quantity.

[0195] 1) Driving feasibility criteria: For It must meet the following requirements:

[0196] ,(twenty two)

[0197] according to The sign of the inequality is defined by the fact that the inequality has a positive or negative sign. The upper or lower bound of the interval is denoted as the feasible interval. .

[0198] 2) Braking feasibility criteria: For It must meet the following requirements:

[0199] ,(twenty three)

[0200] Similarly, solve feasible range .

[0201] The masking operation for discrete actions is as follows:

[0202] 1) If If stability cannot be achieved regardless of adjustments to the driving force, then the discrete strategy header will be... Set as .

[0203] 2) If Then Set as .

[0204] After passing through the Softmax layer, the probability of selecting invalid discrete patterns will be forced to zero.

[0205] After selecting a pattern, the mean of the neural network output is... Projected onto the calculated feasible interval Within this range, ensure that the sampled torque command always remains within the stability region.

[0206] (4) Determine the reward function;

[0207] In this embodiment of the invention, the reward function is designed as follows:

[0208] ,(twenty four)

[0209] in, All represent weights in the reward function, which can be compared with those in the MPC cost function. , and Maintain consistency. Note that since the action mask already guarantees stability, the reward function no longer needs to include the Lyapunov penalty term.

[0210] (5) Train the policy network and value network according to the reward function to construct a hybrid action space loss function and obtain a pre-trained longitudinal velocity control policy network.

[0211] In this embodiment of the invention, the policy network and the value network are trained according to the reward function to construct a hybrid action space loss function, including:

[0212] 51) Initialize the policy network and value network, and determine the physical parameters of the concrete transport vehicle and the slip surface coefficient;

[0213] Specifically, initialize the environment and network parameters: initialize the policy network. and value network Determine the physical parameters (mass) of the concrete transport truck. Wheel radius Transmission ratio (etc.) and sliding surface coefficient .

[0214] 52) Perform trajectory sampling and embed stability constraints;

[0215] Trajectory Sampling and Stability Constraint Embedding: In each round of sampling, for each time step The following processing steps are performed on all of them.

[0216] Specifically, trajectory sampling and stability constraint embedding are performed, including:

[0217] 521) Obtain the hybrid action space vector at the current moment;

[0218] Get the current state .

[0219] 522) Determine the stability boundary of the function decay constraint;

[0220] In this embodiment of the invention, the Lyapunov stability boundary is calculated based on the current sliding surface. Based on equations (22) and (23) above, calculate the stability criterion. The feasible region of control quantity.

[0221] 523) Determine the output mode of the policy network based on the action mask;

[0222] Apply action mask: If the driving or braking mode is determined to be unavailable, then set the logarithmic probability of the corresponding mode output by the policy network to 1. The probability distribution is recalculated using the Softmax function. Decision-making patterns are obtained by sampling from them. In the selected mode Then, the Gaussian distribution output from the policy network. The original torque was obtained by sampling. Subsequently, a forced projection is performed using the corresponding feasible region of the previously calculated mode:

[0223] (25)

[0224] 524) Apply the current action to the vehicle model, update the hybrid action space vector for the next moment according to the dynamic equations, and calculate the current reward according to the reward function.

[0225] Specifically, perform actions and update the state: This involves updating the action... Apply to the vehicle model, update the state according to the dynamic equations. And calculate the current reward based on the reward function. .

[0226] 53) Calculate the advantage value and target value at each time step based on the generalized advantage estimation;

[0227] In this embodiment of the invention, the advantage function and the objective value are calculated: the advantage value at each time step is calculated using generalized advantage estimation. and target value ,Right now:

[0228] (26)

[0229] 54) Calculate the mixed space probability ratio and construct the mixed action space loss function, and update the policy network parameters and value network parameters simultaneously through backpropagation;

[0230] Policy Network and Value Network Update: Calculating the Probability Ratio of the Mixed Space ,Right now:

[0231] (27)

[0232] Constructing the hybrid action space PPO loss function:

[0233] (28)

[0234] The policy network parameters are updated simultaneously through backpropagation. and value network parameters .

[0235] 55) Repeat the above steps until the policy network converges, that is, the velocity tracking error reaches the expected index.

[0236] Finally, online deployment is performed, exporting the trained policy network as an executable model for embedded systems, and deploying it on the onboard computing unit of the concrete transport truck. Figure 3 The intelligent driving domain controller shown. The main steps are as follows:

[0237] Step 1: Real-time status acquisition and observation.

[0238] The vehicle controller communicates via the CAN bus at certain control cycles (e.g.) Real-time vehicle speed data collection Actual engine torque Braking system pressure (converted to braking torque) and tilt sensor data Simultaneously, the anti-saturation variable is calculated iteratively based on the aforementioned discretized equation. and .

[0239] Step 2: Action mask processing

[0240] 1) Forward propagation: Propagate the current state Input the trained policy network, and it directly outputs the Logits of discrete actions and the mean of continuous actions. .

[0241] 2) Real-time motion mask processing: Based on step 2) in the offline training process, obtain the corrected driving and braking mode variables. Continuous actions .

[0242] Step 3: Execution of control commands

[0243] The final control command after mask verification Converted to specific throttle opening Or braking pressure signal It is sent to the underlying execution mechanism.

[0244] Step 4: Anti-saturation state update

[0245] Update the anti-saturation variable for the next time step based on the actual control commands executed and the vehicle response. and This forms a closed-loop control.

[0246] In embodiments of the present invention, the following are employed: Figure 3 The control system architecture is shown. In the longitudinal speed control strategy designed in this invention, the intelligent driving domain controller obtains the combined inertial navigation speed via the CAN bus. With acceleration The vehicle mass is obtained through load sensors or estimators. Road slope is obtained through tilt sensors or slope estimators. After calculation, the drive torque control command is output. Or braking torque control command The designed longitudinal speed control strategy consists of three parts: 1) an auxiliary sliding mode controller, which generates Lyapunov function decay rate constraints to ensure the stability of the closed-loop system under uncertainty in the presence of MPC; 2) a mixed integer model predictive controller, which considers the optimization problems of driving torque and braking torque under a unified framework; and 3) a discrete-continuous hybrid motion space proximal strategy optimization module, which is used to efficiently solve the MPC problem to meet real-time requirements.

[0247] In drive mode, drive torque control command It acts directly on the engine controller, calculating the throttle opening from the engine's calibrated MAP chart, thereby controlling the engine's actual torque output. This torque is amplified through the transmission and driveshaft and then applied to the rear axle wheels, generating ground driving force. In braking mode, it operates according to braking torque control commands. The brake hydraulic pressure can be calculated in reverse, and then the electronic braking system can coordinate and distribute the braking pressure between the front and rear axles to achieve deceleration by utilizing ground friction.

[0248] This invention presents an unmanned longitudinal control method for concrete transport vehicles. It eliminates the need for additional drive / brake switching logic, automatically optimizing the drive / brake mode, drive torque, and braking torque based entirely on actual operating conditions. Simultaneously, drive and braking are considered within a single optimization problem, eliminating the need for iterative optimization of auxiliary sliding mode controller (MPC) parameters. The auxiliary controller is solely responsible for ensuring stability, with tracking performance guaranteed by the optimization problem. Furthermore, both the auxiliary controller and the MPC optimization problem consider control input saturation in the closed-loop system stability analysis, fundamentally avoiding oscillations and instability caused by truncated post-processing. Additionally, the auxiliary sliding mode controller considers uncertainties, and the designed Lyapunov function decay rate constraint ensures the robustness of the MPC. Moreover, due to the designed anti-saturation variables and their update mechanism, the auxiliary sliding mode control law can always serve as a feasible solution for the MPC, making the feasibility of the MPC independent of the prediction time domain. Furthermore, this invention solves the MPC optimization problem using discrete-continuous mixed integer reinforcement learning, significantly improving real-time computation compared to directly using a solver, thus ensuring the normal operation of unmanned concrete transport.

[0249] In summary, the unmanned longitudinal control method for concrete transport vehicles provided by this invention has the following advantages compared to the prior art:

[0250] 1) This invention significantly improves the driving stability and safety of unmanned vehicles under complex working conditions. By introducing Lyapunov function decay as a mandatory constraint and transforming this constraint into an action mask for reinforcement learning through analytical solutions, the system remains within the stable domain during exploration and execution. This theoretically avoids the uninterpretability and potential instability risks caused by the "black box" nature of pure neural network control algorithms, ensuring the operational safety of concrete transport vehicles in extreme scenarios such as heavy-load ramps.

[0251] 2) This invention effectively resolves the contradiction between complex multi-objective optimization control and the limited computing power of the vehicle controller. Utilizing the offline training characteristics of deep reinforcement learning, the complex mixed-integer nonlinear programming solution process of MPC is parameterized through a neural network. During online deployment, the vehicle controller only needs to perform simple neural network forward inference and action mask filtering based on linear analytical solutions to output the optimal mode switching signal and torque command within milliseconds, significantly reducing computational latency within the system sampling interval and improving control frequency and response speed.

[0252] 3) This invention specifically addresses the control failure problem caused by the physical limitations of actuators. By introducing anti-saturation variables independently designed for the drive and braking systems into the dynamic model, and explicitly considering input saturation characteristics in the auxiliary controller and MPC constraints, the controller can perceive the actuator's response boundary. When the vehicle experiences power limitation or braking torque saturation, the anti-saturation variables can dynamically adjust the sliding surface and predict the trajectory, effectively suppressing the vibration caused by frequent actuator starts and stops, improving the service life of the actuator and the ride comfort during concrete transportation.

[0253] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for unmanned longitudinal control of a concrete transport vehicle, characterized in that, include: Real-time acquisition of vehicle operating status data of concrete transport trucks, including at least vehicle speed, driving torque, braking torque and tilt sensor data; The vehicle operating status data is input into a pre-trained longitudinal speed control strategy network to obtain drive control commands. The pre-trained longitudinal speed control strategy network is a function decay rate used for stability constraints after designing an auxiliary sliding mode controller based on the vehicle's longitudinal dynamics model. A hybrid integer model predictive controller is constructed based on the auxiliary sliding mode controller to optimize drive torque and braking torque. The hybrid integer model predictive controller is obtained by solving the hybrid integer model predictive controller according to a near-end optimization strategy for discrete and continuous hybrid motion space. The drive control command is sent to the actuator of the concrete truck to realize unmanned longitudinal control of the concrete truck. The pre-trained longitudinal speed control strategy network is updated based on the vehicle control response results of the concrete transport vehicle and the drive control commands.

2. The unmanned longitudinal control method for a concrete transport vehicle according to claim 1, characterized in that, The pre-trained longitudinal speed control strategy network is obtained by designing an auxiliary sliding controller based on the vehicle's longitudinal dynamics model, using a function decay rate for stability constraints. A hybrid integer model predictive controller is constructed based on this auxiliary sliding controller to optimize driving torque and braking torque. The hybrid integer model predictive controller is then solved using a near-end optimization strategy for a mixed discrete and continuous motion space, including: Based on the longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels of the concrete transport vehicle, a dynamic model of the longitudinal motion of the concrete transport vehicle and the actuator level is constructed. Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, a sliding mode controller under the driving condition and a sliding mode controller under the braking condition are constructed respectively, and the function decay rate for stability constraint under the driving condition and the function decay rate for stability constraint under the braking condition are obtained respectively. A hybrid integer model predictive controller is constructed based on the sliding mode controller under driving conditions and the sliding mode controller under braking conditions; The mixed integer model predictive controller is trained offline using the proximal policy optimization algorithm in the offline continuous mixed action space to obtain a pre-trained longitudinal velocity control policy network.

3. The unmanned longitudinal control method for a concrete transport vehicle according to claim 2, characterized in that, Based on the overall longitudinal dynamics model of the concrete transport vehicle and the rolling dynamics model of the front and rear wheels, a dynamic model of the overall longitudinal motion and actuator level of the concrete transport vehicle is constructed, including: Construct a longitudinal dynamic model of the entire concrete transport vehicle; Construct rolling dynamics models for the front and rear wheels of a concrete transport vehicle, respectively. The relationship between engine speed and speed of concrete transport vehicles is determined based on the transportation scenario of the concrete transport vehicles; Determine the hysteresis coefficient under driving conditions and the hysteresis coefficient under braking conditions; Based on the longitudinal dynamics model of the concrete transport vehicle, the rolling dynamics model of the front wheels, the rolling dynamics model of the rear wheels, the relationship between the vehicle's engine speed and velocity, the hysteresis coefficient under driving conditions, and the hysteresis coefficient under braking conditions, the dynamics model of the longitudinal motion of the concrete transport vehicle and the dynamics model at the actuator level are obtained.

4. The unmanned longitudinal control method for a concrete transport vehicle according to claim 2, characterized in that, Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, a sliding mode controller is constructed for both the driving and braking conditions. The function decay rate for stability constraints under driving and braking conditions is obtained respectively, including: Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the sliding mode variables under the driving and braking conditions are determined respectively. The sliding mode control law under the driving condition is determined based on the sliding mode variables under the driving condition, and the sliding mode control law under the braking condition is determined based on the sliding mode variables under the braking condition. The function decay rate for stability constraints under the driving condition is obtained based on the sliding mode control law under the driving condition, and the function decay rate for stability constraints under the braking condition is obtained based on the sliding mode control law under the braking condition.

5. The unmanned longitudinal control method for a concrete transport vehicle according to claim 4, characterized in that, Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the sliding mode variables under driving and braking conditions are determined respectively, including: Based on the longitudinal motion of the concrete transport vehicle and the dynamic model at the actuator level, the adaptive update law under the driving condition and the adaptive update law under the braking condition are determined respectively. The expression of the adaptive update law under the driving condition is as follows: ， in, This represents the adaptive update law under the driving condition. This represents the anti-saturation variable of the control input under driving conditions. This represents the update coefficient under the driving condition. This indicates the transmission ratio from the engine to the wheels. These represent the dynamic coefficients in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This represents the driving hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates an engine torque control command. This represents the control input saturation function; The expression for the adaptive update law under the braking condition is: ， in, This represents the adaptive update law under braking conditions. This represents the anti-saturation variable of the control input under braking conditions. This represents the update coefficient under braking conditions. This represents the braking hysteresis coefficient in the dynamic model of the longitudinal motion of the concrete transport vehicle at the actuator level. This indicates a torque control command for the braking system; Based on the adaptive update laws under driving and braking conditions, sliding surfaces under driving and braking conditions are designed respectively, and sliding variables under driving and braking conditions are obtained. The sliding variables under driving conditions are expressed as follows: ， in, Represents the sliding mode variable under driving conditions. Indicates the sliding surface coefficient. Indicates control error. The first derivative represents the control error; The sliding mode variable under the braking condition is expressed as follows: ， in, This represents the sliding mode variable under braking conditions.

6. The unmanned longitudinal control method for a concrete transport vehicle according to claim 4, characterized in that, Determining the sliding mode control law under driving conditions based on sliding mode variables under driving conditions, and determining the sliding mode control law under braking conditions based on sliding mode variables under braking conditions, including: The sliding mode reaching law under the driving condition is determined based on the sliding mode variables under the driving condition. The expression of the sliding mode reaching law under the driving condition is as follows: ， in, This represents the sliding mode reaching law under driving conditions. Indicates the parameters of the sliding surface. Represents a symbolic function. This represents the upper bound of the change in the first derivative of the system's uncertainty. The sliding mode control law under the driving condition is determined based on the sliding mode reaching law under the driving condition. The expression of the sliding mode control law under the driving condition is as follows: ， in, This indicates an engine torque control command. This indicates the engine's actual torque. The second derivative of the desired velocity, The first derivative representing the coefficient of drag on a vehicle; The sliding mode reaching law under braking conditions is determined based on the sliding mode variables under braking conditions. The expression for the sliding mode reaching law under braking conditions is as follows: ， in, This represents the sliding mode approach law under braking conditions; Based on the sliding mode approach law under the braking condition, the sliding mode control law under the braking condition is determined, and the expression of the sliding mode control law under the braking condition is as follows: ， in, This indicates a torque control command for the braking system. This indicates the total braking torque.

7. The unmanned longitudinal control method for a concrete transport vehicle according to claim 2, characterized in that, A hybrid integer model predictive controller is constructed based on the sliding mode controller under driving and braking conditions, including: The cost function of the mixed integer model predictive controller is determined based on the control objective of the concrete transport vehicle; The constraint objectives of the mixed integer model predictive controller are determined based on vehicle longitudinal dynamics constraints, drive system hysteresis constraints, braking system hysteresis constraints, function decay rate constraints, and control input saturation constraints.

8. The unmanned longitudinal control method for a concrete transport vehicle according to claim 2, characterized in that, The mixed-integer model predictive controller is trained offline using an offline continuous mixed action space proximal policy optimization algorithm to obtain a pre-trained longitudinal velocity control policy network, including: Based on the speed error and its derivative, the anti-saturation state variables of the drive system and the braking system, the decision mode of the previous moment and the current road slope, a hybrid action space vector is constructed, and the actions of the hybrid action space vector are defined to include discrete actions and continuous actions. Construct a policy network and a value network, wherein the policy network includes discrete policy heads and continuous policy heads; The probability distribution of the policy network is modified according to the action masking mechanism based on function decay constraints so that the sampled torque command is located in the stable region. Determine the reward function; The policy network and value network are trained based on the reward function to construct a hybrid action space loss function and obtain a pre-trained longitudinal velocity control policy network.

9. The unmanned longitudinal control method for a concrete transport vehicle according to claim 8, characterized in that, The policy network and value network are trained based on the reward function to construct a hybrid action space loss function, including: Initialize the policy network and value network, and determine the physical parameters of the concrete transport vehicle and the slip surface coefficient; Perform trajectory sampling and embed stability constraints; Calculate the advantage value and objective value at each moment based on the generalized advantage estimation; Calculate the probability ratio of the mixed space and construct the loss function of the mixed action space. Update the policy network parameters and value network parameters simultaneously through backpropagation. Repeat the above steps until the policy network converges.

10. The unmanned longitudinal control method for a concrete transport vehicle according to claim 9, characterized in that, Perform trajectory sampling and stability constraint embedding, including: Obtain the hybrid action space vector at the current moment; Determine the stability boundary of the function decay constraint; Determine the output pattern of the policy network based on the action mask; The current action is applied to the vehicle model, and the hybrid action space vector for the next moment is updated according to the dynamic equations, and the current reward is calculated according to the reward function.

Images