Speed control method and device of unmanned vehicle and unmanned vehicle

Abstract

The invention relates to the technical field of unmanned vehicle control, and discloses a speed control method and device for an unmanned vehicle and the unmanned vehicle, and the method comprises the steps: obtaining the multi-mode sensor data of the surrounding environment of the vehicle; constructing a dynamic vortex field model and Brownian motion parameter mapping relation based on the sensor data; dynamically adjusting a sensor acquisition strategy according to the vortex field intensity and compressing a prediction time domain; constructing a combined control equation of the coupled vortex field and Brownian motion parameters; and a speed control instruction is generated through variational optimization solution. According to the technical scheme of fusing the hydromechanics vortex field theory and Brownian motion random mapping, unified modeling of traffic flow microscopic disturbance and environment uncertainty is achieved, and the defect that a deterministic model is difficult to adapt to dynamic changes of complex traffic scenes is effectively overcome; therefore, the speed control of the unmanned vehicle has the physical interpretable response capability to sudden disturbance.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned vehicle control, and in particular to a speed control method and device for an unmanned vehicle and an unmanned vehicle. Background Art

[0002] With the rapid development of artificial intelligence and autonomous driving technologies, autonomous vehicles (AVs) have become a research hotspot in the transportation sector. Currently, speed control for AVs relies primarily on environmental data collected by sensors and pre-defined control algorithms to achieve autonomous driving. However, challenges remain, such as complex and changing road environments, high-performance data processing requirements, and dynamic obstacle avoidance. Speed ​​control for AVs not only impacts driving efficiency but also directly impacts safety.

[0003] Existing speed control technologies for autonomous vehicles generally use a discretized traffic element modeling approach, separating parameters such as road obstacles, traffic density, and moving targets into independent variables. This modeling approach struggles to capture the dynamic coupling of multiple targets within traffic flow. This is particularly true in complex interactive scenarios such as vehicle cut-ins and convergences. The linear superposition model established by traditional methods cannot accurately represent the nonlinear interactions between vehicles and the flow field, leading to frequent control command lags and overshoots, severely restricting driving safety in complex urban scenarios.

[0004] In terms of environmental perception and data processing, most technical solutions employ a rigid strategy of fixed scanning frequency and full-time prediction, failing to dynamically allocate perception resources based on traffic conditions. This data collection model results in insufficient data collection accuracy in critical areas and waste of resources in non-critical areas. Furthermore, the excessively long prediction time horizon introduces a significant amount of uncertainty, significantly reducing the real-time performance of control models in emergency scenarios. Summary of the Invention

[0005] The purpose of the present invention is to provide a speed control method and device for an unmanned vehicle, and an unmanned vehicle, which solves the problem that deterministic models are difficult to adapt to the dynamic changes of complex traffic scenes.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A speed control method for an unmanned vehicle, comprising: Acquire multimodal sensor data of the vehicle's surroundings; Constructing a dynamic vortex field model and a Brownian motion parameter mapping relationship based on the sensor data; Dynamically adjust sensor acquisition strategies and compress prediction time domain according to eddy field intensity; Construct the joint governing equations of the coupled vortex field and Brownian motion parameters; Generate speed control instructions through variational optimization solution; Execute control instructions based on dynamic parameter adjustment rules and update model parameters in a closed loop Furthermore, the core of the dynamic vortex field model is to regard traffic flow as a viscous fluid and quantify the local vortex intensity distribution through modified fluid mechanics equations. Unlike traditional methods, this solution embeds dynamic parameters such as the vehicle's own mass and moment of inertia into the calculation of fluid characteristics, so that the model can reflect the active impact of vehicle movement on the surrounding traffic flow in real time. Brownian motion parameter mapping converts traffic density and speed fluctuations into equivalent temperature parameters through thermodynamic analogy, giving physical meaning to the random disturbance terms. The variational optimization process jointly optimizes the vehicle kinetic energy and vortex field potential energy through the principle of energy optimality to generate speed instructions that take into account both smoothness and energy efficiency.

[0007] Preferably, the constructing of the dynamic vortex field model includes: Calculate the local vorticity distribution based on the modified Navier-Stokes equations, where the viscosity coefficient is related to the vehicle mass and moment of inertia; Mapping traffic density to equivalent temperature parameters, constructing vehicle dynamics equations including random disturbance terms; Navier-Stokes equation calculation formula: in: The instantaneous acceleration of the traffic flow velocity field reflects the dynamic changes of the vehicle's surrounding environment; Local traffic flow velocity vector; t: time; v: kinematic viscosity coefficient, calculated from fluid density and dynamic viscosity; k: dimensionless proportionality coefficient; m: vehicle mass; I: The moment of inertia of the vehicle about its center of mass.

[0008] Furthermore, traditional traffic flow models typically use a fixed viscosity coefficient, which fails to reflect the perturbation effects of vehicle inertia on the surrounding environment. This solution, through the functional relationship between dynamic viscosity and vehicle parameters, automatically enables heavy vehicles or vehicles with high moments of inertia to exhibit stronger flow field damping characteristics in the model. The mapping logic of the equivalent temperature parameter converts the product relationship between traffic density and average speed into a thermodynamic temperature analogy, allowing the strength of the random perturbation term to be quantified by the ratio of vehicle mass to flow field parameters, achieving a coupled characterization of traffic flow randomness and vehicle inertia.

[0009] Preferably, the dynamic adjustment of the sensor acquisition strategy includes: When the vortex intensity exceeds a preset threshold, the infrared thermal imaging channel is activated and focuses on the high temperature gradient area; The lidar scanning area is dynamically reduced according to the temperature gradient distribution, and the time domain range of the model prediction control is simultaneously compressed.

[0010] Furthermore, the activation conditions for the infrared thermal imaging channel are linked to the vortex field intensity threshold. When a strong vortex disturbance is detected (such as a traffic conflict at an intersection), the system prioritizes capturing areas with significant temperature gradients (such as vehicle engine heat sources), allowing it to quickly locate potential risk targets. The lidar scanning range is dynamically adjusted using a nonlinear compression strategy, focusing on the core area in front of the vehicle. The prediction time domain is dynamically shortened based on vortex intensity to ensure optimized real-time calculations in high-disturbance scenarios.

[0011] Preferably, the constructing of the joint control equation includes: A partial differential coupling relationship between the vortex field intensity and the Brownian motion parameters is established, where the coupling coefficient is dynamically correlated with the vehicle's yaw rate. A multi-objective optimization function including acceleration smoothing term and energy consumption gradient term is defined, and its weight coefficient is adaptively adjusted with the vortex field intensity; Partial differential coupling relationship formula: in: The rate of change of vortex field intensity per unit time; ω: scalar value of eddy field strength; The real-time yaw rate of the vehicle; kT: traffic density ρ and average speed Generated equivalent temperature parameters; m: vehicle mass; Local traffic flow velocity vector; x: vehicle longitudinal coordinate; The instantaneous acceleration of the traffic flow velocity field.

[0012] Furthermore, the physical quantities calculated on the right side of the equation reflect the impact of random perturbations in traffic flow on vehicle acceleration. When the vehicle is turning, the coupling coefficient automatically adjusts the rate of evolution of the vortex field strength, allowing the turning action to actively alter the surrounding flow field characteristics. In the multi-objective optimization function, the weight of the acceleration smoothing term is inversely proportional to the vortex strength, while the weight of the energy gradient term is directly proportional, achieving an adaptive balance between the control objective and environmental perturbations.

[0013] Preferably, the variational optimization solution includes: Embed the stochastic differential equation into the Hamiltonian action principle: in: δ: variational operator, representing a small change to the action functional, used to solve the extreme path of the vehicle's motion trajectory; ∫: time integration operator; m: vehicle mass; Vehicle velocity vector; vortex field potential energy integral; Spatial location The vortex field strength at ; The volume of small spatial regions in the vehicle's surroundings; dt: discrete time step of the control system; Three-dimensional space vector; t: time; The optimization problem is solved by a numerical discretization method with adaptive step size, and the step size is dynamically switched according to the modified Reynolds number.

[0014] Furthermore, the construction of the energy optimization principle breaks through the limitations of traditional single-point velocity optimization and calculates the total potential energy of the vortex field around the vehicle through three-dimensional spatial integration. The adaptive step size mechanism dynamically adjusts according to the flow field state: a fine step size is used to improve resolution when turbulent characteristics are prominent, and a coarse step size is switched to accelerate convergence during stable flow phases, achieving a coordinated optimization of computational accuracy and efficiency.

[0015] Preferably, the execution control instruction includes: The gain coefficient of the proportional-integral-differential controller is adjusted in real time according to the vortex field intensity and equivalent temperature parameters; Applying exponential smoothing filtering based on relaxation time to the original control instructions; Adjust the proportional gain coefficient in real time according to the vortex field strength: in: K p : Current proportional gain coefficient; K p0 : Base proportional gain coefficient, determined through vehicle calibration test; ω: real-time vortex field strength; ω0: baseline vortex strength threshold.

[0016] Furthermore, when the vortex intensity exceeds a baseline threshold, the linear amplification of the proportional gain enhances the control system's response sensitivity to sudden disturbances. The relaxation time of the exponential smoothing filter is dynamically adjusted based on the vortex field intensity, reducing the filter intensity in high-disturbance scenarios and avoiding control lag caused by oversmoothing.

[0017] Preferably, the closed-loop update model parameters include: Reversely correct the eddy field viscosity coefficient based on the deviation between the actual acceleration and the control target; The mapping relationship of the equivalent temperature parameters is dynamically adjusted according to the energy consumption error.

[0018] Furthermore, the flow field characteristic parameters are corrected using a gradient optimization method, which backpropagates the deviation between the actual acceleration and the target value to the model parameters. Dynamic adjustment of the equivalent temperature parameter mapping relationship is optimized based on energy consumption error, allowing the model to continuously adapt to macroscopic density changes and microscopic speed fluctuations in traffic flow.

[0019] Preferably, the multimodal sensor data includes: 3D point cloud sequences generated by LiDAR, surface temperature field distribution captured by infrared cameras, and relative velocity vectors measured by millimeter-wave radar; The data were pre-processed with spatiotemporal alignment and noise suppression.

[0020] Furthermore, spatiotemporal alignment utilizes a 3D coordinate transformation model to unify the sensor data into the vehicle's reference coordinate system. Noise suppression is achieved through an adaptive filtering algorithm that corrects motion distortion on the LiDAR point cloud and applies flow field intensity-based non-uniform noise reduction to the infrared thermal imager data.

[0021] A speed control device for an unmanned vehicle, comprising: Sensor module, used to collect multimodal environmental data and implement dynamic optimization collection strategy; A processing module is used to construct a joint control model of the coupled vortex field and Brownian motion parameters and generate optimization instructions; The control module is used to perform speed control according to the dynamic parameter adjustment rules and realize the closed-loop update of the model parameters.

[0022] Furthermore, the processing module integrates parallel computing units for rapid solution of fluid dynamics equations. The control module embeds programmable logic devices, enabling microsecond-level adjustment of control parameters to ensure real-time response in highly dynamic scenarios.

[0023] An unmanned vehicle, comprising: Speed ​​control device; A drive actuator for receiving control instructions and adjusting the vehicle driving torque; an onboard computing unit configured to operate a processing module and a control module of the speed control device; When the unmanned vehicle is used, the speed control method of the unmanned vehicle is implemented.

[0024] Furthermore, the drive actuator employs a torque vectoring strategy, breaking down speed commands into independent control signals for each wheel. The onboard computing unit connects the sensor and controller via a high-speed bus, ensuring data throughput and computational latency meet real-time requirements, enabling millisecond-level closed-loop control of perception, decision-making, and execution.

[0025] In summary, the present invention includes at least one of the following beneficial technical effects: 1. This invention achieves unified modeling of traffic flow micro-perturbations and environmental uncertainties by integrating fluid dynamics vortex field theory with the technical solution of Brownian motion random mapping. Compared to the conventional approach of using fixed-parameter traffic flow models and noise separation, this method effectively addresses the limitation of deterministic models in adapting to the dynamic changes of complex traffic scenarios, enabling unmanned vehicle speed control to respond to sudden disturbances in a physically explainable manner.

[0026] 2. This invention utilizes a dynamic sensing strategy driven by vortex field intensity and a prediction time-domain compression technology solution to achieve an intelligent balance between computing resource allocation and sensing accuracy. Compared to traditional full-range continuous scanning sensor control methods, this overcomes the bottlenecks of sensor data redundancy and computational latency in high-density traffic scenarios, significantly improving system real-time performance in complex scenarios.

[0027] 3. This invention, based on a variational optimization solution coupled with multiple physical fields, establishes a joint optimization framework for vehicle kinematics and traffic flow potential fields. Compared to existing single-target speed tracking control methods, this approach breaks through the simplified handling of conflicting control objectives, enabling speed command generation with multi-dimensional optimization features that balance energy efficiency, ride comfort, and safety.

[0028] 4. This invention establishes a fully closed-loop update mechanism, forming an adaptive control system through real-time data-driven reverse correction of model parameters and dynamic gain adjustment. Compared to traditional open-loop parameter-fixed control strategies, this completely resolves the model mismatch problem caused by the time-varying nature of the environment, ensuring the control robustness and scenario generalization capabilities of the unmanned vehicle during long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0030] The following is combined with Figure 1 , the present invention is described in further detail.

[0031] The present invention provides a speed control method for an unmanned vehicle, comprising: S1, obtaining multimodal sensor data of the vehicle's surrounding environment; S2. Constructing a dynamic vortex field model and a Brownian motion parameter mapping relationship based on the sensor data; S3, dynamically adjust the sensor acquisition strategy according to the vortex field intensity and compress the prediction time domain; S4. Construct the joint governing equations of the coupled vortex field and Brownian motion parameters; S5. Generate speed control instructions through variational optimization solution; S6. Execute control instructions based on dynamic parameter adjustment rules and update model parameters in a closed loop.

[0032] Step S1: Acquire multimodal sensor data of the vehicle's surroundings In this embodiment, the acquisition of multimodal sensor data is achieved through the collaborative acquisition of heterogeneous sensors and spatiotemporal alignment and fusion processing. Specifically, the lidar scans the vehicle's surroundings by emitting a laser beam, generating a three-dimensional point cloud sequence containing spatial coordinate information. The laser beam's scanning pattern is dynamically adjusted based on the complexity of the environment, automatically improving angular resolution in high-interest areas in front of the vehicle. The infrared thermal imaging camera simultaneously captures the surface temperature distribution data of the target object and identifies the heat source characteristics of dynamic obstacles through differences in thermal radiation intensity. The millimeter-wave radar measures the relative velocity vector of surrounding targets based on the Doppler effect and distinguishes between stationary and moving objects using multi-channel echo signal separation technology.

[0033] The sensor data synchronization is achieved through a combination of hardware trigger signals and software timestamp alignment, ensuring that all modal data is collected on the same time basis. For lidar point cloud data, an adaptive filtering algorithm based on motion compensation is used to eliminate point cloud distortion caused by the vehicle's own motion. Specifically, the point cloud coordinates are reversely corrected using the real-time pose data provided by the inertial measurement unit. Infrared thermal imaging data undergoes non-uniform noise reduction processing, retaining the original resolution in areas with significant high-temperature gradients, and performing spatial mean filtering to suppress noise in areas with uniform temperature distribution. Millimeter-wave radar data uses a probability hypothesis density filter to separate valid target echoes from clutter interference, improving the detection stability of low-speed targets.

[0034] The spatiotemporal alignment of multimodal data is achieved through the Lie group SE(3) transformation model, which transforms the observation data in each sensor coordinate system into the body coordinate system with the vehicle's center of mass as the origin. Specifically, an extrinsic calibration matrix is ​​established for each sensor, and the extrinsic parameters are dynamically updated by combining offline calibration with online optimization to eliminate coordinate offset errors caused by mechanical vibration or temperature drift. For lidar point cloud data, the conversion formula is: Pvehicle =M lidar ·P lidar ; Among them, P lidar is the original laser radar point cloud coordinate, M lidar is the laser radar external parameter transformation matrix, P vehicle The converted vehicle coordinate system point cloud is shown in Figure 2. The coordinate system conversion of the millimeter-wave radar velocity vector uses a similar principle, aligning the velocity direction with the vehicle's motion direction through matrix operations.

[0035] During the data fusion phase, a dynamic weighting function is constructed, adjusting the fusion strategy based on sensor type, ambient lighting conditions, and target motion. In strong backlight scenarios, the weight coefficient of infrared thermal imaging data is automatically increased to compensate for the reduced detection performance of the lidar. In rainy and foggy weather, the confidence weight of the millimeter-wave radar is increased to overcome the signal attenuation of the optical sensor. The fusion output is a spatiotemporally aligned multiphysics data cube containing the joint distribution of spatial coordinates, temperature gradients, and velocity vectors, providing a highly consistent input data foundation for subsequent vortex field modeling.

[0036] The method significantly improves the robustness of environmental perception in complex scenarios through the complementary advantages and collaborative optimization of heterogeneous sensors, overcomes the detection blind spots and noise interference problems of a single sensor, and ensures the completeness and timeliness of the data required for the construction of the dynamic eddy field model.

[0037] Step S2: Constructing a dynamic vortex field model and a Brownian motion parameter mapping relationship based on the sensor data In this embodiment, the dynamic vortex field model is constructed based on the modified Navier-Stokes equations. By coupling vehicle dynamics parameters with fluid dynamics properties, a quantitative characterization system for traffic flow micro-disturbance is established. Specifically, the velocity field distribution from lidar point cloud data is integrated with millimeter-wave radar velocity measurement results to calculate the local vortex intensity. The viscosity coefficient is dynamically adjusted to reflect the effect of vehicle inertia on traffic flow.

[0038] Navier-Stokes equation calculation formula: in: The instantaneous acceleration of the traffic flow velocity field reflects the dynamic changes of the vehicle's surrounding environment; Local traffic flow velocity vector; t: time; v: kinematic viscosity coefficient, calculated from fluid density and dynamic viscosity; k: dimensionless proportionality coefficient; m: vehicle mass; I: The moment of inertia of the vehicle about its center of mass.

[0039] The Brownian motion parameter mapping is achieved by thermodynamic analogy, which relates the traffic density ρ to the average speed The product relationship is converted into an equivalent temperature parameter: Among them, kT represents the intensity of random disturbance of traffic flow, k is the analogy of Boltzmann constant, which is obtained through calibration of actual traffic scenes; is the average speed of the local traffic flow, which is calculated by the sliding average of the velocity vector measured by the millimeter-wave radar. The macroscopic density fluctuations and microscopic speed randomness of the traffic flow are converted into physical quantities that can be analyzed by vehicle dynamics.

[0040] In the calculation of the viscosity coefficient, the introduction of vehicle mass m and moment of inertia I enables the model to dynamically reflect the damping effect of vehicle inertia on the surrounding flow field. For example, when the vehicle mass is large or the moment of inertia is high (such as when loaded), the square root relationship of the dynamic viscosity μ automatically increases the weight of the flow field resistance term and suppresses sudden changes in vortex intensity. Random perturbation term The generation of is simulated by the Wiener process, whose variance is proportional to the equivalent temperature parameter kT, thereby mapping the random fluctuation of traffic flow into a quantifiable noise source in the control equation.

[0041] Preferably, the dynamic viscosity is calculated using an online learning mechanism, enabling continuous optimization of model parameters to adapt to different road scenarios based on the deviation between actual acceleration and the control target. Furthermore, the mapping of equivalent temperature parameters is verified using energy conservation constraints, ensuring consistency between the statistical characteristics of random traffic flow disturbances and the vehicle's kinematic response.

[0042] Through the above method, this step realizes the bidirectional coupling modeling of traffic flow dynamic characteristics and vehicle kinematics, providing a joint input of both deterministic trends and random disturbances for subsequent control strategies, laying the physical foundation for vehicle autonomous decision-making.

[0043] Step S3: Dynamically adjust the sensor acquisition strategy based on vortex field intensity and compress the prediction time domain. In this embodiment, the dynamic adjustment of the sensor acquisition strategy is achieved based on the real-time calculation of the vortex field intensity distribution. Through the coordinated optimization mechanism of spatial domain focusing and temporal domain compression, computing resource consumption is reduced while ensuring the acquisition of key information. Specifically, when the vortex field intensity exceeds a preset threshold, the infrared thermal imaging channel is activated to preferentially scan the high-temperature gradient area. At the same time, the scanning range of the lidar is dynamically reduced, and the time domain window of the model prediction control is compressed according to a nonlinear relationship, forming a closed-loop optimization of perception and calculation.

[0044] The vortex field intensity threshold is set according to the vehicle motion state and scene complexity and is adaptively adjusted. The judgment conditions are: Wherein, ω is the current scalar value of the vortex field intensity, which is obtained in real time from the model output of step S2; ω0 is the baseline threshold, which is initialized according to the road type (e.g., urban road, highway); is the modulus of the vehicle’s velocity vector; v ref The physical meaning of this formula is that when the vehicle speed is high or the vortex intensity increases significantly, the system automatically reduces the trigger threshold sensitivity and starts the sensor optimization strategy in advance.

[0045] Wherein, ω is the current scalar value of the vortex field intensity, which is obtained in real time from the model output of step S2; ω0 is the reference threshold, which is initialized according to the road type (such as urban road, highway); is the modulus of the vehicle's velocity vector; v ref The reference speed is set at 80% of the current road speed limit. The physical meaning of this formula is that when the vehicle speed is high or the vortex intensity increases significantly, the system automatically reduces the trigger threshold sensitivity and initiates the sensor optimization strategy in advance.

[0046] The activation of the infrared thermal imaging channel adopts a spatial gradient screening mechanism to preferentially scan the temperature change rate Region ( is the gradient threshold), focusing on dynamic obstacles with significant thermal characteristics such as engine heat sources and brake pads.

[0047] Preferably, the lidar scanning range is dynamically compressed according to the spatial distribution of the vortex field intensity: in the area where ω>ω0, the scanning angle is reduced from the default 120° to 60°, and a fan-shaped scanning window is generated with this area as the center, so that the point cloud data collection density in the key area is increased to twice the original level, while reducing redundant data collection in non-focus areas.

[0048] The compression of the prediction time domain is achieved through a time-varying window function, whose time length T is inversely proportional to the vortex field intensity: Among them, T0 is the initial prediction time domain length, and α is the compression coefficient, which is determined by the vehicle dynamics stability constraint. This design allows the prediction time domain to be automatically shortened in high vortex intensity scenarios (such as traffic conflicts at intersections) to reduce the impact of environmental uncertainty on control commands, while improving the system response speed by increasing the control frequency.

[0049] Preferably, the coordinated execution of sensor strategy adjustment and prediction time domain compression utilizes an event-driven mechanism: when the vortex field intensity triggers a threshold condition, infrared thermal imaging instructions, lidar scanning parameters, and prediction time domain adjustment signals are synchronously generated, achieving microsecond-level response via a hardware interrupt channel. At the data processing level, the compressed prediction time domain window uses a sliding mean filter to perform weighted fusion of historical data to avoid control command oscillation caused by sudden changes in the time domain.

[0050] Through the above method, this step realizes the intelligent allocation of perception resources and dynamic optimization of computing load in complex traffic scenarios, which not only ensures high-precision data collection in areas with strong disturbances, but also reduces the solution complexity of model predictive control through time domain compression, providing stable input conditions for real-time control.

[0051] Step S4: Construct the joint governing equations of the coupled vortex field and Brownian motion parameters In this embodiment, the joint control equation is constructed by establishing a partial differential coupling relationship between vortex field intensity and Brownian motion parameters. This multi-physics fusion integrates vehicle dynamics, random traffic flow disturbances, and sensor perception data to form a control model that combines deterministic trends with random noise descriptions. Specifically, based on the dynamic vortex field intensity distribution and equivalent temperature parameters generated in step S2, a partial differential equation is established that incorporates the interaction between vehicle yaw motion and flow field evolution. A multi-objective optimization function is then used to coordinate the smoothness of control instructions with energy efficiency.

[0052] The partial differential coupling relationship is expressed as: in: The rate of change of vortex field intensity per unit time; ω: scalar value of eddy field strength; The real-time yaw rate of the vehicle; kT: traffic density ρ and average speed Generated equivalent temperature parameters; m: vehicle mass; Local traffic flow velocity vector; x: vehicle longitudinal coordinate; The instantaneous acceleration of the traffic flow velocity field.

[0053] The construction of the multi-objective optimization function includes the acceleration smoothing term and the energy consumption gradient term, and its weight coefficient is adaptively adjusted according to the vortex field intensity. The specific expression is: Among them, λ1 is the weight of the acceleration smoothing term, which is inversely proportional to the vortex intensity ω, allowing more drastic acceleration changes in strong vortex disturbance scenarios to avoid risks; ∫ is the time integration operator; λ2 is the weight of the energy consumption gradient term, which is proportional to ω and promotes the optimization of energy efficiency in high disturbance environments; is the vehicle acceleration vector; is the derivative of the acceleration vector with respect to time, which indicates the rate of change of acceleration; dt is the discrete time step of the control system; It is the energy consumption gradient field, which is generated by jointly calculating the motor torque-speed characteristic curve and the battery discharge model.

[0054] Preferably, the discretized solution of the partial differential equation utilizes the finite volume method, which divides the space surrounding the vehicle into a dynamically updated non-uniform grid. The grid nodes are automatically refined in areas with significant vortex intensity gradients (such as the vehicle's lateral blind spots) to improve computational accuracy. Real-time calculation of the coupling coefficient is achieved through a hardware acceleration unit, ensuring that the equation is iteratively solved within a millisecond time window.

[0055] This step incorporates vehicle kinematics, traffic flow fluid dynamics, and energy consumption characteristics into a unified control framework, breaking through the limitations of isolated treatment of each physical quantity in traditional control models and providing a high-fidelity mathematical description basis for variational optimization solutions.

[0056] Step S5: Generate speed control instructions through variational optimization solution In this embodiment, the variational optimization solution is implemented based on the Hamiltonian action principle. The joint control equation constructed in step S4 is transformed into an energy-optimal control problem. Speed ​​control instructions that satisfy multiple objective constraints are generated by dynamically adjusting optimization weights and using a discretized solution strategy. Specifically, by defining an action functional that includes vehicle kinetic energy, vortex field potential energy, and energy consumption gradients, and combining it with real-time updated sensor data and model parameters, an optimal speed trajectory that satisfies boundary conditions and dynamic constraints is determined.

[0057] Embed the stochastic differential equation into the Hamiltonian action principle: in: δ: variational operator, representing a small change to the action functional, used to solve the extreme path of the vehicle's motion trajectory; ∫: time integration operator; m: vehicle mass; Vehicle velocity vector; vortex field potential energy integral; Spatial location The vortex field strength at ; The volume of small spatial regions in the vehicle's surroundings; dt: discrete time step of the control system; Three-dimensional space vector; t: time.

[0058] The construction of the Hamiltonian action functional deeply integrates the vehicle's motion state with the characteristics of the traffic flow environment. Its expression includes the vehicle's kinetic energy term and the vortex field potential energy integral term. The kinetic energy term, consisting of the product of the vehicle's mass and the square of the velocity vector modulus, reflects the energy consumption of the vehicle's own motion state. The potential energy integral term quantifies the potential impact of traffic flow disturbances on vehicle motion by calculating the volume integral of the vortex field intensity in the three-dimensional space surrounding the vehicle. The optimization goal is to find a speed control sequence that minimizes the action functional while satisfying acceleration smoothness, energy efficiency, and safety distance constraints.

[0059] An adaptive discretization strategy is employed during the solution process, dividing the continuous time domain into dynamically adjusted time step intervals. When a sudden change in vortex field intensity or a significant change in the equivalent temperature parameter is detected, the time step is automatically reduced to improve optimization accuracy in the local region; during stable flow periods, the step size is increased to enhance computational efficiency. The optimization variables at each discrete time node include longitudinal velocity, lateral velocity, and yaw rate. Acceleration constraints, road boundary conditions, and following distance safety limits are introduced using the Lagrange product method.

[0060] Adjust the proportional gain coefficient in real time according to the vortex field strength: in: K p : Current proportional gain coefficient; K p0 : Base proportional gain coefficient, determined through vehicle calibration test; ω: real-time vortex field strength; ω0: baseline vortex strength threshold.

[0061] Preferably, the variational optimization problem is solved iteratively using the conjugate gradient method, using the optimization results of the previous time window to initialize the initial value of the current iteration, reducing computational complexity. For the weight distribution of the acceleration smoothing term and the energy consumption gradient term in the multi-objective optimization function, a nonlinear mapping relationship with the vortex field intensity is established: when the vortex field intensity exceeds the threshold, the weight coefficient of the energy consumption gradient term increases exponentially, prompting the control system to prioritize energy efficiency in strong disturbance scenarios; in low vortex intensity areas, the weight of the acceleration smoothing term dominates, ensuring ride smoothness.

[0062] Subsequent processing of the optimization results involves exponential smoothing filtering with adaptive relaxation time, dynamically adjusting the filter time constant based on the vortex field strength. In high-frequency disturbance regions, a smaller time constant is used to maintain the rapid response of the control command; in stable flow regions, a larger time constant is used to suppress command jitter caused by sensor noise. The resulting speed control command, containing speed amplitude, azimuth, and acceleration limits, is transmitted to the vehicle's drive actuator via the CAN bus.

[0063] Through the above method, this step realizes the real-time solution of the multi-physics field coupling control model in complex traffic environments, generates speed control instructions that take into account safety, comfort and energy efficiency, and provides an accurate action planning basis for vehicle autonomous decision-making.

[0064] Step S6: Execute control instructions and update model parameters in a closed-loop based on dynamic parameter adjustment rules. In this embodiment, the execution of control instructions and the closed-loop update of model parameters are achieved through dynamic gain adjustment and a data-driven reverse optimization mechanism, ensuring the real-time performance and environmental adaptability of the control system. Specifically, the speed control instruction generated in step S5 is input to the vehicle drive actuator. Simultaneously, the control parameters are dynamically adjusted based on the deviation between the actual acceleration and the target value. The key coefficients of the mapping between the vortex field model and Brownian motion are reversely corrected using an online learning algorithm.

[0065] The core of the dynamic parameter adjustment rule lies in the design of the proportional gain's vortex intensity dependency. When the vortex field intensity exceeds a baseline threshold, the proportional gain coefficient is linearly amplified with the intensity value, enhancing the system's responsiveness to sudden disturbances. When the vortex field intensity is within a low-fluctuation range, the gain coefficient automatically decays to prevent control oscillations caused by over-response. An inertial hysteresis element is introduced into the gain adjustment process to prevent frequent parameter jumps caused by sensor noise or transient vortex field fluctuations, thereby improving control stability.

[0066] During the execution phase of the control command, an exponential smoothing filter with adaptive relaxation time is applied. The filter time constant is dynamically adjusted based on the spatial gradient of the vortex field intensity. In high-intensity gradient regions (such as disturbances caused by obstacles close to the vehicle), a smaller time constant is used to preserve the high-frequency response characteristics of the command; in low-gradient regions, the time constant is increased to suppress command jitter caused by sensor noise and model errors. The filtered control command is then distributed to each drive motor through a torque distribution strategy, achieving coordinated control of longitudinal velocity tracking and lateral attitude stabilization.

[0067] The closed-loop update of model parameters adopts a dual-channel optimization mechanism: for the dynamic vortex field model, based on the deviation between the actual acceleration and the target acceleration, the gradient direction of the viscosity coefficient and the vehicle dynamics parameters is solved through the adjoint equation, and the proportional coefficient is iteratively updated using the momentum gradient descent method; for the Brownian motion mapping relationship, by monitoring the cumulative error between the actual energy consumption and the predicted value, the density-velocity coupling coefficient in the equivalent temperature parameter is reversely adjusted to make the statistical characteristics of the random disturbance of the traffic flow consistent with the actual motion response of the vehicle.

[0068] Preferably, the closed-loop update process incorporates a sliding time window mechanism, retaining only sensor data and vehicle status information within the most recent time window for parameter optimization, thus avoiding model overfitting caused by the accumulation of historical data. Updated model parameters are written to the controller's non-volatile memory in real time via a hardware acceleration unit, ensuring that the latest calibration results are retained even after power outages. For key parameters (such as the dynamic viscosity scaling factor), physical constraints are set to prevent non-physical numerical drift during the optimization process.

[0069] Through the above method, this step realizes high-precision execution of control instructions and continuous self-correction of model parameters, forming a fully closed-loop control chain of "perception-decision-execution-verification", effectively overcoming the defects of model mismatch and parameter drift in traditional open-loop control systems.

[0070] The speed control device for an unmanned vehicle described below and the speed control method for an unmanned vehicle described above may refer to each other.

[0071] The present invention also provides a speed control device for an unmanned vehicle, comprising: Sensor module, used to collect multimodal environmental data and implement dynamic optimization collection strategy; A processing module is used to construct a joint control model of the coupled vortex field and Brownian motion parameters and generate optimization instructions; The control module is used to perform speed control according to the dynamic parameter adjustment rules and realize closed-loop update of model parameters. The device of this embodiment can be used to execute the above method embodiment. Its principles and technical effects are similar and will not be repeated here.

[0072] The present invention also provides an unmanned vehicle, comprising: Speed ​​control device; A drive actuator for receiving control instructions and adjusting the vehicle driving torque; an onboard computing unit configured to operate a processing module and a control module of the speed control device; When the unmanned vehicle is used, the speed control method of the unmanned vehicle is implemented.

[0073] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A speed control method for an unmanned vehicle, characterized in that: include: Acquire multimodal sensor data of the vehicle's surroundings; Constructing a dynamic vortex field model and a Brownian motion parameter mapping relationship based on the sensor data; Dynamically adjust sensor acquisition strategies and compress prediction time domain according to eddy field intensity; Construct the joint governing equations of the coupled vortex field and Brownian motion parameters; Generate speed control instructions through variational optimization solution; Control instructions are executed based on dynamic parameter adjustment rules and model parameters are updated in a closed loop.

2. The speed control method of an unmanned vehicle according to claim 1, characterized in that: Described constructing dynamic vortex field model comprises: Calculate the local vorticity distribution based on the modified Navier-Stokes equations, where the viscosity coefficient is related to the vehicle mass and moment of inertia; Mapping traffic density to equivalent temperature parameters, constructing vehicle dynamics equations including random disturbance terms; Navier-Stokes equation calculation formula: in: The instantaneous acceleration of the traffic flow velocity field reflects the dynamic changes of the vehicle's surrounding environment; Local traffic flow velocity vector; t: time; v: kinematic viscosity coefficient, calculated from fluid density and dynamic viscosity; k: dimensionless proportionality coefficient; m: vehicle mass; I: The moment of inertia of the vehicle about its center of mass.

3. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The dynamic adjustment of the sensor acquisition strategy includes: When the vortex intensity exceeds a preset threshold, the infrared thermal imaging channel is activated and focuses on the high-temperature gradient area; The lidar scanning area is dynamically reduced according to the temperature gradient distribution, and the time domain range of the model prediction control is simultaneously compressed.

4. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The constructing of the joint control equation includes: A partial differential coupling relationship between the vortex field intensity and the Brownian motion parameters is established, where the coupling coefficient is dynamically correlated with the vehicle's yaw rate. A multi-objective optimization function including acceleration smoothing term and energy consumption gradient term is defined, and its weight coefficient is adaptively adjusted with the vortex field intensity. Partial differential coupling relationship formula: in: The rate of change of vortex field intensity per unit time; ω: scalar value of eddy field strength; The real-time yaw rate of the vehicle; kT: traffic density ρ and average speed Generated equivalent temperature parameters; m: vehicle mass; Local traffic flow velocity vector; x: vehicle longitudinal coordinate; The instantaneous acceleration of the traffic flow velocity field.

5. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The variational optimization solution includes: Embed the stochastic differential equation into the Hamiltonian action principle: in: δ: variational operator, representing a small change to the action functional, used to solve the extreme path of the vehicle's motion trajectory; ∫: time integration operator; m: vehicle mass; Vehicle velocity vector; vortex field potential energy integral; Spatial location The vortex field strength at ; The volume of small spatial regions in the vehicle's surroundings; dt: discrete time step of the control system; Three-dimensional space vector; t: time; The optimization problem is solved by a numerical discretization method with adaptive step size, and the step size is dynamically switched according to the modified Reynolds number.

6. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The execution control instructions include: The gain coefficient of the proportional-integral-differential controller is adjusted in real time according to the vortex field intensity and equivalent temperature parameters; Applying exponential smoothing filtering based on relaxation time to the original control instructions; Adjust the proportional gain coefficient in real time according to the vortex field strength: in: K p : Current proportional gain coefficient; K p0 : Base proportional gain coefficient, determined through vehicle calibration test; ω: real-time vortex field strength; ω0: baseline vortex strength threshold.

7. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The closed-loop update model parameters include: Reversely correct the eddy field viscosity coefficient based on the deviation between the actual acceleration and the control target; The mapping relationship of the equivalent temperature parameters is dynamically adjusted according to the energy consumption error.

8. The speed control method of an unmanned vehicle according to claim 1, characterized in that: The multimodal sensor data includes: 3D point cloud sequences generated by LiDAR, surface temperature field distribution captured by infrared cameras, and relative velocity vectors measured by millimeter-wave radar; The data were pre-processed with spatiotemporal alignment and noise suppression.

9. A speed control device for an unmanned vehicle, applied to a speed control method for an unmanned vehicle according to any one of claims 1 to 8, characterized in that: include: Sensor module, used to collect multimodal environmental data and implement dynamic optimization collection strategy; A processing module is used to construct a joint control model of the coupled vortex field and Brownian motion parameters and generate optimization instructions; The control module is used to perform speed control according to the dynamic parameter adjustment rules and realize the closed-loop update of the model parameters.

10. An unmanned vehicle, characterized in that: include: Speed ​​control device; A drive actuator for receiving control instructions and adjusting the vehicle driving torque; an onboard computing unit configured to operate a processing module and a control module of the speed control device; When the unmanned vehicle is used, the speed control method of the unmanned vehicle as described in any one of claims 1 to 8 is implemented.

Images