Unmanned vehicle multi-mode adaptive control method and system, terminal and medium

Abstract

The invention relates to the field of unmanned automatic control, and particularly discloses an unmanned vehicle multi-mode self-adaptive control method and system, a terminal and a medium, and the method comprises the steps: obtaining a target parameter for multi-mode self-adaptive control; based on the target parameters, working condition identification is carried out through a preset rule engine; determining a corresponding transverse and longitudinal control algorithm combination based on the current working condition identification result according to a mapping relation between a preset working condition type and the transverse and longitudinal control algorithm combination; and generating a control instruction based on the current transverse and longitudinal control algorithm combination and sending the control instruction to an execution mechanism. By matching the most suitable control strategy for different driving stages, the operation efficiency and the task precision are balanced, and the whole-course average time of single operation is effectively shortened.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of unmanned automatic control, in particular to a multi-modal adaptive control method, system, terminal and medium for unmanned vehicles. BACKGROUND

[0002] Unmanned logistics vehicles are increasingly widely used in closed or semi-closed scenarios such as parks and factory areas. Such scenarios have characteristics such as complex road conditions, many dynamic obstacles, and high requirements for work processes. The motion control of park unmanned logistics vehicles often uses a single control algorithm, such as a single linear quadratic regulator (LQR), a proportional-integral-derivative (PID) controller, or a model predictive control (MPC). Such single control strategies are often optimized for a certain specific working condition when designed, and although they can achieve good performance in that working condition, they have obvious shortcomings in adaptability, efficiency, and safety in the complex and variable scenarios of the park. They cannot balance efficiency and accuracy, have poor scene adaptability, and lack sufficient response in emergency working conditions, and the system robustness needs to be improved. For example, a control algorithm designed to ensure stability at medium and high speeds often produces a large steady-state error when the vehicle is performing low-speed precise parking, making it difficult to meet the high-precision pose requirements of automated loading and unloading. Conversely, an algorithm optimized for low-speed high-precision control may lack stability at medium and high speeds due to slow response or overshoot, affecting operational efficiency, and related technologies lack a control method that can identify and adjust itself according to real-time environmental characteristics. SUMMARY

[0003] To solve the above problems, the present application provides a multi-modal adaptive control method, system, terminal and medium for unmanned vehicles, which balances operational efficiency and task accuracy by matching the most suitable control strategy for different driving stages, effectively shortening the average time of a single operation.

[0004] In a first aspect, the technical solution of the present application provides a multi-modal adaptive control method for unmanned vehicles, comprising the following steps: Obtaining target parameters for multi-modal adaptive control; Based on the target parameters, identifying the working condition through a pre-set rule engine; According to the mapping relationship between the pre-set working condition type and the combination of lateral and longitudinal control algorithms, determine the corresponding combination of lateral and longitudinal control algorithms based on the current working condition identification result; Based on the current combination of lateral and longitudinal control algorithms, generate a control instruction and send it to the actuator.

[0005] In an optional embodiment, based on the target parameters, the working condition is identified through a pre-set rule engine, specifically including: Determine whether the emergency arbitrary speed braking working condition trigger condition is met based on the collision time and its rate of change; If yes, the working condition is identified as emergency arbitrary speed braking working condition; If no, it is determined whether the position of the ego vehicle and the position of the target stop point are less than or equal to a preset distance; If yes, the working condition is identified as high-precision extremely low-speed parking working condition; If no, it is determined whether the first future path of the ego vehicle is a curve or a narrow lane according to the path planning data; If yes, the working condition is identified as multi-obstacle medium-low speed passing working condition; If no, a rectangular perception area is defined, the obstacle density in the perception area is calculated, if the obstacle density is 0, the working condition is identified as no-obstacle medium-high speed cruising working condition, if the obstacle density is not 0 and does not exceed a preset density threshold, it is determined whether the safety driving condition is met according to the target future path of the dynamic obstacle and the second future path of the corresponding ego vehicle, if yes, the working condition is also identified as no-obstacle medium-high speed cruising working condition, if no or the obstacle density exceeds the preset density threshold, the working condition is also identified as multi-obstacle medium-low speed passing working condition.

[0006] In an optional embodiment, it is determined whether the emergency arbitrary speed braking working condition triggering condition is met based on the collision time and its rate of change, specifically including: If the collision time is less than a preset collision time threshold and the absolute value of the rate of change of the collision time is greater than an emergency threshold, it is determined that the emergency arbitrary speed braking working condition triggering condition is met.

[0007] In an optional embodiment, it is determined whether the first future path of the ego vehicle is a curve or a narrow lane according to the path planning data, specifically including: Discrete point sets of the future target length path are extracted from the path planning data; The average curvature and the total amount of change in the heading angle of the path are calculated according to the coordinates of all points in the discrete point sets; If the average curvature exceeds a curvature threshold or the total amount of change in the heading angle exceeds an angle threshold, it is determined that it is a curve; A local occupancy grid map corresponding to the future target length path is constructed through laser radar scanning data, the left and right passable widths of the vehicle are calculated according to the grid map, and if the sum of the left and right passable widths is less than a width threshold, it is determined that it is a narrow lane.

[0008] In an optional embodiment, it is determined whether the safety driving condition is met according to the target future path of the dynamic obstacle and the second future path of the corresponding ego vehicle, specifically including: The Kalman filter is used for each dynamic obstacle to predict the trajectory in the future target time length, and the closest point distance and arrival time of each dynamic obstacle to the second future path of the corresponding ego vehicle are calculated, if the closest point distance is greater than the preset distance threshold and the arrival time is greater than the preset time length, it is determined that the safe driving condition is met, otherwise it is determined that the safe driving condition is not met.

[0009] In an optional embodiment, the lateral and longitudinal control algorithm combination corresponding to the high-speed cruise condition without obstacles is a lateral LQR control algorithm and a longitudinal PID control algorithm. The lateral and longitudinal control algorithm combination corresponding to the low-speed passing condition with multiple obstacles is a lateral Stanley control algorithm and a parameter-adjustable longitudinal PID control algorithm. The lateral and longitudinal control algorithm combination corresponding to the high-precision extremely low-speed parking condition is a lateral and longitudinal cooperative MPC control algorithm. The lateral and longitudinal control algorithm combination corresponding to the emergency braking condition at any speed is a lateral SMC control algorithm and a longitudinal logic threshold control algorithm.

[0010] In an optional embodiment, a weighted average algorithm is used to smoothly transition the control instructions during condition switching, specifically including: A fixed transition time window T_transition is defined, and the switching time is denoted as t_switch, then in the time interval t∈[t_switch,t_switch+T_transition], the control instruction u_final(t) output to the actuator is represented as: u_final(t)=α(t)*u_A(t)+(1-α(t))*u_B(t) In the formula, u_A(t) is the control instruction generated before switching, and u_B(t) is the control instruction generated after switching. α(t) is a function that decays with time, and is represented as α(t)=1-((t-t_switch) / T_transition). When t>t_switch+T_transition, α(t) is 0.

[0011] In a second aspect, the technical scheme of the present application provides a multi-modal adaptive control system for unmanned vehicles, comprising: A target parameter acquisition unit is configured to acquire target parameters for multi-modal adaptive control. A condition recognition unit is configured to recognize a condition based on the target parameters through a preset rule engine. A control algorithm determination module is configured to determine a corresponding lateral and longitudinal control algorithm combination based on the current condition recognition result according to a mapping relationship between the preset condition type and the lateral and longitudinal control algorithm combination. The control instruction generation transmission module is configured to generate control instructions based on the current combination of the lateral and longitudinal control algorithms and send the control instructions to the actuator.

[0012] In a third aspect, the present application provides a terminal, comprising: a memory configured to store a multi-modal adaptive control program for an unmanned vehicle; a processor configured to execute the multi-modal adaptive control program for the unmanned vehicle and implement the steps of the multi-modal adaptive control method for the unmanned vehicle according to any one of the preceding aspects.

[0013] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the storage medium stores a multi-modal adaptive control program for an unmanned vehicle, and the multi-modal adaptive control program, when executed by a processor, implements the steps of the multi-modal adaptive control method for the unmanned vehicle according to any one of the preceding aspects.

[0014] As can be seen from the above technical solutions, the present application has the following advantages: By using the pre-set rule engine, the current working condition can be accurately identified in real time based on the target parameters, and the lateral and longitudinal control algorithm combination specially optimized for the working condition can be automatically mapped and switched to, thereby fundamentally solving the problem that a single algorithm cannot maintain optimal performance in all scenarios. Compared with the traditional single LQR or PID control, the present application can take advantage of the corresponding algorithm in different working conditions, for example, the average error of path tracking can be significantly reduced in a curve or multi-obstacle scenario. By matching the most suitable control strategy for different driving stages, the vehicle can cruise stably at high speed in an obstacle-free scenario, pass safely and cautiously in a complex scenario, and be precisely positioned in a parking stage, thereby achieving a balance between operating efficiency and task accuracy, and effectively shortening the average time of the whole process of a single operation. Further, by identifying and processing the highest priority emergency braking working condition, the danger can be quickly determined based on the time to collision (TTC) and its change rate, and the combination of the lateral sliding mode control (SMC) and the longitudinal logic threshold control can be switched to instantaneously, which is more responsive and more decisive, thereby ensuring that obstacle avoidance and braking can be achieved quickly and smoothly in an emergency, and the safety of people and vehicles is maximized. Preferably, the weighted average algorithm is used to smooth the transition of the control instructions at the switching moment, thereby effectively avoiding the sudden change of the longitudinal or lateral dynamics of the vehicle caused by the jump of the control mode, and significantly improving the driving comfort and the stability of the overall control system. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort.

[0016] Figure 1 A flowchart of a multi-modal adaptive control method of an unmanned vehicle is provided for an embodiment of the present application.

[0017] Figure 2 A flowchart of a working condition identification process is provided.

[0018] Figure 3 A structure schematic block diagram of a multi-modal adaptive control system of an unmanned vehicle is provided for an embodiment of the present application.

[0019] Figure 4 A structure schematic diagram of a terminal is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the application purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions protected by the present application will be described clearly and completely by using specific embodiments and drawings. Obviously, the following described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort are within the scope of protection of the present application.

[0021] Unless otherwise defined, all technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application.

[0022] Figure 1 A flowchart of a multi-modal adaptive control method of an unmanned vehicle is provided for an embodiment of the present application. Wherein, Figure 1 The execution subject can be a multi-modal adaptive control system of an unmanned vehicle. The multi-modal adaptive control method of an unmanned vehicle provided by the embodiment of the present application is executed by a computer device, and accordingly, the multi-modal adaptive control system of an unmanned vehicle runs in the computer device. According to different needs, the order of steps in the flowchart can be changed, and some can be omitted.

[0023] As Figure 1 shown, the method comprises the following steps.

[0024] S1, obtaining target parameters for multi-modal adaptive control.

[0025] S2 identifies operating conditions based on target parameters using a pre-defined rule engine.

[0026] S3, based on the mapping relationship between the preset working condition type and the combination of horizontal and vertical control algorithms, determines the corresponding combination of horizontal and vertical control algorithms based on the current working condition identification result.

[0027] S4 generates control commands based on the current combination of horizontal and vertical control algorithms and sends them to the actuator.

[0028] In some alternative implementations, necessary input information is acquired through a sensor sensing layer. This mainly includes: Self-state sensors: including IMU (Inertial Measurement Unit) and wheel speed encoder, used to acquire real-time yaw rate, longitudinal acceleration, vehicle speed, etc.

[0029] Environmental perception sensors include LiDAR, millimeter-wave radar, and cameras. LiDAR is used to generate high-precision point cloud maps to detect the outline, position, and density of obstacles; millimeter-wave radar is used to accurately measure relative speed and distance to obstacles; and cameras are used to assist in the identification of lane lines, special signs, and visual labels for stopping points.

[0030] Positioning and pose sensors: including integrated navigation system (GNSS+RTK+IMU) to provide vehicle centimeter-level high-precision global pose (X, Y, Z coordinates, roll angle, pitch angle, yaw angle).

[0031] In some optional implementations, the autonomous driving domain controller identifies the operating condition based on target parameters using a preset rule engine. Then, based on the mapping relationship between preset operating condition types and combinations of lateral and longitudinal control algorithms, it determines the corresponding combination of lateral and longitudinal control algorithms according to the current operating condition identification result. Specifically, this includes an operating condition identifier and a switching manager, used for operating condition identification and operating condition switching management, respectively.

[0032] The pre-configured operating conditions include barrier-free medium-high speed cruise, multi-obstacle medium-low speed passage, high-precision ultra-low speed parking, and emergency arbitrary speed braking. When the operating conditions are identified through the preset rule engine, the priority of the operating conditions is determined from high to low as follows: emergency arbitrary speed braking, high-precision ultra-low speed parking, multi-obstacle medium-low speed passage, and barrier-free medium-high speed cruise.

[0033] like Figure 2 As shown, based on the working condition type and judgment priority, step S2 identifies the working condition based on the target parameters through a preset rule engine, including the following steps.

[0034] S201, determining whether the emergency arbitrary speed braking condition triggering condition is met based on the TTC and the rate of change of the TTC; if yes, the working condition is identified as the emergency arbitrary speed braking condition; otherwise, the next step is performed.

[0035] If the TTC is less than a preset TTC threshold and the absolute value of the rate of change of the TTC is greater than an emergency threshold, it is determined that the emergency arbitrary speed braking condition triggering condition is met. Specifically, this step is the highest priority judgment, which continuously detects the "sudden cut-in" behavior based on the TTC (TTC = relative distance / relative speed) and the rate of change of the TTC d(TTC) / dt. When an obstacle simultaneously meets the conditions of TTC less than a threshold of 5 seconds and the absolute value of d(TTC) / dt greater than an emergency threshold of 2.0, it is considered that "sudden cut-in" occurs, and the emergency braking condition is triggered immediately.

[0036] S202, determining whether the position of the vehicle and the position of the target stop point are less than or equal to a preset distance; if yes, the working condition is identified as the high-precision extremely low-speed parking condition; otherwise, the next step is performed.

[0037] If there is no emergency detected in step S201, the Euclidean distance d = sqrt((x_veh-x_target)²+(y_veh-y_target)²) is calculated based on the position of the vehicle (x_veh, y_veh) and the position of the target stop point (x_target, y_target) obtained by the navigation system, and it is determined whether the distance d meets d≤10 meters; if yes, the precise parking condition is entered.

[0038] S203, determining whether the first future path of the vehicle is a curve or a narrow lane according to the path planning data; if yes, the working condition is identified as the multi-obstacle low-speed passing condition; otherwise, the next step is performed.

[0039] After excluding the emergency and the need for parking, the road geometry condition is determined first, that is, the road geometry characteristics are analyzed for curve and narrow lane determination.

[0040] Specifically, the curve determination process is as follows: the discrete point set of the future target length path is extracted from the path planning data; the average curvature and the total amount of change of the heading angle of the path are calculated based on the coordinates of all points in the discrete point set; if the average curvature exceeds the curvature threshold or the total amount of change of the heading angle exceeds the angle threshold, it is determined that it is a curve. For example, the discrete point set (x_i, y_i) of the future 10-meter path is extracted from the global path or local path planner, the average curvature or the total amount of change of the heading angle Δθ of the path is calculated, and if the average curvature exceeds the threshold κ_threshold or Δθ exceeds the angle threshold, it is determined that it is a curve.

[0041] The narrow channel process is: constructing a local occupancy grid map corresponding to the future target length path through laser radar scanning data, calculating the left and right passable widths of the vehicle according to the grid map, and if the sum of the left and right passable widths is less than the width threshold, it is determined to be a narrow lane. For example, the left and right passable widths of the vehicle W_left and W_right, if W_left+W_right<W_narrow_threshold (vehicle width+1.0m), it is determined to be a narrow channel.

[0042] It should be noted that both the bend and the narrow channel are met, and any condition is met to enter the multi-obstacle low-speed passing working condition.

[0043] S204, define a rectangular perception area, calculate the obstacle density in the perception area, if the obstacle density is 0, the working condition recognition is no-obstacle medium-speed cruising working condition, if the obstacle density is not 0 and does not exceed the preset density threshold, whether the safe driving condition is met is judged according to the target future path of the dynamic obstacle and the second future path corresponding to the ego vehicle, if it is met, the working condition recognition is also identified as no-obstacle medium-speed cruising working condition, if it is not met or the obstacle density exceeds the preset density threshold, the working condition recognition is also identified as multi-obstacle low-speed passing working condition.

[0044] The working condition recognizer is centered on the ego vehicle, and a rectangular perception area with variable length and width is defined, wherein the area length L_zone is the lookahead distance set by the algorithm (a variable proportional to the vehicle speed), and the area width W_zone is three times the vehicle width. It should be noted that in one embodiment, after the working condition recognizer receives the fusion data from the sensor perception layer, it first defines the rectangular perception area, and then performs multiple basic calculations in parallel, including calculating the total number of obstacles N_obstacles in the perception area based on laser radar point cloud data through the Euclidean clustering algorithm, calculating the Euclidean distance d=sqrt((x_veh-x_target)²+(y_veh-y_target)²) between the ego vehicle position (x_veh, y_veh) and the target parking point position (x_target, y_target) obtained through the integrated navigation system, and analyzing the road geometric features for bend and narrow channel judgment.

[0045] If the first future path through the bend and narrow road conditions is determined to be not a multi-obstacle low-speed passing working condition, the vehicle enters a no-obstacle medium-speed cruising working condition determination link, that is, the obstacle density ρ=N_obstacles / A_zone (where A_zone is the sensing area) is calculated. When ρ=0, it is determined to be a no-obstacle medium-speed cruising working condition. When ρ is not zero but does not exceed the preset density threshold, it is determined whether the safe driving condition is met according to the target future path of the dynamic obstacle and the second future path of the corresponding ego vehicle. If it is met, the working condition recognition is also recognized as a no-obstacle medium-speed cruising working condition. If it is not met or the obstacle density exceeds the preset density threshold, the working condition recognition is also recognized as a multi-obstacle low-speed passing working condition.

[0046] In one embodiment, whether the safe driving condition is met according to the target future path of the dynamic obstacle and the second future path of the corresponding ego vehicle specifically includes: using Kalman filtering to predict the trajectory of each dynamic obstacle within the future target time, and calculating the closest point distance and arrival time of the second future path of the corresponding ego vehicle. If the closest point distance is greater than the preset distance threshold and the arrival time is greater than the preset time, it is determined that the safe driving condition is met, otherwise it is determined that the safe driving condition is not met.

[0047] For example, Kalman filtering is used to predict the trajectory of each dynamic obstacle within 3-5 seconds, and the closest point distance DCPA and the arrival time TCPA of the ego vehicle path are calculated. If all dynamic obstacles meet the safety condition of DCPA>2 meters and TCPA>5 seconds, it is also determined to be a no-obstacle medium-speed cruising working condition. Conversely, if ρ exceeds the density threshold or there is a dynamic obstacle that does not meet the safety condition, it is determined to be a multi-obstacle low-speed passing working condition.

[0048] It should be noted that the default planned path is known, and path planning can be performed by using existing path planning algorithms. For example, a vehicle model is responsible for updating the vehicle state at each time, providing necessary state input for path planning. The path planning algorithm performs real-time path planning according to obstacle information (obstacle x, y position), high-precision map information (target point x, y position), positioning information (ego vehicle x, y position), and state data provided by the vehicle model. Details are not repeated here.

[0049] It should be further noted that the working condition recognizer is only responsible for high-level mode decision, rather than specific obstacle avoidance path generation. Specifically, when it is determined that the "no-obstacle high-speed cruising working condition", it means that the static obstacles in the current environment and the dynamic obstacles confirmed as having no collision risk (satisfying the conditions of DCPA>2 meters and TCPA>5 seconds) after trajectory prediction and risk assessment have been classified by the recognizer as obstacles of a type that can be handled routinely. The specific avoidance strategies for these obstacles, including trajectory fine-tuning, smooth detouring, or maintaining a safe following distance, are all systematically summarized and delegated to the local path planner, which is responsible for implementation. This architecture clearly separates the functional boundaries of the decision layer and the execution layer: the working condition recognizer serves as the decision core to determine "what working condition", and the local path planner serves as the execution unit to solve "how to avoid obstacles", thereby optimizing the system architecture, improving overall decision efficiency, and ensuring that the planning module can focus on the fine implementation of spatial obstacle avoidance logic.

[0050] In this embodiment, different working condition types correspond to different combinations of longitudinal and lateral control algorithms. If the working condition types of the two consecutive cycles are different, the switching manager sends the corresponding mode switching instructions to the multi-mode controller according to the working condition determination results output by the working condition recognizer. To ensure vehicle stability during the control mode switching process and avoid step jumps in control variables such as steering wheel angle, acceleration / braking force, etc., the manager uses a weighted average algorithm to smoothly transition the control instructions at the switching instant. The specific implementation process is as follows: Define a fixed transition time window T_transition, and let t_switch be the switching time. Within the time interval t∈[t_switch,t_switch+T_transition], the control instruction u_final(t) output to the actuator is represented as: u_final(t)=α(t)*u_A(t)+(1-α(t))*u_B(t) where u_A(t) is the control instruction generated by the pre-switching working condition, and u_B(t) is the control instruction generated by the post-switching working condition.

[0051] α(t) is a function that decays over time, i.e., a function that linearly decays from 1 to 0 over time, represented as α(t)=1-((t-t_switch) / T_transition). When t>t_switch+T_transition, α(t) is 0, at which point the control instruction u_B(t) of the post-switching working condition is completely adopted, and the smooth transition phase ends.

[0052] Wherein, the transition time window T transition can be 5 times of the cycle period. Through the above linear weighted fusion, the continuous and smooth change of the control instruction in the transition period is realized, the sudden change of the vehicle longitudinal or lateral dynamic response caused by the direct switching of the control mode is effectively avoided, and the driving comfort and the stability of the control system are significantly improved.

[0053] In some optional embodiments, the multi-mode controller presets four control algorithm combinations, and can realize accurate mode switching and control according to the output of the working condition identifier. The corresponding lateral and longitudinal control algorithm combination for the obstacle-free medium-high speed cruising working condition (15-25km / h) is the lateral LQR control algorithm and the longitudinal PID control algorithm; the corresponding lateral and longitudinal control algorithm combination for the multi-obstacle medium-low speed passing working condition (10-15km / h) is the lateral Stanley control algorithm and the parameter-adjustable longitudinal PID control algorithm; the corresponding lateral and longitudinal control algorithm combination for the high-precision extremely low-speed parking working condition (≤5km / h) is the lateral and longitudinal cooperative MPC control algorithm; and the corresponding lateral and longitudinal control algorithm combination for the emergency arbitrary speed braking working condition is the lateral SMC control algorithm and the longitudinal logic threshold control algorithm.

[0054] (1) The first control mode (high-speed cruising working condition): Lateral control-LQR (linear quadratic regulator) controller: A) State variable definition: x=[e_y,e_ψ,e y,e ψ]

[0055] Wherein: e y is the lateral position error, e_ψ is the heading angle error, e y is the lateral error rate, e ψ is the heading error rate.

[0056] B) State space model: =Ax+Bu A is the system matrix, B is the control matrix, and u is the front wheel steering control input.

[0057] C) Cost function: J=∫[x Qx+u Ru]dt Q is the state weight matrix (the diagonal elements correspond to the importance of each state variable), and R is the control input weight matrix.

[0058] D) Optimal control law: the optimal feedback gain matrix K is obtained by solving the Riccati equation, and the final control output is: u=-Kx.

[0059] E) Specific process: Real-time acquisition of vehicle state, calculation of lateral and heading error, construction of state vector, calculation of optimal front wheel steering angle through LQR algorithm, output to steering actuator.

[0060] Longitudinal control - PID controller: A) Error calculation: e_v = v_set - v_actual B) Control output: u_long = K_p · e_v + K_i · ∫e_vdt + K_d · de_v / dt C) Parameter tuning: Based on the longitudinal dynamics of the vehicle, determine K_p, K_i, K_d parameters through Ziegler-Nichols method or experimental tuning method.

[0061] D) Output limiter: Saturation processing of the final control command to ensure it is within the safe operating range of the motor / brake system.

[0062] (2) Second control mode (low-speed operation condition): Lateral control - Stanley control algorithm: A) Front axle error calculation: e_fa = (y_road - y_fa) · sin(ψ_road) - (x_road - x_fa) · cos(ψ_road) B) Heading error: e_ψ = ψ_vehicle - ψ_road C) Control law: δ = e_ψ + arctan((k · e_fa) / (v + ε)) Where k is the gain coefficient, ε is a small constant to prevent division by zero.

[0063] D) Curvature feedforward: δ_ff = arctan(L · κ), L is the wheelbase, κ is the road curvature E) Final output: δ_total = δ + δ_ff Longitudinal control - Parameter adaptive PID controller: A) Parameter adaptive mechanism: K_p_adaptive = K_p0 + α · |e_lateral| K_i_adaptive = K_i0 / (1 + β · |e_lateral|) B) Speed adjustment strategy: v_adjusted = v_nominal - γ · |e_lateral| C) Control output: u_long = K_p_adaptive · e_v + K_i_adaptive · ∫e_vdt + K_d · de_v / dt D) Safety first principle: when lateral tracking error exceeds threshold, automatically reduce target vehicle speed to prioritize obstacle avoidance safety (3) Third control mode (precise parking condition): Lateral and longitudinal collaborative MPC (model predictive control) controller: A) Prediction model: discrete state space model based on nonlinear vehicle dynamics: x(k+1)=f(x(k),u(k)) State variables: x=[X,Y,ψ,v,β,r]

[0064] Control input: u=[δ,a]

[0065] Where x(k) represents the state vector of the vehicle at time k, u(k) is the control input at time k, and f(x(k), u(k)) is the nonlinear state transition function, i.e. the vehicle dynamics submodel.

[0066] The elements in the vector [X, Y, ψ, v, β, r] represent the X-axis position of the vehicle in the current global coordinate system, the Y-axis position of the vehicle in the current global coordinate system, the yaw angle of the vehicle, the mass center speed of the vehicle, the mass center side slip angle of the vehicle, and the yaw rate of the vehicle, respectively.

[0067] The elements in [δ, a] represent the front wheel steering angle and longitudinal acceleration, respectively.

[0068] B) Optimization problem: min J=∑[x(k)-x_ref(k)] Q[x(k)-x_ref(k)]+u(k) Ru(k) Constraints: u_min≤u(k)≤u_max, Δu_min≤Δu(k)≤Δu_max Where x_ref(k) represents the reference state at time k, given by the path planner, which is the state that the vehicle should ideally reach; Q is the state error weight matrix, and R is the control input weight matrix.

[0069] C) Rolling optimization: solve the finite time optimal control problem at each control period, only execute the first control quantity.

[0070] D) Terminal constraint: x(N)=x_target, ensure accurate final arrival at target pose. Where N is the last step of the prediction horizon, and x_target is the terminal target state, i.e. the precise pose that the vehicle needs to reach at the parking point.

[0071] E) Real-time linearization: Apply continuous linearization technique to handle nonlinear prediction model, ensure real-time performance.

[0072] Fourth control mode (emergency braking condition): Lateral control - sliding mode control (SMC): A) Sliding surface design: s = _y + λ · e_y, where λ > 0 is a design parameter.

[0073] B) Equivalent control: u_eq = -( h / x) · f(x), based on nominal system model.

[0074] C) Switching control: u_sw = -K · sgn(s), K is switching gain.

[0075] D) Final control law: δ = u_eq + u_sw.

[0076] E) Boundary layer design: Use saturation function sat(s / Φ) instead of sign function to suppress chattering phenomenon.

[0077] Longitudinal control - logic threshold control: A) Emergency judgment logic: IF (TTC < TTC_threshold) OR (sudden cut-in flag = TRUE).

[0078] B) Control rule: THEN a_cmd = a_max_brake C) Exit condition: WHILE (v > 0.1 m / s) AND (obstacle distance < safety distance) D) Priority setting: Longitudinal braking priority higher than lateral obstacle avoidance, ensure the shortest braking distance.

[0079] E) Execution mechanism: Directly send maximum pressure command to brake system, bypass regular control loop.

[0080] Mode switching smoothing processing: Each controller uses state initialization and command smoothing strategy at switching moment, ensure the continuity of control quantity, avoid vehicle dynamic response mutation, improve driving comfort and control system stability.

[0081] Actuator driver receives the command of multi-mode controller (target front wheel steering angle and target acceleration / deceleration), and converts it into physical signal of underlying actuator. In lateral execution aspect, target steering angle is sent to steer-by-wire system through CAN bus to drive steering motor execution. In longitudinal execution aspect, target acceleration / deceleration is converted into throttle opening or brake pressure request, sent to steer-by-wire system and brake-by-wire system through CAN bus for execution. ​

[0082] The above describes in detail an embodiment of the unmanned vehicle multi-modal adaptive control method. Based on the unmanned vehicle multi-modal adaptive control method described in the above embodiment, the embodiment of the present application further provides an unmanned vehicle multi-modal adaptive control device corresponding to the method.

[0083] Figure 3 The present embodiment provides a structural schematic block diagram of an unmanned vehicle multi-modal adaptive control system. In the present embodiment, the unmanned vehicle multi-modal adaptive control system 300 can be divided into multiple functional units according to the functions performed by the system, as shown in the figure. The unit referred to in the present application refers to a series of computer program segments capable of being executed by at least one processor and capable of completing a fixed function, which are stored in the memory. Figure 3

[0084] The target parameter acquisition unit 310 is configured to acquire target parameters for multi-modal adaptive control.

[0085] The working condition recognition unit 320 is configured to recognize a working condition based on the target parameters through a preset rule engine.

[0086] The control algorithm determination unit 330 is configured to determine a corresponding combination of lateral and longitudinal control algorithms based on a current working condition recognition result according to a mapping relationship between preset working condition types and the combination of lateral and longitudinal control algorithms.

[0087] The control instruction generation and transmission unit 340 is configured to generate a control instruction based on the current combination of lateral and longitudinal control algorithms and send the control instruction to an execution mechanism.

[0088] The unmanned vehicle multi-modal adaptive control system of the present embodiment is used to implement the aforementioned unmanned vehicle multi-modal adaptive control method. Therefore, the specific embodiments in the system can be found in the aforementioned embodiment part of the unmanned vehicle multi-modal adaptive control method. Therefore, the specific embodiments can be referred to the description of the corresponding embodiment part, and will not be introduced here.

[0089] In addition, since the unmanned vehicle multi-modal adaptive control system of the present embodiment is used to implement the aforementioned unmanned vehicle multi-modal adaptive control method, its role corresponds to the role of the aforementioned method, and will not be described here.

[0090] Figure 4 The structural schematic diagram of the terminal 400 provided in the present embodiment includes a processor 410, a memory 420, and a communication unit 430. The processor 410 is configured to implement the flow steps of the aforementioned unmanned vehicle multi-modal adaptive control method embodiment when implementing the unmanned vehicle multi-modal adaptive control program saved in the memory 420. ​

[0091] The application further provides a computer storage medium. The storage medium can be a magnetic disc, an optical disc, a Read-Only Memory (ROM) or a Random Access Memory (RAM), etc. The computer storage medium stores an unmanned vehicle multi-mode adaptive control program. When the unmanned vehicle multi-mode adaptive control program is executed by a processor, the flow steps of the unmanned vehicle multi-mode adaptive control method embodiment described above are realized.

[0092] The above description of disclosed embodiments enables a person skilled in the art to implement or use the application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown in the application, but will conform to the widest scope consistent with the principles and novel features disclosed in the application.

Claims

1. A multimodal adaptive control method for unmanned vehicles, characterized in that, Includes the following steps: Obtain the target parameters for multimodal adaptive control; Based on the target parameters, the working condition is identified through a preset rule engine; Based on the mapping relationship between the preset working condition type and the combination of horizontal and vertical control algorithms, the corresponding combination of horizontal and vertical control algorithms is determined based on the current working condition identification result. Control commands are generated based on the current combination of horizontal and vertical control algorithms and sent to the actuator.

2. The multimodal adaptive control method for unmanned vehicles according to claim 1, characterized in that, Based on the target parameters, operating conditions are identified using a pre-defined rule engine, specifically including: Determine whether the emergency arbitrary speed braking condition is met based on the collision time and its rate of change. If so, the operating condition is identified as an emergency arbitrary speed braking condition; If not, determine whether the distance between the vehicle's position and the target parking point is less than or equal to the preset distance; If so, the operating condition is identified as a high-precision, extremely low-speed parking condition. If not, determine whether the vehicle's first future path is a curve or a narrow path based on the path planning data; If so, the working condition is identified as a low-to-medium speed passage condition with multiple obstacles; If not, a rectangular perception area is defined, and the obstacle density within the perception area is calculated. If the obstacle density is 0, the driving condition is identified as an obstacle-free medium-high speed cruise driving condition. If the obstacle density is not 0 and does not exceed the preset density threshold, the driving condition is determined based on the target future path of the dynamic obstacle and the corresponding second future path of the vehicle. If the condition is met, the driving condition is also identified as an obstacle-free medium-high speed cruise driving condition. If the condition is not met or the obstacle density exceeds the preset density threshold, the driving condition is also identified as a multi-obstacle medium-low speed driving condition.

3. The multimodal adaptive control method for unmanned vehicles according to claim 2, characterized in that, The determination of whether the emergency arbitrary speed braking condition is met is based on the collision time and its rate of change, specifically including: If the collision time is less than the preset collision time threshold and the absolute value of the collision time change rate is greater than the emergency threshold, then it is determined that the emergency arbitrary speed braking condition trigger condition is met.

4. The multimodal adaptive control method for unmanned vehicles according to claim 2, characterized in that, Based on path planning data, determine whether the vehicle's first future path is a curve or a narrow path, specifically including: Extract the discrete point set of the path to the future target length from the path planning data; Calculate the average curvature and total change in heading angle of the path segment based on the coordinates of all points in the discrete point set; If the average curvature exceeds the curvature threshold or the total change in heading angle exceeds the angle threshold, it is determined to be a curve; A local occupancy grid map corresponding to the future target length path is constructed using LiDAR scanning data. The passable width of the vehicle on the left and right is calculated based on the grid map. If the sum of the passable widths on the left and right is less than the width threshold, it is determined to be a narrow passage.

5. The multimodal adaptive control method for unmanned vehicles according to claim 2, characterized in that, Determine whether safe driving conditions are met based on the target future path of the dynamic obstacle and the corresponding second future path of the vehicle, specifically including: For each dynamic obstacle, Kalman filtering is used to predict its trajectory within the future target time period, and the distance and arrival time of the closest point to the corresponding second future path of the vehicle are calculated. If the distance of the closest point is greater than the preset distance threshold and the arrival time is greater than the preset time period, the safe driving conditions are determined to be met; otherwise, the safe driving conditions are determined not to be met.

6. The multimodal adaptive control method for unmanned vehicles according to any one of claims 1 to 5, characterized in that, The combination of lateral and longitudinal control algorithms for unobstructed high-speed cruise conditions is the lateral LQR control algorithm and the longitudinal PID control algorithm. The combination of lateral and longitudinal control algorithms for low-speed passage under multiple obstacles is the lateral Stanley control algorithm and the parameter-adjustable longitudinal PID control algorithm. The combination of lateral and longitudinal control algorithms corresponding to high-precision ultra-low speed docking conditions is the lateral and longitudinal collaborative MPC control algorithm. The combination of lateral and longitudinal control algorithms for emergency arbitrary speed braking is the lateral SMC control algorithm and the longitudinal logic threshold control algorithm.

7. The multimodal adaptive control method for unmanned vehicles according to any one of claims 1 to 5, characterized in that, When switching operating conditions, a weighted average algorithm is used to smoothly transition control commands, specifically including: Define a fixed transition time window T_transition, and denote the switching time as t_switch. Then, within the time interval t∈[t_switch,t_switch+T_transition], the control instruction u_final(t) output to the actuator is expressed as: u_final(t)=α(t)*u_A(t)+(1-α(t))*u_B(t) In the formula, u_A(t) is the control command generated under the operating condition before the switch, and u_B(t) is the control command generated under the operating condition after the switch; α(t) is a function that decays with time, expressed as α(t) = 1 - ((t - t_switch) / T_transition); When t > t_switch + T_transition, α(t) becomes 0.

8. A multimodal adaptive control system for unmanned vehicles, characterized in that, include: Target parameter acquisition unit: used to acquire target parameters for multimodal adaptive control; The working condition identification unit is used to identify working conditions based on target parameters and through a preset rule engine. The control algorithm determination unit is used to determine the corresponding combination of horizontal and vertical control algorithms based on the current working condition identification result, according to the mapping relationship between the preset working condition type and the combination of horizontal and vertical control algorithms. The control command generation and transmission unit is used to generate control commands based on the current combination of horizontal and vertical control algorithms and send them to the actuator.

9. A terminal, characterized in that, include: The memory is used to store the multimodal adaptive control program for autonomous vehicles. A processor is configured to implement the steps of the multimodal adaptive control method for an unmanned vehicle as described in any one of claims 1 to 7 when executing the multimodal adaptive control program for the unmanned vehicle.

10. A computer-readable storage medium, characterized in that, The readable storage medium stores an autonomous vehicle multimodal adaptive control program, which, when executed by a processor, implements the steps of the autonomous vehicle multimodal adaptive control method as described in any one of claims 1 to 7.

Images