An emergency obstacle avoidance method for intelligent connected vehicles, involving multi-target trajectory planning and lateral stability tracking.

Abstract

This invention discloses an emergency obstacle avoidance method for intelligent connected vehicles, involving multi-objective trajectory planning and lateral stability tracking. The method comprises four steps: Step 1, generating smooth candidate trajectories in different time domains based on the Pontryagin maximum principle, and selecting a set of safe trajectories with no collision risk by combining swept-body collision detection; Step 2, constructing a multi-objective optimization function that integrates comfort, efficiency, and safety, and applying constraints on vehicle speed and acceleration; Step 3, employing an adaptive elite sparrow search algorithm, efficiently solving for the optimal reference trajectory through stagnation detection scheduling and elite learning mechanisms; Step 4, linearly discretizing the vehicle dynamics model, explicitly constraining lateral acceleration, accurately tracking the trajectory, and suppressing the risks of sideslip and rollover. This invention integrates multi-objective trajectory optimization and lateral stability tracking, improves solution efficiency through algorithm improvements, explicitly constrains sideslip risks, and achieves high trajectory tracking accuracy and robustness, significantly improving the safety and reliability of autonomous driving.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle safety, and more specifically, to an emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, the safety and reliability of autonomous driving systems have received widespread attention. Currently, more than 1.3 million people die and approximately 50 million are injured in road traffic accidents globally each year, making traffic accidents a major cause of personal injury and death worldwide. Obstacle avoidance maneuvers and emergency lane changes, among the contributing factors to fatal traffic accidents, account for a significant proportion. Especially under low-traction conditions such as rain and snow, slippery roads and significantly reduced tire-road adhesion coefficients make vehicles more prone to rear-end collisions, lateral skidding, and even rollovers during braking and steering maneuvers, thus significantly increasing the risk of accidents. Statistics show that approximately 19.9% ​​of traffic accidents are related to adverse weather conditions such as rain or ice. Therefore, ensuring the driving safety and lateral stability of autonomous vehicles during emergency obstacle avoidance on slippery roads has become a critical technical problem that urgently needs to be solved.

[0003] Currently, autonomous driving technology can be mainly divided into two categories: end-to-end learning methods and rule-based modular methods. End-to-end methods rely on deep learning models to map perception input to control output, possessing strong feature learning and scenario adaptability capabilities. However, due to the opaque nature of their internal mechanisms and limited interpretability, their safety assessment and verification are challenging. In contrast, rule-based modular methods typically follow a structured process of "perception—localization—decision-planning—control—execution," with clear boundaries between functional modules, facilitating performance analysis, fault location, and safety constraint design and verification. Therefore, they still hold significant application value in safety-critical scenarios. The technical solution described in this paper falls under the rule-based category, focusing on trajectory planning and control. This stage directly determines the feasibility, safety, and stability of the vehicle in complex traffic environments and extreme road conditions, and is a core component for achieving safe autonomous driving.

[0004] In trajectory planning, existing modular autonomous driving systems often employ search-based planning methods, curve generation methods, or artificial potential field methods to generate feasible trajectories. For trajectory tracking and control, strategies such as proportional-integral-derivative (PID) control, linear quadratic regulation (LQR), or model predictive control (MPC) are commonly used. To improve feasibility in complex scenarios, some existing solutions introduce spatial constraints such as road boundaries and obstacle safety distances during the planning phase to adapt to restricted road environments. Some solutions use model predictive control frameworks to achieve obstacle avoidance, embedding traffic conditions, vehicle operating conditions, or environmental limitations as constraints into the optimization solution process. Other solutions combine filtering or observers for state estimation, along with robust control or zone control, to reduce the impact of uncertainties on trajectory tracking accuracy. Furthermore, some solutions combine artificial potential fields with predictive control and utilize parameters such as road width and vehicle geometry to construct collision constraints, thereby improving the safety and feasibility of the obstacle avoidance process.

[0005] While the aforementioned existing technologies can achieve trajectory planning and tracking control to a certain extent, they still have shortcomings in emergency obstacle avoidance scenarios on slippery roads. On the one hand, the multi-objective trade-offs of comprehensive trajectory performance indicators (such as comfort, traffic efficiency, and safety margin) are not sufficiently considered, making it difficult for the generated trajectory to balance stability and efficiency under extreme conditions. On the other hand, the constraints on vehicle lateral stability (such as sideslip angle, yaw rate, tire force saturation, and adhesion utilization) are not adequately characterized or are not tightly coupled with planning and control, making it difficult to effectively suppress instability risks such as sideslip and fishtailing under low-adhesion conditions. These factors are precisely the key constraints affecting the safe and stable driving of vehicles during emergency obstacle avoidance on low-adhesion roads. Therefore, there is an urgent need for a trajectory planning and control method that can take into account both comprehensive trajectory performance and explicit lateral stability to improve the safety and stability of autonomous vehicles during emergency obstacle avoidance on slippery roads. Summary of the Invention

[0006] To address the aforementioned issues, this paper proposes a safe autonomous driving method for emergency obstacle avoidance in intelligent connected vehicles that integrates multi-objective trajectory optimization and lateral stability tracking. This method first generates smooth candidate trajectories across multiple time domains based on the Pontryagin Maximum (PMP) principle and employs a sweep-body-based collision detection mechanism to quickly eliminate infeasible trajectories. Building upon this, a unified optimization objective function balancing comfort, efficiency, and safety is constructed, introducing an Artificial Potential Field (APF) term to quantify the safety margin between the vehicle and obstacles and road boundaries. Simultaneously, an Adaptive Elite Sparrow Search (AESSA) algorithm, integrating an adaptive scheduling strategy and an elite learning reinforcement mechanism, is proposed to improve global search efficiency and the optimality of the trajectory solution. For the trajectory tracking stage, a linear time-varying model predictive control method (LSLTV-MPC) considering lateral stability is designed, incorporating lateral acceleration (α) into the target value. y), centroid side slip angle ( ), yaw rate ( Lateral dynamics parameters such as the lateral load transfer rate (LTR) are explicitly incorporated into the constraints to reduce the risk of sideslip and rollover during emergency obstacle avoidance on wet and slippery surfaces while ensuring tracking stability. Finally, a joint simulation platform of MATLAB / Simulink and CarSim was built and real-vehicle tests were conducted. The experimental results show that the proposed method can achieve a comprehensive balance of comfort, efficiency, and safety during emergency obstacle avoidance, and maintains high trajectory tracking accuracy and lateral stability under high-speed conditions on wet and slippery surfaces.

[0007] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0008] An emergency obstacle avoidance method for intelligent connected vehicles, involving multi-target trajectory planning and lateral stability tracking, specifically includes the following steps:

[0009] Step 1, candidate trajectory generation and collision screening: smooth candidate trajectories in different time domains are generated based on PMP, and then the intersection of the spatiotemporal occupied areas of the vehicle and dynamic obstacles is simulated by a collision detection method based on swept volume to screen out a set of safe trajectories with no collision risk.

[0010] Step 2: Construct a multi-objective optimization function. Construct a unified objective function that covers comfort (square integral of longitudinal and lateral acceleration), efficiency (total obstacle avoidance time), and safety (APF-based safety margin). Eliminate differences in index scales through max-min normalization, and apply constraints such as vehicle speed and acceleration.

[0011] Step 3, optimal trajectory solution: AESSA is used to efficiently search for the optimal reference trajectory that balances multiple objectives from the set of safe trajectories through an adaptive scheduling strategy based on stagnation detection and an elite learning enhancement mechanism, taking into account both convergence speed and optimality of solution.

[0012] Step 4, Lateral stability trajectory tracking, linearizes and discretizes the nonlinear vehicle dynamics model, and based on LSLTV-MPC, explicitly constrains lateral stability indicators such as lateral acceleration and centroid sideslip angle, so as to suppress the risk of sideslip and rollover while accurately tracking the optimal trajectory.

[0013] Furthermore, step 1 includes the following steps:

[0014] Step 1.1, setting optimization objectives and constraints: First, with minimizing vehicle lateral vibration as the core objective, the optimization index is constructed as lateral jerk. The square integral is used to simultaneously constrain the initial and terminal boundary conditions of lateral position, velocity, and acceleration, ensuring that the trajectory meets the requirements of the driving scenario:

[0015] ,

[0016] In the formula, t f For the trajectory terminal time, The vehicle's lateral position. For lateral velocity, For lateral acceleration, For lateral jerk, the initial and terminal boundary conditions are:

[0017] ,

[0018] Step 1.2, Construction and solution of the Hamiltonian function, by introducing costate variables. , and Constructing the Hamiltonian function:

[0019] ;

[0020] Furthermore, step 1.2 includes the following steps:

[0021] Step 1.2.1, based on the necessary conditions of the Pontryagin maximum principle, derive the differential equations for the costate variables:

[0022] ,

[0023] Step 1.2.2: The system state equations can be expressed using the partial derivatives of the Hamiltonian function with respect to the costate variables:

[0024] ,

[0025] Step 1.2.3, regarding the Hamiltonian function with respect to... Taking the partial derivative and setting it to 0, we obtain the optimal control input:

[0026] ,

[0027] By solving the above two-point boundary value problem, different time domains are generated ( The smooth candidate trajectories are shown below, each containing the vertical position. Horizontal position With heading angle , forming a trajectory set ,in ( (For predicting longitudinal velocity) ;

[0028] Step 1.3, Collision detection and safe trajectory selection based on swept body.

[0029] Furthermore, step 1.3 includes the following steps:

[0030] Step 1.3.1, Trajectory Discretization and Occupied Region Modeling: Each candidate trajectory is discretized at a fixed time step. Discretized ( , , For the first The time domain of the trajectory); the vehicle profile is modeled as a rectangle B(L,W) ( For the train commander, (where the vehicle width is a reference value), the rectangle in the vehicle coordinate system is mapped to the world coordinate system through rigid body transformation to obtain the area occupied by the vehicle at each discrete time point:

[0031] ,

[0032] in, , It is a planar rotation matrix;

[0033] Step 1.3.2, Collision Detection and Filtering Logic, the dynamic obstacle set is represented as follows: ( For the first (the spatiotemporal contour of an obstacle), if a certain trajectory Satisfy all and All have If a trajectory is not found to be a collision-free trajectory, then all collision-free trajectories constitute a safe trajectory set. Infeasible solutions that pose a collision risk are eliminated.

[0034] Furthermore, step 2 includes the following steps:

[0035] Step 2.1 Comfort function design: Comfort is directly related to the vehicle's longitudinal and lateral acceleration and jerk. The acceleration and jerk are constrained to not exceed preset maximum values.

[0036] ,

[0037] in, , For the maximum permissible longitudinal and lateral acceleration, , To the maximum permissible longitudinal and lateral jerk, , The comfort function is defined as the square integral of the acceleration over the planning time domain, where the acceleration is the time derivative.

[0038]

[0039] Step 2.2, Efficiency function design, based on the total time the vehicle takes to complete obstacle avoidance. As an efficiency indicator, minimizing obstacle avoidance time reduces the impact on traffic flow.

[0040] ,

[0041] Step 2.3, Safety Function Design: The safety function is established using the APF method, which mainly includes road potential field modeling and obstacle vehicle potential field modeling, thereby forming a comprehensive safety potential field.

[0042] Furthermore, step 2.3 includes the following steps:

[0043] Step 2.3.1, Modeling the road potential field. The road potential field includes the boundary line potential field and the centerline potential field, and the formula is:

[0044] ,

[0045] in, For the potential field at the road boundary, For the first The center line of the lane has a potential field. The total number of lane centerlines is expressed as follows:

[0046] ,

[0047] In the formula , The potential coefficient (which determines the maximum potential value). , It is the convergence factor (which determines the range of influence of the potential field). The lateral position of the vehicle. , The left and right boundaries of the road. For the first Lane markings location;

[0048] Step 2.3.2, obstacle vehicle potential field modeling. The isosurface of the obstacle vehicle potential field is an ellipse, guiding the vehicle to smoothly avoid obstacles along the outer contour of the ellipse. The formula is:

[0049] ,

[0050] in, Let be the potential field coefficient of the obstacle vehicle. , For vertical and horizontal convergence factors, , For the first The location of the obstructing vehicle;

[0051] Step 2.3.3: Integrating the safety potential field and safety cost, and integrating the road potential field and the obstacle vehicle potential field, we obtain the comprehensive safety potential field:

[0052] ,

[0053] in, Given the total number of vehicles with obstacles, the safety cost is defined as the time-domain normalized integral of the comprehensive potential field, ensuring that the safety of trajectories of different durations is comparable:

[0054] .

[0055] Step 2.4: Comprehensively optimize the objective function, which includes comfort, efficiency, and safety functions. Max-min normalization is used to eliminate differences in indicator scales.

[0056] ,

[0057] in, , To avoid extremely small constants with a denominator of 0, a unified optimization objective is constructed and constraints are imposed:

[0058] ,

[0059] in, For the lateral limit of the road, This is the maximum permissible speed.

[0060] Furthermore, step 3 includes the following steps:

[0061] Step 3.1, Standard Sparrow Search Algorithm (SSA) basic update, planning time domain. As an optimization variable, the population size is The maximum number of iterations is Individuals are divided into three categories: discoverers, followers, and vigilants. The update rules are as follows:

[0062] (1) Discoverer update

[0063] ,

[0064] in, This is a random warning value. As a safety threshold, For Gaussian perturbation, This is the control parameter for the exponential decay rate;

[0065] (2) Follower update

[0066] ,

[0067] in, For the first Next iteration's optimal discoverer As the worst individual, Random symbols;

[0068] (3) Guardian updates and boundary handling

[0069] ;

[0070] in, For the first The optimal individual in the next iteration. The parameters are scaled to reflect the perturbation magnitude. After updating, boundary conditions are used to maintain the variables within the specified range. (Within the feasible timeframe of the planning period):

[0071] ;

[0072] Step 3.2: An improved sparrow optimization algorithm is formed through adaptive scheduling and elite learning enhancement mechanism.

[0073] Furthermore, step 3.2 includes the following steps:

[0074] Step 3.2.1, Adaptive scheduling based on stagnation detection. This is achieved through a sliding window (length...) Calculate the optimal fitness improvement margin:

[0075]

[0076] in, ( (For the fitness function). If ( (For the stagnation threshold), global exploration is enhanced using Levi's flight:

[0077]

[0078] in, To explore step length, The Lévy distribution characteristic index is used; otherwise, local development is performed through Gaussian refinement.

[0079]

[0080] Refine the scale A linear decay strategy is adopted:

[0081]

[0082] in, This is the initial refinement scale.

[0083] Step 3.2.2, Elite Learning Enhancement Mechanism. Maintaining the Elite Memory Set. Before storage Find the optimal feasible solution and calculate the mean and variance of the elite set:

[0084]

[0085] Generate based on elite set statistical features One enhanced candidate solution:

[0086]

[0087] in, To control the diversity parameters, enhanced candidate solutions are incorporated into the population before selection, thereby improving convergence robustness.

[0088] Step 3.3, Optimal trajectory output: The comprehensive optimization objective is solved iteratively using AESSA to obtain the optimal trajectory from the safe trajectory set. The optimal reference trajectory that balances comfort, efficiency, and safety is selected from the data. .

[0089] Furthermore, step 4 includes the following steps:

[0090] Step 4.1, linearization and discretization of the vehicle dynamics model. The continuous state-space equations of vehicle dynamics are as follows:

[0091] ,

[0092] Among them, state variables Control input Output variables , To output the matrix, the forward Euler method is used for discretization:

[0093] ,

[0094] in, The discrete time step;

[0095] Furthermore, linearization and control increment transformation are carried out at the operating point. After Taylor expansion and retaining the first-order terms, we obtain the discrete linearized equation:

[0096]

[0097] in, , , To linearize the error, it is transformed into a function based on control increments. Extended form:

[0098]

[0099] in, , , , , ;

[0100] Step 4.2, Model Prediction and Objective Function Construction: First, conduct multi-step prediction and set the prediction step size. Control step size Predict future states and outputs:

[0101] ,

[0102] The system equations within the prediction domain simplify to:

[0103] ;

[0104] Furthermore, a tracking objective function is defined to minimize tracking error and control incremental smoothness, and a relaxation factor is introduced. Ensure a feasible solution:

[0105] ,

[0106] in, , , This is the weight matrix. Output as reference trajectory;

[0107] Step 4.3, Constraint Design and Solution: Basic constraints include applying inequality constraints to the output, control input, and increment.

[0108] ,

[0109] in, As a relaxation factor, The weights are for the slack terms;

[0110] Lateral stability constraints: This invention designs a system including lateral acceleration. , centroid side slip angle yaw rate Lateral load transfer rate The indicators are as follows:

[0111] ,

[0112] in, The calculation formula is:

[0113] ,

[0114] For the sprung mass, The roll angle is... This is the roll acceleration. For the wheelbase, the nonlinear constraints are linearized and discretized:

[0115] ,

[0116] For a linear polynomial of state variables, For the control variable coefficient matrix, This is for linearization error;

[0117] Step 4.4, Quadratic Programming Solution and Control Execution: The constrained optimization problem is transformed into a quadratic programming problem to be solved, and the control command at the current moment (front wheel steering angle) is output. Longitudinal acceleration It achieves closed-loop trajectory tracking through real-time feedback, ensuring vehicle stability under low-adhesion and high-speed conditions.

[0118] Beneficial effects: This invention integrates multi-objective trajectory optimization and lateral stability tracking, balancing safety, efficiency, and comfort in emergency obstacle avoidance on slippery roads. By improving the algorithm, it enhances solution efficiency, explicitly constrains sideslip risk, and achieves high trajectory tracking accuracy and robustness, significantly improving the safety and reliability of autonomous driving. Specifically, this is reflected in the following aspects:

[0119] High safety: Explicitly constrained lateral stability indicators effectively suppress the risk of skidding and rollover on low-adhesion road surfaces, and provide sufficient safety margin for obstacle avoidance;

[0120] Good comfort: Minimal lateral acceleration, smooth trajectory, and excellent riding experience;

[0121] Excellent efficiency: The improved algorithm quickly finds the optimal solution, has a short obstacle avoidance time, and strong real-time performance;

[0122] Highly robust: adaptable to harsh working conditions such as rain and snow, with high tracking accuracy and wide applicability. Attached Figure Description

[0123] Figure 1 This diagram illustrates the steps of multi-objective trajectory planning and lateral stability tracking for intelligent connected vehicles.

[0124] Figure 2 A flowchart illustrating multi-objective trajectory planning and lateral stability tracking for intelligent connected vehicles is shown.

[0125] Figure 3 This is a schematic diagram of the trajectory planning process for intelligent connected vehicles;

[0126] Figure 4 This is a schematic diagram comparing the fitness convergence curves of the trajectory planning process for intelligent connected vehicles.

[0127] Figure 5This is a comparative diagram of trajectory planning and obstacle avoidance for intelligent connected vehicles;

[0128] Figure 6 This is a comparative diagram showing the trackability of trajectories generated by intelligent connected vehicle planning methods;

[0129] Figure 7 This is a multi-dimensional quantitative comparison diagram of trajectory planning for intelligent connected vehicles;

[0130] Figure 8 This is a diagram comparing the trajectory tracking performance of intelligent connected vehicles under different road surface adhesion coefficients;

[0131] Figure 9 This is a diagram comparing the trajectory tracking errors of intelligent connected vehicles in different scenarios;

[0132] Figure 10 This is a diagram comparing the trajectory tracking calculation time of intelligent connected vehicles in different scenarios.

[0133] Figure 11 This is a diagram comparing the real-time computing time of intelligent connected vehicles in different scenarios. Detailed Implementation

[0134] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings: These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0135] like Figure 1 The diagram illustrates the steps of an emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of an intelligent connected vehicle according to the present invention. Figure 2 The diagram shows the specific process of target trajectory planning and lateral stability tracking for intelligent connected vehicles.

[0136] This invention utilizes a CarSim2020 and MATLAB / Simulink co-simulation platform for simulation verification. A Class C sedan model from CarSim2020 is used to simulate the dynamics and driving environment of an autonomous vehicle. The experimental scenario is a three-lane highway with a standard lane width of 3.75m, where the vehicle's speed during obstacle avoidance is no less than 25m / s. Different adverse weather conditions, including rain, snow, and icing (with road surface adhesion coefficients of 0.4, 0.28, and 0.18 respectively), are incorporated to systematically evaluate the performance of the proposed method under these demanding conditions. To comprehensively verify the performance of the proposed framework, the simulation is divided into two progressive stages: baseline trajectory planning verification and robustness analysis of trajectory tracking under extreme conditions.

[0137] This invention uses a high-speed emergency obstacle avoidance test scenario as an example to carry out an embodiment, including the following steps:

[0138] Step 1: Candidate Trajectory Generation and Collision Screening. Smooth candidate trajectories in different time domains are generated based on PMP. Then, a collision detection method based on swept volumes is used to simulate the intersection of the spatiotemporal occupied areas of the vehicle and dynamic obstacles, and a set of safe trajectories with no collision risk is screened out.

[0139] Step 2: Construction of Multi-Objective Optimization Function. A unified objective function is constructed that covers comfort (square integral of longitudinal and lateral acceleration), efficiency (total obstacle avoidance time), and safety (APF quantified safety margin). The difference in index scale is eliminated by max-min normalization, and constraints such as vehicle speed and acceleration are applied.

[0140] Step 3: Optimal Trajectory Solution. AESSA is employed, using an adaptive scheduling strategy based on stagnation detection and an elite learning reinforcement mechanism to efficiently search for the optimal reference trajectory balancing multiple objectives from a set of safe trajectories, taking into account both convergence speed and solution optimality.

[0141] Step 4: Lateral stability trajectory tracking. The nonlinear vehicle dynamics model is linearized and discretized. Based on LSLTV-MPC, lateral stability indices such as lateral acceleration and center of gravity sideslip angle are explicitly constrained to suppress the risk of sideslip and rollover while accurately tracking the optimal trajectory.

[0142] The following is a complete description of the specific process of this embodiment:

[0143] (1.1) In the first stage, the performance of the trajectory planning layer is verified: the vehicle travels at a speed of 25 m / s on a dry asphalt road, avoiding obstacles and overtaking. The speed of the obstacle vehicle in front is 15 m / s. The focus is on examining the planner's ability to generate the optimal reference trajectory. The trajectory is required to take into account multiple objectives, maintain a strict safety margin, ensure high traffic efficiency, and optimize ride comfort by minimizing lateral acceleration.

[0144] (1.1.1) The trajectory planning process sets up an overtaking and obstacle avoidance scenario where the vehicle travels at 25 m / s on a standard dry road surface to verify the effectiveness of the proposed trajectory planning method. The trajectory generation process is as follows: Figure 3 As shown, it includes the potential field of obstacle vehicles, the potential field of the road, the set of feasible trajectories, the set of infeasible trajectories (i.e., trajectories that violate lateral acceleration constraints or collision avoidance constraints), and the final optimal trajectory generated by AESSA that takes into account comfort, efficiency and safety.

[0145] (1.1.1.1) First, a set of smooth candidate trajectories is generated using PMP, such as Figure 3 As shown in t min and t max The internal trajectory set is then used to eliminate trajectories with potential collisions using swept-body collision detection. Subsequently, a comprehensive objective function is constructed, introducing an APF (Advanced Persistent Risk Factor) term as a safety component to quantify the safety margin between the vehicle and surrounding obstacles and road boundaries. Figure 3 The light red area represents the safe trajectory set, and the gray area represents the potential collision / constraint violation trajectory set. Finally, AESSA is used to filter out the trajectory that achieves the optimal trade-off between comfort, efficiency, and safety, such as... Figure 3 The dark red trajectory is shown.

[0146] (1.1.1.2) To verify the convergence performance of the proposed AESSA algorithm in solving the optimal trajectory planning problem, a comprehensive comparative experiment was conducted, consisting of two parts: First, an ablation experiment comparing AESSA with its variants (ASSA including adaptive scheduling, and ESSA including elite learning) and the benchmark algorithm SSA; second, a comparative experiment comparing AESSA with mainstream swarm intelligence algorithms, including the Improved Whale Optimization Algorithm (NIWOA), the Hybrid Gray Wolf Optimization Algorithm (HCGO), the Multi-Swarm Mutant Moth-Flame Optimization Algorithm (MM-MFO), and the Improved Particle Swarm Optimization Algorithm (IPSO). The fitness convergence curves of each algorithm in the simulated trajectory planning scenario are shown below. Figure 4 As shown.

[0147] (1.1.1.2.1) by Figure 4 It can be seen that AESSA's convergence performance is significantly better than all the compared algorithms: in the early stages of iteration, its fitness decreases at the steepest rate, indicating that the algorithm searches faster and can quickly locate high-quality regions in the solution space; in the local zoom-in range of 40-90 iterations, NIWOA, HCGO, MM-MFO, and IPSO all exhibit convergence stagnation or premature convergence to suboptimal solutions, while AESSA's fitness continues to optimize, eventually reaching its lowest value. AESSA's final fitness value is approximately 0.52, achieving the optimal trade-off between comfort, efficiency, and safety.

[0148] (1.1.1.2.2) In addition, Figure 4 Ablation experiments further demonstrate that ASSA and ESSA outperform SSA, while AESSA surpasses both variants, proving that the fusion design of stagnation-aware scheduling and elite learning-enhanced generation can deliver synergistic gains. Simultaneously, AESSA maintains high computational efficiency, with a single planning time of only 4.53 ms, comparable to SSA-based benchmark algorithms and significantly lower than NIWOA (9.92 ms), HCGO (8.76 ms), MM-MFO (7.32 ms), and IPSO (7.65 ms), validating its feasibility for real-time applications.

[0149] (1.1.2) To verify the effectiveness of the proposed method, Figure 5The complete obstacle avoidance trajectory generated by the proposed PMP-APF-AESSA method is presented and compared with three benchmark methods: the improved artificial potential field method (AutoPFied), the AutoPFied-FS method (applying a smoothing filter to the artificial potential field output), and the MPC-APF method (integrating the artificial potential field as a safety constraint into the model predictive control framework). From the trajectory characteristics, the single AutoPFied method is prone to getting trapped in local optima and has poor trajectory smoothness; while the AutoPFied-FS, MPC-APF, and PMP-APF-AESSA methods all improve the smoothness of the generated trajectories. Furthermore, the AutoPFied-based method initiates obstacle avoidance earlier, allowing for a rapid response to obstacles, but resulting in more noticeable detour maneuvers; the MPC-APF method exhibits a delayed obstacle avoidance response and a narrow safety margin between the trajectory and obstacles, making it difficult to consistently maintain a sufficient safe distance; in contrast, the PMP-APF-AESSA method achieves a better balance between obstacle avoidance timing, efficiency, and comfort.

[0150] (1.1.3) Furthermore, linear time-varying model predictive control (LTV-MPC) is used as the baseline controller to verify the trackability of trajectories generated by each planning method, such as... Figure 6 As shown, a comprehensive quantitative comparison was conducted from five dimensions: comfort (lateral acceleration), efficiency (lane change time), safety (average safe distance during lane change), trackability (tracking error), and average planning time. The results are as follows. Figure 7 As shown in the figure. The results show that the PMP-APF-AESSA method has extremely high tracking accuracy, with a lateral deviation of only 0.013m and a lateral jerk of 0.017 m / s³, exhibiting excellent ride comfort. The MPC-APF method has the shortest lane-changing time (4.9s) and superior traffic efficiency, but its safe distance from obstacles is only 30.86m, which is relatively small. The AutoPFied-based method has the highest planning efficiency, with an average processing time of only 2.1 ms and 2.3 ms, but it has significant shortcomings in trajectory smoothness and tracking accuracy. Overall, the PMP-APF-AESSA trajectory planning method achieves the best comprehensive performance in terms of comfort, efficiency, and safety, while ensuring high tracking accuracy and acceptable planning efficiency, with an average planning time of 6.4 ms.

[0151] (2.1) In the second stage, sensitivity analysis was performed on the trajectory tracking control layer to explore its performance boundary under different environmental constraints: different road surface adhesion coefficients: the vehicle maintained a constant speed of 25 m / s, and the tracking accuracy and vehicle stability of the controller were compared on rainy (μ=0.4), snowy (μ=0.28), and icy (μ=0.18) road surfaces; different longitudinal vehicle speeds: under typical rainy conditions (μ=0.4), the dynamic response and tracking accuracy of the algorithm at high speeds of 27 m / s, 30 m / s, and 33 m / s were further tested.

[0152] (2.1.1) To verify the trajectory tracking performance under extreme weather conditions, three test scenarios were designed based on the reference trajectory generated by PMP-APF-AESSA: rainfall (μ=0.4), snowfall (μ=0.28), and icing (μ=0.18), with the vehicle's constant speed set at 25 m / s. The proposed LSLTV-MPC was compared with linear time-invariant model predictive control (LTI-MPC), the current best time-optimal linear time-varying model predictive control (TOLTV-MPC), and Lyapunov-based nonlinear model predictive control (LNMPC). The tracking results are as follows: Figure 8 As shown. Specifically: Under rainy conditions, the LTI-MPC exhibits lateral displacement deviation, with a yaw angle... lateral acceleration a y , centroid side slip angle Both LTI-MPC and TOLTV-MPC showed slight fluctuations in lateral stability index LTR, while other methods maintained relatively stable vehicle conditions; under snow and ice conditions, both LTI-MPC and TOLTV-MPC exhibited significant lateral displacement deviations, and a y , The fluctuation range of LTR increased significantly, while LNMPC and LSLTV-MPC maintained vehicle stability and demonstrated good lateral stability. These performance differences primarily stem from the controller's modeling strategy: LTI-MPC uses a fixed linear prediction model, which cannot adapt to changes in vehicle dynamics caused by variations in road adhesion coefficient. On low-adhesion surfaces, model mismatch worsens, leading to significant vehicle sideslip. a y , Stability / rollover related indicators such as LTR deteriorated sharply; although TOLTV-MPC improved its adaptability by updating the model online linearized, it is still essentially a linear approximation and cannot accurately capture strong nonlinear characteristics such as tire force saturation. Its performance dropped significantly in extremely low-adhesion conditions such as snow and ice. LNMPC uses a nonlinear dynamic model, which can accurately describe the nonlinear characteristics of the vehicle and exhibits excellent control performance across the entire adhesion coefficient range. It achieves high-precision trajectory tracking and effectively suppresses [rollover] during lane changes. a y , The increase in LTR; LSLTV-MPC incorporates lateral dynamic constraints on the basis of LTV-MPC. Although it adopts a linear model, it effectively suppresses vehicle sideslip through reasonable constraint design and controls the vehicle state within the stable boundary, exhibiting good stability under low-adhesion conditions.

[0153] (2.1.2) To verify the trajectory tracking performance in the overtaking scenario under high-speed driving conditions, the vehicle overtook other vehicles at speeds of 27 m / s, 30 m / s, and 33 m / s, respectively, with the road surface adhesion coefficient set to 0.4. The trajectory tracking performance of LSLTV-MPC, LTI-MPC, TOLTV-MPC, and LNMPC was compared. The trajectory tracking performance at different speeds is as follows: Figure 9 As shown in the figure, the results indicate that as the vehicle speed increases from 27 m / s to 30 m / s and 33 m / s, the lateral stability and vehicle state of the four MPC controllers exhibit significant differences: LTI-MPC, using a fixed linear model, can complete basic lane-changing operations at 27 m / s, but exhibits significant control lag and sideslip at 30 m / s, and completely loses lateral control at 33 m / s, posing an extremely high risk of instability; TOLTV-MPC shows good control performance at 27 m / s and 30 m / s, achieving high-precision trajectory tracking, but at 33 m / s, its control accuracy decreases due to tire force saturation and vehicle nonlinear dynamics. a y , The LTR (Long-Range Trajectory) increases significantly, leading to a decrease in overall vehicle stability. LNMPC, benefiting from comprehensive modeling of nonlinear dynamics, exhibits excellent control performance at all three speeds, achieving high-precision trajectory tracking during lane changes while effectively suppressing the increase in stability indicators. LSLTV-MPC, by incorporating lateral stability constraints into its control framework, effectively reduces the risk of vehicle instability at 30 m / s and 33 m / s. Although the trajectory deviation increases slightly, reasonable constraints on the control inputs mitigate this risk. a y , Both LTR and LTR are consistently kept within safe thresholds, demonstrating strong robustness and security.

[0154] (2.1.3) To evaluate the tracking accuracy of the proposed method in different scenarios, the tracking errors of each algorithm under different road surface adhesion coefficients and different driving speeds were calculated. The real-time absolute tracking errors are as follows: Figure 10 As shown in the figure. The results show that under low-friction road surface or high-speed driving conditions, the tracking errors of LTI-MPC and TOLTV-MPC increase significantly, while LNMPC and LSLTV-MPC maintain smaller error values ​​and show better trajectory tracking performance in all scenarios.

[0155] (2.1.4) To further analyze the computational efficiency of each method, the computation time was calculated under different road surface adhesion coefficients and different driving speeds. The real-time computation time is as follows: Figure 11 As shown in the figure. The results indicate that the computation time of LTI-MPC, TOLTV-MPC, and LSLTV-MPC is relatively low, while the computation time of LNMPC is significantly higher.

[0156] The reason for this is that LNMPC uses a nonlinear dynamic model to describe the system, and its optimization problem is essentially a non-convex nonlinear programming problem. It requires iterative algorithms to calculate gradients, and the computational complexity increases exponentially with the number of iterations. Therefore, although the tracking performance is excellent, the computation time is too long. LTI-MPC and TOLTV-MPC are based on linear models, which transform the optimization problem into a convex optimization problem and have higher computational efficiency. However, the linear model cannot accurately capture the nonlinear dynamic characteristics of the vehicle, and the tracking error is significant under low-adhesion and high-speed conditions. In contrast, the proposed LSLTV-MPC retains the convex optimization characteristics of the linear model, and its computational efficiency is comparable to that of LTI-MPC. At the same time, by optimizing the lateral stability design, it overcomes the limitations of traditional linear models in trajectory tracking. Finally, it achieves a dual guarantee of excellent trajectory tracking performance and computational efficiency in various scenarios.

[0157] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An emergency obstacle avoidance method for intelligent connected vehicles, characterized by multi-target trajectory planning and lateral stability tracking, wherein... Specifically, the following steps are included: Step 1, candidate trajectory generation and collision screening: smooth candidate trajectories in different time domains are generated based on PMP, and then the intersection of the spatiotemporal occupied areas of the vehicle and dynamic obstacles is simulated by a collision detection method based on swept volume to screen out a set of safe trajectories with no collision risk. Step 2: Construct a multi-objective optimization function. Construct a unified objective function that covers comfort (square integral of longitudinal and lateral acceleration), efficiency (total obstacle avoidance time), and safety (APF-based safety margin). Eliminate differences in index scales through max-min normalization, and apply constraints such as vehicle speed and acceleration. Step 3, optimal trajectory solution: AESSA is used to efficiently search for the optimal reference trajectory that balances multiple objectives from the set of safe trajectories through an adaptive scheduling strategy based on stagnation detection and an elite learning enhancement mechanism, taking into account both convergence speed and optimality of solution. Step 4, Lateral stability trajectory tracking, linearizes and discretizes the nonlinear vehicle dynamics model, and based on LSLTV-MPC, explicitly constrains lateral stability indicators such as lateral acceleration and centroid sideslip angle, so as to suppress the risk of sideslip and rollover while accurately tracking the optimal trajectory.

2. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 1, characterized in that, Step 1 includes the following steps: Step 1.1, setting optimization objectives and constraints: First, with minimizing vehicle lateral vibration as the core objective, the optimization index is constructed as lateral jerk. The square integral is used to simultaneously constrain the initial and terminal boundary conditions of lateral position, velocity, and acceleration, ensuring that the trajectory meets the requirements of the driving scenario: ， In the formula, t f For the trajectory terminal time, The vehicle's lateral position. For lateral velocity, For lateral acceleration, For lateral jerk, the initial and terminal boundary conditions are: ， Step 1.2, Construction and solution of the Hamiltonian function, by introducing costate variables. , and Constructing the Hamiltonian function: ； Step 1.3, Collision detection and safe trajectory selection based on swept body.

3. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 1, characterized in that, Step 2 includes the following steps: Step 2.1 Comfort function design: Comfort is directly related to the vehicle's longitudinal and lateral acceleration and jerk. The acceleration and jerk are constrained to not exceed preset maximum values. ， in, , For the maximum permissible longitudinal and lateral acceleration, , To the maximum permissible longitudinal and lateral jerk, , The comfort function is defined as the square integral of the acceleration over the planning time domain, where the acceleration is the time derivative. ， Step 2.2, Efficiency function design, based on the total time the vehicle takes to complete obstacle avoidance. As an efficiency indicator, minimizing obstacle avoidance time reduces the impact on traffic flow. ， Step 2.3, Safety Function Design: The safety function is established using the APF method, which mainly includes road potential field modeling and obstacle vehicle potential field modeling, thereby forming a comprehensive safety potential field. Step 2.4: Comprehensively optimize the objective function, which includes comfort, efficiency, and safety functions. Max-min normalization is used to eliminate differences in indicator scales. ， in, , To avoid extremely small constants with a denominator of 0, a unified optimization objective is constructed and constraints are imposed: ， in, For the lateral limit of the road, This is the maximum permissible speed.

4. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 1, characterized in that, Step 3 includes the following steps: Step 3.1, Standard Sparrow Search Algorithm (SSA) basic update, planning time domain. As an optimization variable, the population size is The maximum number of iterations is Individuals are divided into three categories: discoverers, followers, and vigilants. The update rules are as follows: (1) Discoverer update ， in, This is a random warning value. As a safety threshold, For Gaussian perturbation, This is the control parameter for the exponential decay rate; (2) Follower update ， in, For the first Next iteration's optimal discoverer As the worst individual, The symbol is random. (3) Guardian updates and boundary handling ； in, For the first The optimal individual in the next iteration. The parameters are scaled to the disturbance magnitude, and after updating, boundary conditions are applied to maintain the variable within the range. (Within the feasible timeframe of the planning period): ； Step 3.2: An improved sparrow optimization algorithm is formed through adaptive scheduling and elite learning enhancement mechanisms; Step 3.3, Optimal trajectory output: The comprehensive optimization objective is solved iteratively using AESSA to obtain the optimal trajectory from the safe trajectory set. The optimal reference trajectory that balances comfort, efficiency, and safety is selected from the data. .

5. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 1, characterized in that, Step 4 includes the following steps: Step 4.1, linearization and discretization of the vehicle dynamics model. The continuous state-space equations of vehicle dynamics are as follows: ， Among them, state variables Control input Output variables , To output the matrix, the forward Euler method is used for discretization: ， in, The discrete time step; Step 4.2, Model Prediction and Objective Function Construction: First, conduct multi-step prediction and set the prediction step size. Control step size Predict future states and outputs: ， The system equations within the prediction domain simplify to: ； Furthermore, a tracking objective function is defined to minimize tracking error and control incremental smoothness, and a relaxation factor is introduced. Ensure a feasible solution: ， in, , , This is the weight matrix. Output as reference trajectory; Step 4.3, Constraint Design and Solution: Basic constraints include applying inequality constraints to the output, control input, and increment. ， in, As a relaxation factor, The weights are for the slack terms; Lateral stability constraints: This invention designs a system including lateral acceleration. , centroid side slip angle yaw rate Lateral load transfer rate The indicators are as follows: ， in, The calculation formula is: ， For the sprung mass, The roll angle is... This is the roll acceleration. For the wheelbase, the nonlinear constraints are linearized and discretized: ， For a linear polynomial of state variables, For the control variable coefficient matrix, This is for linearization error; Step 4.4, Quadratic Programming Solution and Control Execution: The constrained optimization problem is transformed into a quadratic programming problem to be solved, and the control command at the current moment (front wheel steering angle) is output. Longitudinal acceleration It achieves closed-loop trajectory tracking through real-time feedback, ensuring vehicle stability under low-adhesion and high-speed conditions.

6. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 2, characterized in that, Furthermore, step 1.2 includes the following steps: Step 1.2.1, based on the necessary conditions of the Pontryagin maximum principle, derive the differential equations for the costate variables: ， Step 1.2.2: The system state equations can be expressed using the partial derivatives of the Hamiltonian function with respect to the costate variables: ， Step 1.2.3, regarding the Hamiltonian function with respect to... Taking the partial derivative and setting it to 0, we obtain the optimal control input: ， By solving the above two-point boundary value problem, different time domains are generated ( The smooth candidate trajectories are shown below, each containing the vertical position. Horizontal position With heading angle , forming a trajectory set ,in ( (For predicting longitudinal velocity) .

7. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 2, characterized in that, Furthermore, step 1.3 includes the following steps: Step 1.3.1, Trajectory Discretization and Occupied Region Modeling: Each candidate trajectory is discretized at a fixed time step. Discretized ( , , For the first The time domain of the trajectory); the vehicle contour is modeled as a rectangle B(L,W) ( For the train commander, (where the vehicle width is a reference value), the rectangle in the vehicle coordinate system is mapped to the world coordinate system through rigid body transformation to obtain the area occupied by the vehicle at each discrete time point: ， in, , It is a planar rotation matrix; Step 1.3.2, Collision Detection and Filtering Logic, the dynamic obstacle set is represented as follows: ( For the first (the spatiotemporal contour of an obstacle), if a certain trajectory Satisfy all and All have If a trajectory is not found to be a collision-free trajectory, then all collision-free trajectories constitute a safe trajectory set. Infeasible solutions that pose a collision risk are eliminated.

8. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 3, characterized in that, Furthermore, step 2.3 includes the following steps: Step 2.3.1, Modeling the road potential field. The road potential field includes the boundary line potential field and the centerline potential field, and the formula is: ， in, For the potential field at the road boundary, For the first The center line of each lane has a potential field. The total number of lane centerlines is expressed as follows: ， In the formula , This is the potential field coefficient (which determines the maximum potential field value). , It is the convergence factor (which determines the range of influence of the potential field). The lateral position of the vehicle. , The left and right boundaries of the road. For the first Lane markings location; Step 2.3.2: Model the potential field of the obstacle vehicle. The isosurface of the obstacle vehicle's potential field is an ellipse. Guide the vehicle to smoothly avoid obstacles along the outer contour of the ellipse. The formula is: ， in, The potential field coefficient for the obstacle vehicle. , For vertical and horizontal convergence factors, , For the first The location of the obstructing vehicle; Step 2.3.3: Integrating the safety potential field and safety cost, and integrating the road potential field and the obstacle vehicle potential field, we obtain the comprehensive safety potential field: ， in, Given the total number of vehicles with obstacles, the safety cost is defined as the time-domain normalized integral of the comprehensive potential field, ensuring that the safety of trajectories of different durations is comparable: 。 9. The emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 4, characterized in that, Furthermore, step 3.2 includes the following steps: Step 3.2.1, adaptive scheduling based on stagnation detection, using a sliding window (length) Calculate the optimal fitness improvement margin: ， in, ( (for the fitness function), if ( (For the stagnation threshold), global exploration is enhanced using Levi's flight: ， in, To explore step length, The Lévy distribution characteristic index is used; otherwise, local development is performed through Gaussian refinement. ， Refine the scale A linear decay strategy is adopted: ， in, This is the initial refinement scale; Step 3.2.2, Elite Learning Enhancement Mechanism, Maintaining Elite Memory Sets Before storage Find the optimal feasible solution and calculate the mean and variance of the elite set: ， Generate based on elite set statistical features One enhanced candidate solution: ， in, To control the diversity parameters, enhanced candidate solutions are incorporated into the population before selection, thereby improving convergence robustness.

10. An emergency obstacle avoidance method for multi-target trajectory planning and lateral stability tracking of intelligent connected vehicles according to claim 5, characterized in that, In step 4.1, linearization and control increment transformation are performed at the operating point. After Taylor expansion and retaining the first-order terms, we obtain the discrete linearized equation: ， in, , , To linearize the error, it is transformed into a function based on control increments. Extended form: ， in, , , , , .

Images