An adaptive obstacle avoidance method for an autonomous vehicle

Abstract

In order to complete the emergency obstacle avoidance operation under different road friction coefficients and vehicle driving speeds, improve the lateral dynamic stability of the automatic driving vehicle, the application discloses an automatic driving vehicle adaptive obstacle avoidance method. The method considers different road friction coefficients and vehicle speeds, generates corresponding expected obstacle avoidance speed by using fuzzy fitting method, and generates expected obstacle avoidance trajectory by using hybrid A* method, so as to reduce the steering burden of the vehicle during obstacle avoidance. A new steering enhancement method is constructed, a direct yaw turning controller is designed, a steering torque is generated, the dynamic stability of the vehicle during emergency obstacle avoidance is improved, a fuzzy speed controller based on BP neural network prediction is designed, and a driving torque is generated. The tracking accuracy of the expected obstacle avoidance speed is improved. The planning method provides a new solution for the emergency obstacle avoidance problem of the vehicle, and provides a new driving force for the development of the industry.

Description

Technical Field

[0001] This invention belongs to the field of path planning technology, and in particular relates to an adaptive obstacle avoidance method for autonomous vehicles. Background Technology

[0002] Autonomous vehicles can plan routes, track, and avoid collisions. Due to their faster reaction times, lower fuel consumption, and safer driving characteristics, autonomous vehicles can effectively improve road capacity and traffic safety. Significant progress has been made technologically thanks to years of investment in autonomous driving technology by researchers, governments, and automakers. However, the difficulties and challenges that need to be overcome remain enormous. The implementation of autonomous driving environments involves not only complex automotive technologies but also human behavior, ethics, and social responsibility.

[0003] From a technical perspective, many issues related to perception, navigation, and control must be addressed to ensure safe operation in everyday traffic or harsh off-road environments. Among these, obstacle avoidance in emergency situations is a fundamental technical problem. However, current obstacle avoidance algorithms do not consider the impact of road friction coefficient and vehicle speed on the vehicle. Consequently, when the friction coefficient is low or the autonomous vehicle's speed is high, it is easy to lose lateral dynamic stability, causing the vehicle to lose control and leading to serious traffic accidents. Summary of the Invention

[0004] To address the above technical problems, this invention provides an adaptive obstacle avoidance method for autonomous vehicles, which adapts to different road conditions and vehicle speeds, and improves the lateral dynamic stability of the vehicle during driving.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] An adaptive obstacle avoidance method for autonomous vehicles, comprising the following steps:

[0007] S100: Obtain the road friction coefficient for different road surfaces. The obstacle avoidance speed planner generates the corresponding expected obstacle avoidance speed based on fuzzy fitting and different road friction coefficients.

[0008] S200: Obtain the initial speed of the vehicle. The obstacle avoidance trajectory planner is based on the improved hybrid A* algorithm. It combines the generated desired obstacle avoidance speed, the initial speed of the vehicle and the friction coefficient of the road surface to generate the corresponding desired obstacle avoidance trajectory.

[0009] S300: Acquires vehicle parameter information, designs a direct yaw turning controller, calculates the required direct yaw torque based on the vehicle parameter information and the desired obstacle avoidance trajectory, and distributes the direct yaw torque to the motors in each wheel according to a preset torque distribution method to generate steering torque;

[0010] S400: Construct a fuzzy speed controller based on BP neural network prediction. The BP neural network is used to predict the vehicle speed error. The predicted speed error is used as the input of the fuzzy speed controller to calculate the driving torque.

[0011] S500: Adds the steering torque and drive torque to generate control torque, thus completing adaptive obstacle avoidance.

[0012] Preferably, S100 includes:

[0013] S110: Obtain the road friction coefficient of different road surfaces and set the fuzzy variables corresponding to the different road friction coefficients;

[0014] S120: By sampling through emergency obstacle avoidance experiments at different speeds, the expected obstacle avoidance speed under different friction coefficients in the experiment is obtained;

[0015] S130: By using Gaussian membership functions, the fuzzy variables corresponding to different road surface friction coefficients and the expected obstacle avoidance speeds under different friction coefficients in the experiment are fuzzified, and the mapping relationship between different road surface friction coefficients and expected obstacle avoidance speeds is established.

[0016] Preferably, S200 includes:

[0017] S210: Establish a node expansion model for the hybrid A* algorithm, where each node in the hybrid A* algorithm contains vehicle state information. Where x and y are the vehicle coordinates. It is the vehicle yaw angle, and each current node includes a preset number of extension nodes;

[0018] S220: Design a cost function f(n) of the hybrid A* algorithm, calculate the cost of all the preset number of extended nodes, and determine the node corresponding to the minimum cost as the next node. Combine the generated expected obstacle avoidance speed, the initial speed of the vehicle and the friction coefficient of the road surface to obtain the maximum front wheel steering angle from the current node to the next node.

[0019] S230: Take the next node as the current node, repeat S220, and generate the corresponding desired obstacle avoidance trajectory.

[0020] Preferably, S210 further includes:

[0021] Within the sampling period, let the extended value δ f Given a value, the vehicle's state at the next moment is obtained based on vehicle kinematics simulation, where δ f -δ fmax -δ fmax / 2、0、δ fmax / 2、δ fmaxWhen the vehicle is in the unfolded sub-node action, the five states are maximum left turn, half maximum left turn, no turn, half maximum right turn and maximum right turn.

[0022] Preferably, the cost function in S220 includes a cumulative cost function g(n) and a heuristic function h(n), as shown in the following formula:

[0023] f(n=g(n)+h(n)

[0024] Wherein, the cumulative cost function g(n) is the distance from the starting point to the current node, and the heuristic function h(n) is used to estimate the cost from the current node to the target node;

[0025] Suppose there are two parallel heuristic functions, and the maximum value of the two is taken as shown in the following equation:

[0026] h(n=max{h no_obs ,h obs}

[0027] Among them, h no_obs h is the shortest path length from the current node to the target node that satisfies kinematic constraints, ignoring obstacles. obs The length of the collision avoidance path that takes obstacle information into account but not the non-kinematic constraints of the vehicle.

[0028] Preferably, in S220, the maximum front wheel steering angle from the current node to the next node is obtained by combining the generated desired obstacle avoidance speed, the initial vehicle speed, and the road surface friction coefficient, specifically as follows:

[0029]

[0030] Where, δ fmax It is the maximum front wheel steering angle of the vehicle, μ is the road friction coefficient, and v I It is the initial speed of the vehicle, v xd It is the vehicle's desired obstacle avoidance speed.

[0031] Preferably, S300 includes:

[0032] S310: Acquire vehicle parameter information and perform lateral dynamics modeling of the vehicle:

[0033]

[0034] The yaw rate can be updated in the following ways:

[0035]

[0036] Where, m v It is the mass of the vehicle, F yrl It is the longitudinal force F acting on the left rear tire.yrr It is the longitudinal force F acting on the right rear tire. yfr It is the longitudinal force F acting on the right front tire. yfl It is the longitudinal force F acting on the left front tire. xfr It is the lateral force F acting on the right front tire. xfl γ is the lateral force on the left front tire, γ is the yaw rate, δ is the front wheel steering angle, and l z It is the vehicle's moment of inertia, l f and l r It is the front and rear wheelbase, d f and d r These are the front track and rear track, respectively. (M) z It is the direct yaw torque, Δd1 and Δd2 represent the corresponding suspension and aerodynamic effects, respectively, v x and v y These are the vehicle's lateral and longitudinal speeds;

[0037] S320: The direct yaw torque is distributed to the motors in each wheel according to a preset torque distribution method to generate steering torque. The preset torque distribution method is expressed as follows:

[0038]

[0039]

[0040]

[0041]

[0042] Among them, F zfl It is the vertical load of the left front tire, F zfr It is the vertical load of the right front tire, F z It is the sum of all vertical loads, 'a' is the vehicle acceleration, 'R' is the wheel radius, and 'T' is the total vertical load. mfl It is the steering torque of the left front tire, T mfr It is the steering torque of the right front tire, T mrl It is the steering torque of the left rear tire, T mrr It is the steering torque of the right rear tire;

[0043] Preferably, S400 includes:

[0044] S410: The input vector of a BP neural network is defined as follows:

[0045] x=[β,δ,v x ,v y ,γ,T s ]

[0046] The output vector is represented as:

[0047]

[0048] The fuzzy velocity controller can be considered as a black-box function, denoted as:

[0049] y p =Net(x)

[0050] Where Net(·) is a BP neural network, β is the sideslip angle, δ is the front wheel steering angle, and v x and v y These are the vehicle's lateral and longitudinal velocities, γ is the yaw rate, and T is the lateral and longitudinal velocities. s It is the vehicle's control torque;

[0051] The sideslip angle is updated using the following formula:

[0052]

[0053] Control error e pc The definition is as follows:

[0054]

[0055] Where, k n This represents the compensation factor for the nth prediction step, where N is the total number of prediction steps, and e p (n) represents the predictive control at n th The error at the predicted step size is obtained by the following formula:

[0056] e p (k+n)=v xd (k+n)-v xp (k+n)

[0057] Among them, v xp (k+n) represents the predicted longitudinal speed of the vehicle in the nth prediction step within the k sampling period, v xp It can be obtained through the following two methods:

[0058] When n=1, v xp The calculation is as follows:

[0059] v xp (k+1)=Net3[β(k),δ(k),v x (k),v y (k),γ(k),T s ]

[0060] Where Net3(·) is the three-dimensional output of Net(·), and based on this prediction result, the predicted torque T to be used in the next prediction step is... p for:

[0061] Tp =FIS(e p ,ec p )

[0062] Where FIS(·) is a fuzzy inference system, e p =v xd -v xp e p It is the corresponding differential;

[0063] e p (k+n)=v xd (k+n)-v xp (k+n)

[0064] When n≠1, v xp The calculation is as follows:

[0065] v xp (k+n)=Net3[β(m),δ(m0,v x (m),v y (m),γ(m),T p (m)]

[0066] Where m = k + n - 1, and n ≥ 2;

[0067] S420: Design a fuzzy inference mechanism, calculate speed error compensation, and denote the input of the fuzzy controller as [e v ,ec v The corresponding output variable is T. s ; where e v For longitudinal speed control error, ec v It is e v The differential, T s It is the driving torque output by the fuzzy controller;

[0068] S430: Utilizing e pc As the final control error, the final driving torque T is obtained. s :

[0069] T s =FIS(e pc ,ec pc )

[0070] Among them, ec pc It is e pc The differential.

[0071] Preferably, the input and output variables in S420 are normalized to a specific range using the following formula:

[0072]

[0073] Where, r ev , and They are e v , ec v and T s The normalization factors are 8, 8, and 6, respectively.

[0074] Preferably, S500 specifically comprises:

[0075] T fl =T mfl +T s

[0076] T fr =T mfr +T s

[0077] T rl =T mrl +T s

[0078] T rr =T mrr +T s

[0079] Among them, T fl It is the control torque of the left front tire, T fr It is the control torque of the right front tire, T rl It is the control torque of the left rear tire, T rr It is the control torque of the right rear tire, T s It is the driving torque, T mfl It is the steering torque of the left front tire, T mfr It is the steering torque of the right front tire, T mrl It is the steering torque of the left rear tire, T mrr It is the steering torque of the right rear tire.

[0080] To improve the lateral dynamic stability of autonomous vehicles and facilitate emergency obstacle avoidance under varying road surface friction coefficients and vehicle speeds, this invention proposes an adaptive obstacle avoidance method for autonomous vehicles. This method considers different road surface friction coefficients and vehicle speeds, employing a fuzzy fitting method to generate the desired obstacle avoidance speed and a hybrid A* method to generate the desired obstacle avoidance trajectory, thus reducing the steering load during obstacle avoidance. A novel steering enhancement method is constructed, designing a direct yaw-turn controller to generate steering torque, improving the vehicle's dynamic stability during emergency obstacle avoidance. A fuzzy speed controller based on BP neural network prediction is designed to generate driving torque, improving the tracking accuracy of the desired obstacle avoidance speed. This planning method offers a novel solution to the vehicle emergency obstacle avoidance problem, providing new impetus for the development of this industry. Attached Figure Description

[0081] Figure 1 This is a flowchart of an adaptive obstacle avoidance method for an autonomous vehicle according to an embodiment of the present invention;

[0082] Figure 2 This is a flowchart of an adaptive obstacle avoidance method for autonomous vehicles according to an embodiment of the present invention;

[0083] Figure 3 This is a schematic diagram of vehicle lateral dynamic instability in one embodiment of the present invention;

[0084] Figure 4 This is a schematic diagram of the hybrid A* extension principle in one embodiment of the present invention. Detailed Implementation

[0085] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0086] In one embodiment, such as Figure 1 and 2 As shown, an adaptive obstacle avoidance method for autonomous vehicles includes the following steps:

[0087] S100: Obtain the road friction coefficient for different road surfaces. The obstacle avoidance speed planner generates the corresponding expected obstacle avoidance speed based on fuzzy fitting and different road friction coefficients.

[0088] S200: Obtain the initial speed of the vehicle. The obstacle avoidance trajectory planner is based on the improved hybrid A* algorithm. It combines the generated desired obstacle avoidance speed, the initial speed of the vehicle and the friction coefficient of the road surface to generate the corresponding desired obstacle avoidance trajectory.

[0089] S300: Acquires vehicle parameter information, designs a direct yaw turning controller, calculates the required direct yaw torque based on the vehicle parameter information and the desired obstacle avoidance trajectory, and distributes the direct yaw torque to the motors in each wheel according to a preset torque distribution method to generate steering torque;

[0090] S400: Construct a fuzzy speed controller based on BP neural network prediction. The BP neural network is used to predict the vehicle speed error. The predicted speed error is used as the input of the fuzzy speed controller to calculate the driving torque.

[0091] S500: Adds the steering torque and drive torque to generate control torque, thus completing adaptive obstacle avoidance.

[0092] Specifically, see Figure 3When a vehicle performs emergency obstacle avoidance, due to slippery road surfaces or high vehicle speeds, the steering radius increases significantly, causing the vehicle to lose lateral dynamic stability. This severely impacts driving safety and can lead to traffic accidents. To improve the lateral dynamic stability of autonomous vehicles and enable emergency obstacle avoidance under varying road friction coefficients and vehicle speeds, this invention proposes an adaptive obstacle avoidance method for autonomous vehicles. This method considers different road friction coefficients and vehicle speeds, using a fuzzy fitting method to generate the desired obstacle avoidance speed and a hybrid A* method to generate the desired obstacle avoidance trajectory, reducing the steering load during obstacle avoidance. A novel steering enhancement method is constructed, designing a direct yaw turning controller to generate steering torque, improving the vehicle's dynamic stability during emergency obstacle avoidance. A fuzzy speed controller based on BP neural network prediction is designed to generate driving torque, improving the tracking accuracy of the desired obstacle avoidance speed. This planning method offers a new solution to the vehicle emergency obstacle avoidance problem, providing new impetus for the development of this industry.

[0093] Furthermore, the trajectory planning part generates the corresponding expected obstacle avoidance trajectory based on the road surface and vehicle information, including two parts: an obstacle avoidance speed planner and an obstacle avoidance trajectory planner. The obstacle avoidance speed planner is based on fuzzy fitting and generates the corresponding expected obstacle avoidance speed according to different road surface friction coefficients. The obstacle avoidance trajectory planner is based on an improved hybrid A* algorithm and combines the generated expected obstacle avoidance speed, the vehicle's initial speed, and the road surface friction coefficient to generate the corresponding expected obstacle avoidance trajectory.

[0094] In one embodiment, S100 includes:

[0095] S110: Obtain the road friction coefficient of different road surfaces and set the fuzzy variables corresponding to the different road friction coefficients;

[0096] S120: By sampling through emergency obstacle avoidance experiments at different speeds, the expected obstacle avoidance speed under different friction coefficients in the experiment is obtained;

[0097] S130: By using Gaussian membership functions, the fuzzy variables corresponding to different road surface friction coefficients and the expected obstacle avoidance speeds under different friction coefficients in the experiment are fuzzified, and the mapping relationship between different road surface friction coefficients and expected obstacle avoidance speeds is established.

[0098] In one embodiment, S200 includes:

[0099] S210: Establish a node expansion model for the hybrid A* algorithm, where each node in the hybrid A* algorithm contains vehicle state information. Where x and y are the vehicle coordinates. It is the vehicle yaw angle, and each current node includes a preset number of extension nodes;

[0100] S220: Design the cost function f(n) of the hybrid A* algorithm, calculate the cost of all the preset number of expansion nodes, and determine the node corresponding to the minimum cost as the next node. Combine the generated expected obstacle avoidance speed, the initial speed of the vehicle and the friction coefficient of the road surface to obtain the maximum front wheel steering angle from the current node to the next node.

[0101] S230: Take the next node as the current node, repeat S220, and generate the corresponding desired obstacle avoidance trajectory.

[0102] Specifically, the hybrid A* algorithm node expansion model is as follows: Figure 4 As shown.

[0103] In one embodiment, S210 further includes:

[0104] Within the sampling period, let the extended value δ f Given a value, the vehicle's state at the next moment is obtained based on vehicle kinematics simulation, where δ f -δ fmax -δ fmax / 2、0、δ fmax / 2、δ fmax When the vehicle is in the unfolded sub-node action, the five states are maximum left turn, half maximum left turn, no turn, half maximum right turn and maximum right turn.

[0105] In one embodiment, the cost function in S220 includes a cumulative cost function g(n) and a heuristic function h(n), as shown in the following formula:

[0106] f(n) = g(n) + h(n)

[0107] Wherein, the cumulative cost function g(n) is the distance from the starting point to the current node, and the heuristic function h(n) is used to estimate the cost from the current node to the target node;

[0108] Suppose there are two parallel heuristic functions, and the maximum value of the two is taken as shown in the following equation:

[0109] h(n) = max{h no_obs ,h obs}

[0110] Among them, h no_obs h is the shortest path length from the current node to the target node that satisfies kinematic constraints, ignoring obstacles. obs The length of the collision avoidance path that takes obstacle information into account but not the non-kinematic constraints of the vehicle.

[0111] In one embodiment, in S220, the maximum front wheel steering angle from the current node to the next node is obtained by combining the generated desired obstacle avoidance speed, the initial vehicle speed, and the road surface friction coefficient. Specifically:

[0112]

[0113] Where, δ fmax It is the maximum front wheel steering angle of the vehicle, μ is the road friction coefficient, and v I It is the initial speed of the vehicle, v xd It is the vehicle's desired obstacle avoidance speed.

[0114] Specifically, an adaptive constraint method for the front wheel steering angle is proposed, which adjusts the maximum steering angle of the front wheels according to the road surface and vehicle speed to avoid the problem of lateral dynamic instability of autonomous vehicles.

[0115] Furthermore, after trajectory planning is completed, trajectory tracking is performed. The trajectory tracking function tracks the generated desired obstacle avoidance trajectory and includes two parts: a steering enhancement method and a fuzzy speed controller. The steering enhancement method uses a direct yaw steering controller and a torque distribution method to generate a steering torque T. m This improves vehicle dynamic stability; the fuzzy speed controller uses a BP neural network to predict the speed error, and uses the predicted error as the input to the fuzzy controller to calculate the driving torque T. s This improves the tracking accuracy for the desired obstacle avoidance speed.

[0116] In one embodiment, S300 includes:

[0117] S310: Acquire vehicle parameter information and perform lateral dynamics modeling of the vehicle:

[0118]

[0119] The yaw rate can be updated in the following ways:

[0120]

[0121] Where, m v It is the mass of the vehicle, F yrl It is the longitudinal force F acting on the left rear tire. yrr It is the longitudinal force F acting on the right rear tire. yfr It is the longitudinal force F acting on the right front tire. yfl It is the longitudinal force F acting on the left front tire. xfr It is the lateral force F acting on the right front tire. xfl γ is the lateral force on the left front tire, γ is the yaw rate, δ is the front wheel steering angle, and I is the lateral force. z It is the vehicle's moment of inertia, l f and lr It is the front and rear wheelbase, d f and d r These are the front track and rear track, respectively. (M) z It is the direct yaw torque, Δd1 and Δd2 represent the corresponding suspension and aerodynamic effects, respectively, v x and v y These are the vehicle's lateral and longitudinal speeds;

[0122] S320: The direct yaw torque is distributed to the motors in each wheel according to a preset torque distribution method to generate steering torque. The preset torque distribution method is expressed as follows:

[0123]

[0124]

[0125]

[0126]

[0127] Among them, F zfl It is the vertical load of the left front tire, F zfl It is the vertical load of the right front tire, F z It is the sum of all vertical loads, 'a' is the vehicle acceleration, 'R' is the wheel radius, and 'T' is the total vertical load. mfl It is the steering torque of the left front tire, T mfr It is the steering torque of the right front tire, R mrl It is the steering torque of the left rear tire, T mrr It is the steering torque of the right rear tire.

[0128] Furthermore, a BP neural network trained offline is used to predict the velocity error, and the prediction is achieved through online fine-tuning using the backpropagation algorithm. This prediction compensation method aims to improve the control performance of the fuzzy controller.

[0129] In one embodiment, S400 includes:

[0130] S410: The input vector of a BP neural network is defined as follows:

[0131] x=[β,δ,v x ,v y ,γ,T s ]

[0132] The output vector is represented as:

[0133]

[0134] The fuzzy velocity controller can be considered as a black-box function, denoted as:

[0135] y p =Net(x)

[0136] Where Net(·) is a BP neural network, β is the sideslip angle, δ is the front wheel steering angle, and v x and v y These are the vehicle's lateral and longitudinal velocities, γ is the yaw rate, and T is the lateral and longitudinal velocities. s It is the vehicle's control torque;

[0137] The sideslip angle is updated using the following formula:

[0138]

[0139] Control error e pc The definition is as follows:

[0140]

[0141] Where, k n This represents the compensation factor for the nth prediction step, where N is the total number of prediction steps, and e p (n) represents the predictive control at n th The error at the predicted step size is obtained by the following formula:

[0142] e p (k+n)=v xd (k+n)-v xp (k+n)

[0143] Among them, v xp (k+n) represents the predicted longitudinal speed of the vehicle in the nth prediction step within the k sampling period, v xp It can be obtained through the following two methods:

[0144] When n=1, v xp The calculation is as follows:

[0145] v xp (k+1)=Net3[β(k),δ(k),v x (k),v y (k),γ(k),T s ]

[0146] Where Net3(·) is the three-dimensional output of Net(·), and based on this prediction result, the predicted torque T to be used in the next prediction step is... p for:

[0147] T p =FIS(e p ,ec p )

[0148] Where FIS(·) is a fuzzy inference system, e p =v xd -v xp e p It is the corresponding differential;

[0149] e p (k+n)=v xd (k+n)-v xp (k+n)

[0150] When n≠1, v xp The calculation is as follows:

[0151] v xp (k+n)=Net3[β(m),δ(m),v x (m),v y (m),γ(m),T p (m)]

[0152] Where m = k + n - 1, and n ≥ 2;

[0153] S420: Design a fuzzy inference mechanism, calculate speed error compensation, and denote the input of the fuzzy controller as [e v ,ec v The corresponding output variable is T. s ; where e v For longitudinal speed control error, ec v It is e v The differential, T s It is the driving torque output by the fuzzy controller;

[0154] S430: Utilizing e pc As the final control error, the final driving torque T is obtained. s :

[0155] T s =FIS(e pc ,ec pc )

[0156] Among them, ec pc It is e pc The differential.

[0157] In one embodiment, the input and output variables in S420 are normalized to a specific range using the following formula:

[0158]

[0159] Where, r ev , and They are e v , ecv and T s The normalization factors are 8, 8, and 6, respectively.

[0160] Set the following fuzzy rules:

[0161]

[0162] In one embodiment, S500 specifically refers to:

[0163] T fl =T mfl +T s

[0164] T fr =T mfr +T s

[0165] T rl =T mrl +T s

[0166] T rr =T mrr +T s

[0167] Among them, T fl It is the control torque of the left front tire, T fr It is the control torque of the right front tire, T rl It is the control torque of the left rear tire, T rr It is the control torque of the right rear tire, T s It is the driving torque, T mfl It is the steering torque of the left front tire, T mfr It is the steering torque of the right front tire, T mrl It is the steering torque of the left rear tire, T mrr It is the steering torque of the right rear tire.

[0168] To improve the lateral dynamic stability of autonomous vehicles and facilitate emergency obstacle avoidance under varying road surface friction coefficients and vehicle speeds, this invention proposes an adaptive obstacle avoidance method for autonomous vehicles. This method considers different road surface friction coefficients and vehicle speeds, employing a fuzzy fitting method to generate the desired obstacle avoidance speed and a hybrid A* method to generate the desired obstacle avoidance trajectory, thus reducing the steering load during obstacle avoidance. A novel steering enhancement method is constructed, designing a direct yaw-turn controller to generate steering torque, improving the vehicle's dynamic stability during emergency obstacle avoidance. A fuzzy speed controller based on BP neural network prediction is designed to generate driving torque, improving the tracking accuracy of the desired obstacle avoidance speed. This planning method offers a novel solution to the vehicle emergency obstacle avoidance problem, providing new impetus for the development of this industry.

[0169] The above provides a detailed description of an adaptive obstacle avoidance method for autonomous vehicles provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. An adaptive obstacle avoidance method for autonomous vehicles, characterized in that, The method includes the following steps: S100: Obtain the road friction coefficient for different road surfaces. The obstacle avoidance speed planner generates the corresponding expected obstacle avoidance speed based on fuzzy fitting and different road friction coefficients. S200: Obtain the initial vehicle speed. The obstacle avoidance trajectory planner, based on the improved hybrid A* algorithm, combines the generated desired obstacle avoidance speed, the initial vehicle speed, and the road surface friction coefficient to generate the corresponding desired obstacle avoidance trajectory. S200 includes: S210: Establish a hybrid A* algorithm node expansion model, wherein each node of the hybrid A* algorithm contains vehicle state information. ,in These are vehicle coordinates. It is the vehicle yaw angle, and each current node includes a preset number of extension nodes; S220: Designing the cost function of a hybrid A* algorithm Calculate the cost of all preset number of expansion nodes, and determine the node corresponding to the minimum cost as the next node. Combine the generated expected obstacle avoidance speed, the initial speed of the vehicle, and the road friction coefficient to obtain the maximum front wheel steering angle from the current node to the next node. S230: Take the next node as the current node, repeat S220, and generate the corresponding desired obstacle avoidance trajectory; S300: Obtain vehicle parameter information, design a direct yaw turning controller, calculate the required direct yaw torque based on the vehicle parameter information and the desired obstacle avoidance trajectory, and distribute the direct yaw torque to the motors in each wheel according to a preset torque distribution method to generate steering torque; S400: Construct a fuzzy speed controller based on BP neural network prediction. The BP neural network is used to predict the vehicle speed error. The predicted speed error is used as the input of the fuzzy speed controller to calculate the driving torque. S500: The steering torque and the driving torque are added together to generate the control torque, thereby completing adaptive obstacle avoidance.

2. The method according to claim 1, characterized in that, S100 includes: S110: Obtain the road friction coefficient of different road surfaces and set the fuzzy variables corresponding to the different road friction coefficients; S120: By sampling through emergency obstacle avoidance experiments at different speeds, the expected obstacle avoidance speed under different friction coefficients in the experiment is obtained; S130: The fuzzy variables corresponding to the different road surface friction coefficients and the expected obstacle avoidance speeds under different friction coefficients in the experiment are fuzzified by using Gaussian membership functions to establish a mapping relationship between different road surface friction coefficients and expected obstacle avoidance speeds.

3. The method according to claim 2, characterized in that, S210 also includes: Within the sampling period, set the extended value. Given a value, the vehicle's state at the next moment is obtained based on vehicle kinematics simulation, where, when They are respectively When the vehicle is in the unfolded sub-node action, the five states are maximum left turn, half maximum left turn, no turn, half maximum right turn and maximum right turn.

4. The method according to claim 3, characterized in that, The cost function in S220 includes a cumulative cost function. and heuristic functions The formula is as follows: Wherein, the cumulative cost function It is the distance from the starting point to the current node, a heuristic function. Used to estimate the cost from the current node to the target node; Suppose there are two parallel heuristic functions, and the maximum value of the two is taken as shown in the following equation: in, The shortest path length from the current node to the target node that satisfies kinematic constraints, ignoring obstacles. The length of the collision avoidance path that takes obstacle information into account but not the non-kinematic constraints of the vehicle.

5. The method according to claim 4, characterized in that, In S220, the maximum front wheel steering angle from the current node to the next node is obtained by combining the generated desired obstacle avoidance speed, the initial speed of the vehicle, and the road surface friction coefficient. Specifically: in, It is the maximum front wheel steering angle of the vehicle. It is the road surface friction coefficient. It is the vehicle's initial speed. It is the vehicle's desired obstacle avoidance speed.

6. The method according to claim 5, characterized in that, The S300 includes: S310: Acquire vehicle parameter information and perform lateral dynamics modeling of the vehicle: The yaw rate can be updated in the following ways: in, It's about the quality of the vehicle. It is the longitudinal force acting on the left rear tire. It is the longitudinal force acting on the right rear tire. It is the longitudinal force on the right front tire. It is the longitudinal force on the left front tire. It is the lateral force acting on the right front tire. It is the lateral force acting on the left front tire. It is the yaw rate. It is the front wheel steering angle. It is the vehicle's moment of inertia. and It refers to the front and rear wheelbase. and These are the front track and the rear track, It is the direct yaw torque. and These represent the corresponding suspension and aerodynamic effects, and These are the vehicle's lateral and longitudinal speeds; S320: The direct yaw torque is distributed to the in-wheel motors according to a preset torque distribution method to generate steering torque. The preset torque distribution method is expressed as follows: in, It is the vertical load on the left front tire. It is the vertical load on the right front tire. It is the sum of all vertical loads. It is vehicle acceleration. It is the radius of the wheel. It is the steering torque of the left front tire. It is the steering torque of the right front tire. It is the steering torque of the left rear tire. It is the steering torque of the right rear tire.

7. The method according to claim 6, characterized in that, The S400 includes: S410: The input vector of a BP neural network is defined as follows: The output vector is represented as: The fuzzy velocity controller can be considered as a black-box function, denoted as: in, It is a backpropagation (BP) neural network, in which, It is the sideslip angle. It is the front wheel steering angle. and These are the vehicle's lateral and longitudinal speeds. It is the yaw rate. It is the vehicle's control torque; The sideslip angle is updated using the following formula: Control error The definition is as follows: in, Indicates the first The compensation factor for each prediction step, where N is the total number of prediction steps. For predictive control The error at the predicted step size is obtained by the following formula: in, For the first Each prediction step in Predicted longitudinal speed of vehicles within the sampling period It can be obtained through the following two methods: When n=1, The calculation is as follows: in, yes Based on the three-dimensional output of the prediction result, the predicted torque to be used in the next prediction step is determined. for: in, For fuzzy reasoning systems, , It is the corresponding differential; When n≠1, The calculation is as follows: in, ,and ; S420: Design a fuzzy inference mechanism, calculate speed error compensation, and denote the input of the fuzzy controller as... The corresponding output variable is ;in, For longitudinal speed control error, yes The differential, It is the driving torque output by the fuzzy controller; S430: Utilizing As the final control error, the final driving torque is obtained. : in, yes The differential.

8. The method according to claim 7, characterized in that, The input and output variables in S420 are normalized to a specific range using the following formula: in, , and They are , and The normalization factors are 8, 8, and 6, respectively.

9. The method according to claim 8, characterized in that, The S500 specifically refers to: in, It is the control torque of the left front tire. It is the control torque of the right front tire. It is the control torque of the left rear tire. It is the control torque of the right rear tire. It is the driving torque. It is the steering torque of the left front tire. It is the steering torque of the right front tire. It is the steering torque of the left rear tire. It is the steering torque of the right rear tire.

Images