A method for speed control of autonomous vehicles based on environmental complexity

Abstract

This invention discloses a vehicle speed control method for autonomous vehicles based on environmental complexity. The control method executes the following four steps sequentially with a cycle time of less than 50ms: Step 1: Simultaneously collect road geometry, traffic flow density, visibility distance, road surface adhesion, and meteorological data, and fuse them using a Kalman filter algorithm to obtain an environmental parameter vector; Step 2: Input this vector into a two-layer evaluation network, which outputs four levels of environmental complexity labels; Step 3: Input this vector into a control model constructed based on a hierarchical adaptive complexity speed control algorithm to generate a target vehicle speed; Step 4: If the target vehicle speed is greater than or equal to a collision risk index, immediately increase the environmental complexity level and return to Step 3 to recalculate the target vehicle speed until the target vehicle speed is less than the collision risk index, at which point a vehicle speed execution command is output. This invention significantly improves vehicle safety and driving efficiency, and ensures the stability and reliability of the autonomous driving system.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method for controlling the speed of autonomous vehicles based on environmental complexity. Background Technology

[0002] In recent years, with the rapid development of autonomous driving technology, the accuracy, safety and real-time performance of vehicle speed control during autonomous driving have become research hotspots. Traditional vehicle speed control methods are usually based on a single or few parameters, such as speed limit information or simple obstacle detection, which cannot accurately cope with complex and ever-changing traffic scenarios. In order to improve driving safety and efficiency, vehicle speed control methods are gradually shifting from static, single-rule control to dynamic, multi-source data fusion control.

[0003] CN111376912A discloses a speed control method for autonomous vehicles. This method determines the corresponding maximum speed based on information about the driving scenario, driving environment, and road conditions, and then selects the minimum speed as the final speed. Although it considers many factors, it only uses the minimum speed as a safety guarantee and fails to dynamically adjust according to the complexity of the environment, resulting in the vehicle speed not responding to environmental changes in a timely manner in some scenarios.

[0004] CN113689721A discloses a vehicle speed control method based on traffic flow prediction at traffic light intersections. This method uses a neural network model to predict traffic flow and optimize the speed of autonomous vehicles at intersections. While this method effectively improves the efficiency of traffic intersections, its focus is mainly on energy saving and speed optimization at intersections. It has shortcomings in safety control and real-time risk response in complex road environments. In real-world complex environments, road geometry is diverse, traffic flow density changes drastically, and weather conditions are unstable. This method lacks dynamic, real-time, and accurate environmental perception and assessment methods, resulting in delayed vehicle speed adjustments and potential safety hazards or reduced comfort.

[0005] In summary, existing vehicle speed control technologies suffer from the inability to accurately adapt to complex environmental changes. This invention proposes a vehicle speed control method for autonomous vehicles based on environmental complexity. Addressing the deficiencies in the real-time performance and accuracy of environmental information perception and assessment, this method dynamically assesses environmental complexity through multi-source sensor data fusion, fuzzy logic, and meta-learning methods. Real-time risk indicators are used to provide feedback and perform closed-loop control adjustments to the target vehicle speed, thus solving the key technical challenge of real-time safe vehicle speed control in complex driving environments. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: the control method executes the following four steps sequentially with a cycle period of less than 50ms, including:

[0009] Step 1: Simultaneously collect road geometry, traffic flow density, visibility distance, road surface adhesion, and meteorological data through lidar, millimeter-wave radar, camera, vehicle-road cooperative communication device, and inertial measurement unit. After alignment in the time domain, use Kalman filter algorithm to fuse them to obtain environmental parameter vectors.

[0010] Step 2: Input the environmental parameter vector into a two-layer evaluation network consisting of a fuzzy logic layer and a meta-learning layer. First, fuzzify the parameters of the same category. Then, use a meta-policy optimization algorithm with safety margin as the feedback signal to update the weights of each parameter online. Finally, output four levels of environmental complexity labels, representing four driving scenarios: simple, general, complex, and high-risk.

[0011] Step 3: Input the environmental complexity label into the control model constructed based on the hierarchical adaptive complexity vehicle speed control algorithm to generate the target vehicle speed;

[0012] Step 4: If the target vehicle speed is greater than or equal to the collision risk index, the environmental complexity level is immediately increased and the process returns to Step 3 to recalculate the target vehicle speed until the target vehicle speed is less than the collision risk index. Then, a vehicle speed execution command is output.

[0013] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the step of fusing environmental parameter vectors using the Kalman filter algorithm includes:

[0014] Based on the vehicle's global positioning information and time source, the raw data from each sensor are aligned frame by frame.

[0015] The current prior environmental state is calculated based on the environmental state and vehicle motion state output from the previous cycle.

[0016] Four types of observations—road geometry, traffic flow, meteorology, and road surface adhesion—are introduced sequentially, and Kalman gain is used to weight and correct the current environmental prior state.

[0017] The corrected state vector is processed by outlier removal and coordinate system unification, and the output includes an environmental parameter vector containing road curvature, number of available lanes, lane width, average traffic speed, traffic density, distance to the nearest vehicle, visibility distance, road surface friction level, road surface humidity, precipitation intensity, light intensity, lateral wind speed, and ambient temperature.

[0018] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the environmental parameter vector is input into a two-layer evaluation network, and the output environmental complexity label includes:

[0019] Road geometry and traffic flow observations are used as the first set of inputs, and meteorological and pavement adhesion observations are used as the second set of inputs, and fuzzification is performed on each of them.

[0020] Using the safety margin as a real-time feedback signal, the weights of the two sets of inputs are adjusted through a meta-policy optimization algorithm;

[0021] The fuzzy output is then weighted and compared with a preset complexity classification rule to generate an environment complexity label.

[0022] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the preset complexity classification rule includes:

[0023] When the weighted output comprehensive score is lower than the first threshold, it is marked as a simple driving scenario, allowing the vehicle to drive at a speed close to the road speed limit.

[0024] When the weighted output comprehensive score is between the first threshold and the second threshold, it is marked as a general driving scenario, and the vehicle speed is limited to 80% of the road speed limit.

[0025] When the weighted output comprehensive score is between the second and third thresholds, it is marked as a complex driving scenario, and the vehicle speed is limited to 60% of the road speed limit.

[0026] When the weighted output comprehensive score is higher than the third threshold, it is marked as a high-risk driving scenario, and the braking preload and distance keeping tasks are forcibly activated.

[0027] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the two-layer evaluation network specifically includes:

[0028] The fuzzy logic layer is used to map each environmental parameter to a fuzzy set according to a preset membership function;

[0029] The meta-learning layer is used to dynamically adjust the weights of the fuzzy set based on the safety margin feedback and through a policy optimization mechanism.

[0030] The classification decision unit is used to weight and summarize the corrected fuzzy output to obtain a comprehensive score, and output the environmental complexity label according to the classification rules.

[0031] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, step three specifically includes:

[0032] The smaller value between the road speed limit and the vehicle dynamic limit speed is selected as the benchmark safe speed.

[0033] The environmental complexity label is multiplied by the corresponding scaling factor on the baseline safe speed, where 100% is used for simple driving scenarios, 80% for general driving scenarios, 60% for complex driving scenarios, and 40% for high-risk driving scenarios.

[0034] The scaled vehicle speed is input into a smoothing function to eliminate speed jumps between adjacent cycle periods.

[0035] In high-risk driving scenarios, an additional braking preload command is applied, and the final target vehicle speed is output.

[0036] As a preferred embodiment of the autonomous vehicle speed control method based on environmental complexity described in this invention, the generation of the target vehicle speed specifically includes:

[0037] When the environmental complexity label is simple, the smaller value between the maximum permissible vehicle speed and the vehicle dynamic limit speed is selected as the target vehicle speed.

[0038] When the environmental complexity label is "general scenario", the smaller value is multiplied by 80% to obtain the target vehicle speed.

[0039] When the environmental complexity label is "complex scene", the smaller value is multiplied by 60% to obtain the target vehicle speed.

[0040] When the environmental complexity is labeled as a high-risk scenario, the smaller value is multiplied by 40% to obtain the target vehicle speed, and the braking preload task is activated in advance.

[0041] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, step four specifically includes:

[0042] In each cycle, calculate the remaining braking distance, the difference between the estimated collision time and the system braking response time;

[0043] The target vehicle speed is compared with the remaining braking distance, the difference between the estimated collision time and the difference between the system braking response time. If any of the danger quantities reaches the danger threshold, a collision risk is triggered.

[0044] If a collision risk is triggered and the current environmental complexity level has not yet reached a high risk level, immediately raise the environmental complexity level by one level and return to step three to recalculate the target vehicle speed.

[0045] If the target vehicle speed cannot be lowered below the collision risk index after two consecutive increases, a downgraded safety mode will be activated, instructing the vehicle to decelerate to half the road speed limit and periodically checking the collision risk index until it is cleared.

[0046] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the collision risk indicators include at least the remaining braking distance, the expected collision time difference, the closing speed between the vehicle and the obstacle in front, the road adhesion level correction coefficient, and the difference in system braking response time.

[0047] The remaining braking distance, the expected collision time difference, the closing speed of the vehicle and the obstacle ahead, the road adhesion level correction coefficient, and the difference of the system braking response time are normalized according to preset weights, weighted and summed, and then compared with the danger threshold to determine the risk level.

[0048] As a preferred embodiment of the vehicle speed control method for autonomous vehicles based on environmental complexity described in this invention, the vehicle speed execution command includes at least a drive torque request, a braking torque request, a regenerative braking ratio, a gear or torque distribution request, and a vehicle-road cooperative risk notification.

[0049] The vehicle speed execution command is sent to each actuator within 10ms after being generated by the real-time scheduling system, and execution confirmation is received within 5ms after being sent. If no confirmation is received within the time limit, the system switches to the degraded safety mode.

[0050] In the degraded safety mode, the driving torque is limited to less than 30% of the road speed limit, the braking torque is kept continuously output, and the vehicle safety status and environmental complexity level information are sent to the remote monitoring platform once per second until the risk is eliminated.

[0051] The beneficial effects of this invention are as follows: This invention comprehensively realizes the precise fusion of environmental parameters, dynamic adaptive assessment of complexity, real-time dynamic planning of vehicle speed, and proactive closed-loop intervention for risks; it effectively solves the problems of insufficient environmental perception accuracy, coarse classification of complex environments, static control strategies, and lack of dynamic intervention for collision risks in traditional autonomous driving speed control methods. While significantly improving vehicle safety and driving efficiency, it ensures the stability, reliability, and comfort of the autonomous driving system. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0053] Figure 1 This is a flowchart illustrating the vehicle speed control method for autonomous vehicles based on environmental complexity as shown in this invention. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0055] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a vehicle speed control method for autonomous vehicles based on environmental complexity. This control method executes the following four steps sequentially with a cycle time of less than 50ms, specifically including the following steps:

[0058] Step 1: Simultaneously collect road geometry, traffic flow density, visibility distance, road surface adhesion, and meteorological data using lidar, millimeter-wave radar, cameras, vehicle-to-infrastructure communication devices, and inertial measurement units. After time-domain alignment, fuse these data using a Kalman filter algorithm to obtain an environmental parameter vector. Note that the following points should be noted in this step:

[0059] Based on the vehicle's global positioning information and time source, the raw data from each sensor are aligned frame by frame.

[0060] The current prior environmental state is calculated based on the environmental state and vehicle motion state output from the previous cycle.

[0061] Four types of observations are introduced in sequence: road geometry, traffic flow, meteorology and road surface adhesion. Kalman gain is used to weight and correct the current prior environmental state.

[0062] The corrected state vector is processed by outlier removal and coordinate system unification, and the output includes an environmental parameter vector containing road curvature, number of available lanes, lane width, average traffic speed, traffic density, distance to the nearest vehicle, visibility distance, road surface friction level, road surface humidity, precipitation intensity, light intensity, lateral wind speed, and ambient temperature.

[0063] It should be further explained that at the beginning of the current cycle, based on the environmental state, vehicle speed, acceleration and heading changes output by the previous cycle, the road curvature, drivable lane boundaries, traffic density and weather evolution trends are extrapolated using the assumption of uniform motion to obtain the current environmental prior state and its confidence level.

[0064] Four types of observations are introduced in sequence: road geometry, traffic flow, meteorology, and road surface adhesion;

[0065] For each category of observations, the difference between the observed value and the prior state is calculated, and the magnitude of the difference and the reliability of each sensor are combined to form an adaptive weighting coefficient.

[0066] The prior states are corrected in descending order of confidence, while the overall uncertainty is reduced, thus achieving real-time multi-source fusion.

[0067] It should be noted that this step effectively eliminates the blind spots and noise interference of single sensors by collecting data on road geometry, traffic conditions, weather conditions and road surface conditions in a synchronous fusion manner, which significantly improves the stability and accuracy of environmental perception, and thus provides reliable data support for real-time vehicle decision-making.

[0068] Step 2: Input the environmental parameter vector into a two-layer evaluation network consisting of a fuzzy logic layer and a meta-learning layer. First, fuzzify parameters of the same category. Then, use a meta-policy optimization algorithm with safety margin as feedback signal to update the weights of each parameter online. Finally, output four levels of environmental complexity labels, representing simple, general, complex, and high-risk driving scenarios, respectively. Note that the following should be noted in this step:

[0069] Road geometry and traffic flow observations are used as the first set of inputs, and meteorological and pavement adhesion observations are used as the second set of inputs, and fuzzification is performed on each of them.

[0070] Using the safety margin as a real-time feedback signal, the weights of the two sets of inputs are adjusted through a meta-policy optimization algorithm;

[0071] The fuzzy output is then weighted and compared with a preset complexity classification rule to generate an environment complexity label L.

[0072] Furthermore, preset complexity grading rules include:

[0073] When the weighted output comprehensive score is lower than the first threshold, it is marked as a simple driving scenario (label is 0), and the vehicle is allowed to drive at a speed close to the road speed limit.

[0074] When the weighted output comprehensive score is between the first threshold and the second threshold, it is marked as a general driving scenario (label 1), and the vehicle speed is limited to within 80% of the road speed limit.

[0075] When the weighted output comprehensive score is between the second and third thresholds, it is marked as a complex driving scenario (label 2), and the vehicle speed is limited to 60% of the road speed limit.

[0076] When the weighted output comprehensive score is higher than the third threshold, it is marked as a high-risk driving scenario (label 3), and the braking preload and distance keeping tasks are forcibly activated.

[0077] Furthermore, the two-layer evaluation network specifically includes:

[0078] The fuzzy logic layer is used to map each environmental parameter to a fuzzy set according to a preset membership function;

[0079] The meta-learning layer is used to dynamically adjust the weights of the fuzzy set based on the safety margin feedback and through a policy optimization mechanism.

[0080] The classification decision unit is used to weight and summarize the corrected fuzzy output to obtain a comprehensive score, and output the environmental complexity label according to the classification rules.

[0081] It should be noted that the first threshold, second threshold, and third threshold in this embodiment are set based on medium tire grip, normal braking ability, and standard road speed limit constraints. For example, the first threshold is defined as 0.30, the second threshold is defined as 0.60, and the third threshold is defined as 0.85, in order to improve vehicle control stability and scene resolution.

[0082] As an example, in the fuzzy logic layer, a three-valued membership function (low, medium, high) is constructed for each observation, and the initial membership interval is determined using expert experience;

[0083] Calculate the membership degree value of each observation at the current time to obtain the membership degree vector μ = [μ1, μ2, ..., μ]. m ];

[0084] Where, μ i Let be the fuzzy membership value of the i-th environmental feature;

[0085] Vehicle obstacle safety margin Δ s As a feedback signal for the meta-learning layer;

[0086] A policy gradient-based meta-policy optimization algorithm is employed to optimize the feature weights w = [w1, w2, ..., w] of the two input sets. m Perform online updates, with the following update rules:

[0087]

[0088] Where η is the learning rate, Δ s To ensure a safety margin, To ensure that the weight of high-risk features is automatically increased when the safety margin is insufficient, the loss function is designed with safety risk and ride comfort as joint objectives. i This represents the weight coefficient of the i-th environmental feature in the current loop;

[0089] In the classification decision unit, the comprehensive score S is calculated based on the updated weight vector and membership vector:

[0090]

[0091] Where S∈[0,1] is the scene comprehensive complexity score, and m is the total feature dimension.

[0092] Preferably, this step constructs a two-layer evaluation network to achieve fuzzy processing of environmental parameter vectors and online updating of weights, enabling the environmental complexity assessment to dynamically adapt to constantly changing actual driving conditions. The meta-policy optimization algorithm in this embodiment uses safety margin as a feedback signal to achieve real-time and accurate output of environmental complexity labels, effectively avoiding the risk of misjudgment caused by the lack of adaptability in traditional static evaluation methods, improving the accuracy of complexity level classification, and thus effectively enhancing the safety of vehicles in complex environments.

[0093] Step 3: Input the environmental complexity label into the control model constructed based on the hierarchical adaptive complexity vehicle speed control algorithm to generate the target vehicle speed. Note that the following should be noted in this step:

[0094] The speed limit value for this road segment is obtained by retrieving the electronic road speed limit database, and the vehicle dynamics limit estimator is called in real time to obtain the limit safe speed under the current adhesion coefficient constraint. The smaller value between the two is defined as the benchmark safe speed.

[0095] Read the environment complexity label L∈{0,1,2,3} output from step two, which corresponds to four levels of scenarios: simple, normal, complex, and high-risk, respectively, with (1+L) as the unit of measurement. -1 The initial target speed is obtained by multiplying the first-order scaling factor by the baseline safe speed.

[0096] The initial target vehicle speed is input into a first-order exponential smoothing function, and the friction curvature integral factor and the visible distance error function are simultaneously superimposed to construct a hierarchical adaptive complexity vehicle speed control model.

[0097] When L=3 and the safety margin is lower than the threshold, a preload request is sent to the brake actuator to reduce the target vehicle speed by an additional 10%, and the final target vehicle speed is output.

[0098] Furthermore, the target vehicle speed is generated, specifically including:

[0099] When the environmental complexity label is simple, the smaller value between the maximum permissible vehicle speed and the vehicle dynamic limit speed is selected as the target vehicle speed.

[0100] When the environmental complexity label is "general scenario", multiply the smaller value by 80% to obtain the target vehicle speed.

[0101] When the environment complexity label is "complex scene", multiply the smaller value by 60% to get the target vehicle speed.

[0102] When the environmental complexity is labeled as a high-risk scenario, the smaller value is multiplied by 40% to obtain the target vehicle speed, and the braking preload task is activated in advance.

[0103] As an example, the control model constructed in this embodiment is used to calculate the target vehicle speed, and the mathematical expression of the model is as follows:

[0104]

[0105] Among them, V tar V is the target vehicle speed generated in the current cycle. lim Let L be the smaller of the road speed limit and the vehicle dynamics limit speed, H be the forward predicted sight distance, κ(h) be the curvature of the predicted curve at a distance h from the vehicle, χ be the curvature attenuation coefficient, sgn(·) be the curvature sign function, erf be the Gaussian error function, σ be the sight distance sensitivity coefficient, and d be the smaller of the road speed limit and the vehicle dynamics limit speed. vis R represents the real-time visibility distance, and R represents the number of dimensions of the lateral wind force and road surface disturbance characteristics. ω is the normalized weight for the r-th perturbation feature. r For the corresponding disturbance amplitude, δ is the disturbance exponential amplification factor, β is the complexity gain reduction factor, φ is the adhesion exponential scaling factor, μ is the real-time tire-road adhesion coefficient, and λ is the adhesion nonlinear attenuation coefficient.

[0106] The overall value of the control model satisfies 0 <V tar ≤V lim When the integral-fractional term approaches 0 or the exponential suppression term approaches 0, the formula output approaches 0, triggering the vehicle's minimum protection speed. When the numerator is much larger than the denominator and the road adhesion coefficient is high, the exponential suppression effect weakens, and the formula output approaches the upper bound V. lim / (1+L).

[0107] It should be noted that this step constructs a control model based on a hierarchical adaptive complexity vehicle speed control algorithm, outputting target vehicle speeds that are closely related to the actual environmental conditions for different environmental complexity labels, which greatly improves the differentiation and accuracy of the vehicle speed control strategy.

[0108] Preferably, the method provided in this embodiment can automatically adjust the vehicle speed control strategy according to the level of environmental complexity, thereby avoiding the safety hazards that may be caused by a single speed strategy, effectively balancing vehicle driving safety and traffic efficiency, and improving the comfort and economy of driving and riding.

[0109] Step 4: If the target vehicle speed is greater than or equal to the collision risk index, immediately increase the environmental complexity level and return to Step 3 to recalculate the target vehicle speed until the target vehicle speed is less than the collision risk index, then output the vehicle speed execution command. Note that the following should be noted in this step:

[0110] In each cycle, calculate the remaining braking distance, the difference between the estimated collision time and the system braking response time;

[0111] The target vehicle speed is compared with the remaining braking distance, the difference between the estimated collision time and the difference between the system braking response time. If any of the dangerous quantities reaches the danger threshold, the collision risk is triggered.

[0112] If a collision risk is triggered and the current environmental complexity level has not yet reached a high risk level, immediately raise the environmental complexity level by one level and return to step three to recalculate the target vehicle speed.

[0113] If the target vehicle speed cannot be lowered below the collision risk index after two consecutive increases, a downgraded safety mode will be activated, instructing the vehicle to slow down to half the road speed limit and periodically checking the collision risk index until it is cleared.

[0114] In an optional implementation, the collision risk indicators include at least the remaining braking distance, the expected collision time difference, the closing speed of the vehicle and the obstacle ahead, the road adhesion level correction factor, and the difference in system braking response time.

[0115] Among them, the remaining braking distance, the expected collision time difference, the closing speed of the vehicle and the obstacle in front, the road adhesion level correction coefficient, and the difference of the system braking response time are normalized according to preset weights, weighted and summed, and then compared with the danger threshold to determine the risk level.

[0116] It should be further noted that the vehicle speed execution command includes at least the drive torque request, braking torque request, regenerative braking ratio, gear or torque distribution request, and vehicle-road cooperative risk notification.

[0117] Among them, the vehicle speed execution command is sent to each execution mechanism within 10ms after being generated by the real-time dispatch system, and the execution confirmation is received within 5ms after being sent. If no confirmation is received within the time limit, the system switches to the degraded safety mode.

[0118] In degraded safety mode, the driving torque is limited to less than 30% of the road speed limit, the braking torque is kept continuously output, and the vehicle safety status and environmental complexity level information are sent to the remote monitoring platform once per second until the risk is eliminated.

[0119] As an example, the formula for calculating the remaining braking distance is:

[0120]

[0121] The formula for calculating the expected collision time difference is:

[0122]

[0123] The formula for calculating the difference in system braking response time is:

[0124]

[0125] The obstacle closing speed of the vehicle and the correction factor for the road adhesion level are respectively:

[0126]

[0127] The comprehensive risk indicators are:

[0128]

[0129] The rules for determining risk levels are as follows:

[0130]

[0131] Among them, D b τ is the remaining braking distance, μ is the real-time tire-road adhesion coefficient, g is the gravitational acceleration constant, and τ is the ground contact distance. r Let v(τ) be the total delay of the driving control system, v(τ) be the instantaneous vehicle speed function during the delay period, V0 be the vehicle speed at the end of the delay period, and α be the total delay of the driving control system. max ΔT represents the maximum achievable braking deceleration under the current adhesion conditions. c To predict the collision time difference, d ego-obj V represents the longitudinal distance between the vehicle and the target ahead. cl Let denoted as φ, sigm(·), be the sigm-normalized function, γ be the time difference exponential decay coefficient, Φ be the road adhesion level correction coefficient, and Δt be the closing velocity of the vehicle relative to the target ahead. sys ω is the difference in braking response time of the system. i Let ξ be the normalized weight of the i-th risk factor.i To correspond to the normalized risk, R c For the comprehensive collision risk index, ρ1 and ρ2 are the two-stage dividing values ​​of risk level, RiskLevel is the final risk level label, and t 90 Braking torque M after the self-braking command is issued b (t) The desired braking torque M is reached for the first time. req 90% of the time, t cmd t is the time when the braking command is issued in this cycle. ref V is the baseline braking response time recorded by the vehicle platform during factory calibration. cl v is the closing velocity of the vehicle relative to the obstacle ahead; a positive value indicates approaching and a negative value indicates moving away. ego v is the current longitudinal speed of the vehicle. obj Let μ be the longitudinal vehicle speed projected onto the obstacle along the road coordinate system, ψ be the cosine correction factor for the angle between the vehicle's longitudinal axis and the line connecting the vehicle and the obstacle, and μ be the longitudinal speed of the obstacle projected onto the road coordinate system. dry The reference adhesion coefficient for dry asphalt pavement is μ. min This is the minimum adhesion coefficient threshold required for the platform's safety margin.

[0132] It should be noted that ρ1 = 0.35, ρ2 = 0.70, and the danger threshold is 1. When R c ≥1 and the target vehicle speed is still not lower than R c The downgraded security mode is triggered at this time.

[0133] Preferably, this step sets a collision risk index as a safety threshold. When the target vehicle speed is greater than or equal to this index, the environmental complexity level is actively increased and the target vehicle speed is replanned, forming a closed-loop dynamic speed control mechanism. This method enables the vehicle to avoid potential collisions in a timely manner, effectively making up for the safety deficiencies caused by the lack of a collision prediction feedback mechanism in traditional static speed control strategies. It enhances the system's risk prediction and active safety intervention capabilities, thereby significantly improving vehicle driving safety.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling the speed of an autonomous vehicle based on environmental complexity, characterized in that, The control method executes the following four steps sequentially with a cycle period of less than 50ms, including: Step 1: Simultaneously collect road geometry, traffic flow density, visibility distance, road surface adhesion, and meteorological data through lidar, millimeter-wave radar, camera, vehicle-road cooperative communication device, and inertial measurement unit. After alignment in the time domain, use Kalman filter algorithm to fuse them to obtain environmental parameter vectors. The process of fusing environmental parameter vectors using the Kalman filter algorithm includes: aligning the raw data from each sensor frame by frame based on the vehicle's global positioning information and time source; calculating the current prior environmental state based on the environmental state output from the previous cycle and the vehicle's motion state; sequentially introducing four types of observations—road geometry, traffic flow, meteorology, and road surface adhesion—and using Kalman gain to weight and correct the current prior environmental state; performing outlier removal and coordinate system unification on the corrected state vector, and outputting an environmental parameter vector containing road curvature, number of available lanes, lane width, average traffic speed, traffic density, nearest preceding vehicle distance, visibility distance, road surface friction level, road surface humidity, precipitation intensity, light intensity, lateral wind speed, and ambient temperature. Step 2: Input the environmental parameter vector into a two-layer evaluation network consisting of a fuzzy logic layer and a meta-learning layer. First, fuzzify the parameters of the same category. Then, use a meta-policy optimization algorithm with safety margin as the feedback signal to update the weights of each parameter online. Finally, output four levels of environmental complexity labels, representing four driving scenarios: simple, general, complex, and high-risk. The environmental parameter vector is input into a two-layer evaluation network to output an environmental complexity label. This includes: taking road geometry and traffic flow observations as the first set of inputs and meteorological and road surface adhesion observations as the second set of inputs, and performing fuzzification processing on each; using safety margin as a real-time feedback signal, adjusting the weights of the two sets of inputs through a meta-policy optimization algorithm; combining the fuzzified outputs, weighting them, and comparing them with a preset complexity classification rule to generate an environmental complexity label. Step 3: Input the environmental complexity label into the control model constructed based on the hierarchical adaptive complexity vehicle speed control algorithm to generate the target vehicle speed; Step three specifically includes: selecting the smaller value between the road speed limit and the vehicle dynamic limit speed as the baseline safe speed; multiplying the baseline safe speed by the corresponding scaling ratio according to the environmental complexity label, where 100% is used for simple driving scenarios, 80% for general driving scenarios, 60% for complex driving scenarios, and 40% for high-risk driving scenarios; inputting the scaled speed into a smoothing function to eliminate speed jumps between adjacent cycles; adding a braking preload command in high-risk driving scenarios, and outputting the final target speed. Step 4: If the target vehicle speed is greater than or equal to the collision risk index, the environmental complexity level is immediately increased and the process returns to Step 3 to recalculate the target vehicle speed until the target vehicle speed is less than the collision risk index. Then, a vehicle speed execution command is output.

2. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 1, characterized in that, The preset complexity grading rules include: When the weighted output comprehensive score is lower than the first threshold, it is marked as a simple driving scenario, allowing the vehicle to drive at the speed limit of the standard road. When the weighted output comprehensive score is between the first threshold and the second threshold, it is marked as a general driving scenario, and the vehicle speed is limited to 80% of the road speed limit. When the weighted output comprehensive score is between the second and third thresholds, it is marked as a complex driving scenario, and the vehicle speed is limited to 60% of the road speed limit. When the weighted output comprehensive score is higher than the third threshold, it is marked as a high-risk driving scenario, and the braking preload and distance keeping tasks are forcibly activated.

3. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 1, characterized in that, The two-layer evaluation network specifically includes: The fuzzy logic layer is used to map each environmental parameter to a fuzzy set according to a preset membership function; The meta-learning layer is used to dynamically adjust the weights of the fuzzy set based on the safety margin feedback and through a policy optimization mechanism. The classification decision unit is used to weight and summarize the corrected fuzzy output to obtain a comprehensive score, and output the environmental complexity label according to the classification rules.

4. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 1, characterized in that, The generation of the target vehicle speed specifically includes: When the environmental complexity label is simple, the smaller value between the maximum permissible vehicle speed and the vehicle dynamic limit speed is selected as the target vehicle speed. When the environmental complexity label is "general scenario", the smaller value is multiplied by 80% to obtain the target vehicle speed. When the environmental complexity label is "complex scene", the smaller value is multiplied by 60% to obtain the target vehicle speed. When the environmental complexity is labeled as a high-risk scenario, the smaller value is multiplied by 40% to obtain the target vehicle speed, and the braking preload task is activated in advance.

5. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 4, characterized in that, Step four specifically includes: In each cycle, calculate the remaining braking distance, the difference between the estimated collision time and the system braking response time; The target vehicle speed is compared with the remaining braking distance, the difference between the estimated collision time and the difference between the system braking response time. If any of the danger quantities reaches the danger threshold, a collision risk is triggered. If a collision risk is triggered and the current environmental complexity level has not yet reached a high risk level, immediately raise the environmental complexity level by one level and return to step three to recalculate the target vehicle speed. If the target vehicle speed cannot be lowered below the collision risk index after two consecutive increases, a downgraded safety mode will be activated, instructing the vehicle to decelerate to half the road speed limit and periodically checking the collision risk index until it is cleared.

6. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 5, characterized in that, The collision risk indicators include at least the remaining braking distance, the expected collision time difference, the closing speed of the vehicle and the obstacle in front, the road adhesion level correction factor, and the difference in system braking response time. The remaining braking distance, the expected collision time difference, the closing speed of the vehicle and the obstacle ahead, the road adhesion level correction coefficient, and the difference of the system braking response time are normalized according to preset weights, weighted and summed, and then compared with the danger threshold to determine the risk level.

7. The method for controlling the speed of an autonomous vehicle based on environmental complexity according to claim 6, characterized in that, The vehicle speed execution command includes at least the drive torque request, braking torque request, regenerative braking ratio, gear or torque distribution request, and vehicle-road cooperative risk notification. The vehicle speed execution command is sent to each actuator within 10ms after being generated by the real-time scheduling system, and execution confirmation is received within 5ms after being sent. If no confirmation is received within the time limit, the system switches to the degraded safety mode. In the degraded safety mode, the driving torque is limited to less than 30% of the road speed limit, the braking torque is kept continuously output, and the vehicle safety status and environmental complexity level information are sent to the remote monitoring platform once per second until the risk is eliminated.

Images