Artificial intelligence-based autonomous driving evaluation system

Abstract

The application discloses an automatic driving evaluation system based on artificial intelligence and relates to the technical field of automatic driving.The system comprises the following components: a meteorological model coupling module, a sensor parameter adjustment module, an automatic driving simulation module and an evaluation analysis module;the meteorological model coupling module is connected with an external meteorological model to acquire comprehensive meteorological data including air temperature, humidity, wind speed, rainfall, snowfall, visibility and the like in real time, and an extreme weather scenario is constructed based on the data;the sensor parameter adjustment module automatically adjusts the performance parameters of vehicle sensors according to different weather types to ensure the accuracy of sensor data under extreme weather conditions;and the automatic driving simulation module runs an automatic driving system model in the simulated extreme weather scenario to simulate the driving process of a vehicle.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to an autonomous driving evaluation system based on artificial intelligence. Background Technology

[0002] With the rapid development of autonomous driving technology, its safety and reliability have become the focus of public and industry attention. Autonomous driving systems need to operate stably under various complex and extreme weather conditions to ensure the safety of passengers and pedestrians. However, extreme weather conditions such as heavy rain, dense fog, and heavy snow can interfere with vehicle sensors, affecting their ability to perceive the environment, and thus affecting the decision-making and control accuracy of the autonomous driving system.

[0003] Traditional evaluation methods for autonomous driving systems often rely on laboratory simulations or limited field tests. These methods are significantly inadequate when simulating extreme weather conditions. Laboratory simulations cannot fully reproduce the complexity and dynamism of the real world, while field tests are affected by a variety of uncontrollable factors such as weather and traffic conditions, making it difficult to comprehensively and accurately evaluate the performance of autonomous driving systems under extreme weather conditions. In addition, traditional evaluation methods often focus on single-dimensional performance indicators, such as driving safety or decision rationality, while neglecting other important factors such as the accuracy of sensor data, resulting in evaluation results that are not comprehensive and objective enough.

[0004] To address the shortcomings of traditional autonomous driving system evaluation methods under extreme weather conditions, this invention proposes an artificial intelligence-based autonomous driving evaluation system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an artificial intelligence-based autonomous driving evaluation system. This system can establish a data connection with an external meteorological model through a meteorological model coupling module, acquire comprehensive meteorological data in real time, and construct simulated extreme weather scenarios based on this data. Simultaneously, a sensor parameter adjustment module automatically adjusts the performance parameters of vehicle sensors according to different weather types to ensure the accuracy of sensor data under extreme weather conditions. An autonomous driving simulation module runs an autonomous driving system model in the simulated extreme weather scenarios, simulating the vehicle's driving process. Finally, an evaluation and analysis module uses artificial intelligence algorithms to analyze the operational data, quantitatively evaluating the performance of the autonomous driving system under simulated extreme weather scenarios. This provides a comprehensive and dynamic evaluation method, significantly improving the accuracy of performance evaluation of autonomous driving systems under extreme weather conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an artificial intelligence-based autonomous driving evaluation system, which includes the following components: a meteorological model coupling module, a sensor parameter adjustment module, an autonomous driving simulation module, and an evaluation and analysis module;

[0007] The meteorological model coupling module is used to establish a data connection with an external meteorological model, acquire meteorological data in real time, including but not limited to temperature, humidity, wind speed, rainfall, snowfall, and visibility information, and construct simulated extreme weather scenarios based on this meteorological data.

[0008] The sensor parameter adjustment module: When the meteorological model coupling module generates extreme weather scenarios, this module automatically adjusts the performance parameters of the vehicle sensors according to different weather types.

[0009] The autonomous driving simulation module: Based on the adjusted sensor performance parameters, it runs an autonomous driving system model in simulated extreme weather scenarios, simulating the vehicle's driving process in this environment, including the vehicle's acceleration, deceleration, steering, and avoidance operations, as well as the autonomous driving system's perception, decision-making, and control processes of environmental information.

[0010] The evaluation and analysis module uses artificial intelligence algorithms to analyze the operational data of the autonomous driving simulation module. Through the set evaluation indicators, such as vehicle driving safety, decision rationality, and sensor data accuracy, it quantitatively evaluates the performance of the autonomous driving system under simulated extreme weather scenarios, generates a detailed evaluation report, and points out the advantages and disadvantages of the autonomous driving system in responding to climate change.

[0011] Furthermore, when constructing simulated extreme weather scenarios, the meteorological model coupling module employs a dynamic climate parameter generation algorithm. This algorithm is based on the fusion analysis of historical meteorological data and real-time meteorological data. It generates the simulated extreme weather scenario parameter S using the formula S = α × H + (1 - α) × R, where H is the parameter set corresponding to the extreme weather type in the historical meteorological data, R is the parameter set in the real-time meteorological data, and α is the historical data weight coefficient, with a value range of [0.3, 0.7]. The method for determining α is as follows: the goodness of fit between historical data and real-time data of extreme weather events in the target area over the past ten years is statistically analyzed, and the α value that maximizes the goodness of fit is found through an optimization algorithm. This makes the generated simulated extreme weather scenario more closely resemble the actual situation, improving the authenticity and reliability of subsequent evaluations.

[0012] Furthermore, when adjusting the camera's imaging quality parameters, the sensor parameter adjustment module employs a dynamic adjustment algorithm based on visual blur. For rainstorm scenes, the camera image sharpness adjustment formula is as follows:

[0013]

[0014] Where C old To improve the image clarity of the camera before adjustment, C newTo adjust the sharpness, β is the rainfall impact coefficient, with a value range of [0.1, 0.5]. This coefficient is determined by collecting actual image quality data from the camera under different rainfall intensities and using regression analysis. V is the real-time rainfall amount. max By setting a maximum rainfall threshold, the algorithm can dynamically adjust the camera's resolution based on real-time rainfall, more accurately simulating the interference of heavy rain on camera imaging, and providing more realistic sensor data input for autonomous driving simulation.

[0015] Furthermore, the algorithm used in the evaluation and analysis module to assess vehicle driving safety employs a multi-dimensional risk assessment model, the calculation formula of which is:

[0016] R s =ω1×D+ω2×A+ω3×L

[0017] Where R s Let D be the vehicle driving safety risk value, A be the vehicle acceleration, L be the lane departure degree, and ω1, ω2, and ω3 be the weights of the corresponding parameters, with values ​​ranging from [0.4, 0.6], [0.2, 0.4], and [0.2, 0.4], respectively. The weights are determined by collecting a large amount of actual traffic accident case data, analyzing the influence of each factor in the occurrence of the accident, and using the analytic hierarchy process (AHP) to determine the weights of each factor. This enables the model to comprehensively and accurately assess the driving safety risks of vehicles in simulated extreme weather scenarios, providing a scientific basis for the safety evaluation of autonomous driving systems.

[0018] Furthermore, the evaluation and analysis module employs a decision evaluation algorithm based on traffic rule semantic understanding when assessing the rationality of the autonomous driving system's decisions. First, it semantically parses the traffic rules and constructs a rule knowledge base. For each decision of the autonomous driving system, it uses a formula... Calculate the probability P that the decision complies with traffic rules. r Where N is the total number of traffic rules involved in the decision, and M i Let M be the matching degree of the i-th rule, where M is the matching degree when the decision perfectly matches the rule. i =1, when partially conforming, a value between [0, 1) is assigned based on the degree of conformity; when completely non-conforming, M = 1. i =0, by combining expert experience with actual traffic data, the judgment criteria for the matching degree of each rule are determined, thereby realizing a quantitative assessment of the rationality of the autonomous driving system's decision-making and effectively evaluating whether the system's decision-making is compliant under extreme weather conditions.

[0019] Furthermore, the evaluation and analysis module uses interactive experience comfort evaluation indicators when assessing passenger comfort in the autonomous driving system:

[0020] Decision transparency: The autonomous driving system shall inform passengers of the decision logic in real time through the dashboard or voice, with a coverage rate of ≥90%;

[0021] Route planning rationality: Avoid frequent lane changes or choosing bumpy road sections, generate a comfortable route model by training with historical navigation data, and calculate the deviation between the actual route and the comfortable route;

[0022] Emergency response smoothness: When avoiding obstacles or braking, the operating force is adjusted in advance through the pre-aiming algorithm to avoid sudden braking or sharp steering wheel turns. This can be quantitatively evaluated through passenger body displacement sensor data.

[0023] Furthermore, the autonomous driving simulation module employs a reinforcement learning-based dynamic environment adaptation algorithm during simulated vehicle driving. It defines the state space as the vehicle's current position, speed, sensor data, and surrounding environmental information, and the action space as the vehicle's acceleration, deceleration, and steering operations, using formulas... The action value function Q(s,a) is updated, where η is the learning rate, ranging from [0.1, 0.3], and γ is the discount factor, ranging from [0.7, 0.9]. Through trials in a large number of simulated scenarios, with the goal of enabling the system to reach stable and reasonable decisions as quickly as possible, these two parameters are optimized and determined. r is the immediate reward obtained for the current action. A reward mechanism is set according to whether a collision occurs or whether the vehicle deviates from its lane. This algorithm enables the autonomous driving simulation module to continuously learn and adapt to dynamically changing environments in simulated extreme weather scenarios, improving the realism and reliability of the simulation and providing richer and more effective data for the evaluation and analysis module.

[0024] Furthermore, when adjusting the performance parameters of the lidar, the sensor parameter adjustment module employs a penetration rate adjustment algorithm based on particle scattering theory. In dense fog scenarios, the lidar penetration rate P... t The calculation formula is P t =e -σ×N×d Where σ is the scattering cross section of the fog droplets on the laser, and this parameter is obtained by laser scattering experiments on fog droplets of different concentrations and by data fitting. N is the number of fog droplets per unit volume, which can be determined according to the dense fog scene parameters generated by the meteorological model coupling module. d is the laser propagation distance. This algorithm is based on physical principles and can accurately simulate the attenuation effect of dense fog on the lidar signal, making the sensor parameter adjustment more in line with the actual physical laws and improving the accuracy of lidar performance evaluation by the autonomous driving evaluation system.

[0025] Furthermore, when generating the evaluation report, the evaluation and analysis module employs a report generation algorithm based on natural language generation. Specifically, this includes: first, semantically converting the quantitative evaluation results to construct a mapping relationship between different evaluation dimensions and semantic descriptions; then, determining a weight allocation mechanism based on the importance of each evaluation dimension. This mechanism is determined by industry experts scoring the impact of each evaluation dimension on the overall performance of the autonomous driving system and statistically analyzing the scoring results using the Delphi method; finally, weighted fusion processing of the results of each evaluation dimension is performed according to the determined weights to generate a topic framework for the evaluation text; and finally, using a pre-trained language model, based on the generated topic framework and combined with specific simulated data cases, a complete evaluation report text containing targeted improvement suggestions is generated.

[0026] Furthermore, after acquiring meteorological data, the meteorological model coupling module performs spatiotemporal interpolation processing on the data, employing an optimization algorithm based on Kriging interpolation. Specifically, this includes: constructing a spatial interpolation model based on known monitoring point data for locations within the target area where no meteorological monitoring points are set up; determining the influence weight of each known monitoring point on the estimated meteorological parameters at the target location by analyzing the spatial autocorrelation of meteorological parameters; minimizing the estimation variance by a target function using the Lagrange multiplier method; weighting and combining the meteorological parameter values ​​of each known monitoring point using the determined weights to obtain the estimated meteorological parameters at the target location; cross-validating the interpolation results; and optimizing and adjusting the interpolation model by comparing the estimation errors of different interpolation methods at the validation points. This improves the spatial integrity and accuracy of the meteorological data, making the constructed simulated extreme weather scenario more uniform and realistic within the region, thus providing a higher quality data foundation for subsequent autonomous driving evaluation.

[0027] Compared with existing technologies, this AI-based autonomous driving evaluation system has the following advantages:

[0028] First, the system establishes a data connection with an external meteorological model through a meteorological model coupling module, acquiring comprehensive meteorological data in real time, including temperature, humidity, wind speed, rainfall, snowfall, and visibility. Based on this data, it constructs simulated extreme weather scenarios. The sensor parameter adjustment module automatically adjusts the performance parameters of the vehicle's sensors according to different weather types to ensure the accuracy of sensor data under extreme weather conditions. The autonomous driving simulation module runs the autonomous driving system model in the simulated extreme weather scenarios, simulating the vehicle's driving process. Finally, the evaluation and analysis module uses artificial intelligence algorithms to analyze the operational data and quantitatively evaluate the performance of the autonomous driving system under simulated extreme weather scenarios. This comprehensive and dynamic evaluation method significantly improves the accuracy of performance evaluation of the autonomous driving system under extreme weather conditions, providing strong support for the further optimization of autonomous driving technology.

[0029] Second, when generating evaluation reports, the system employs a natural language-based report generation algorithm in its evaluation and analysis module. This algorithm performs semantic transformation on the quantitative evaluation results, constructs a mapping relationship between different evaluation dimensions and semantic descriptions, and determines a weight allocation mechanism based on the importance of each evaluation dimension. The results of each evaluation dimension are then weighted and fused to generate a thematic framework for the evaluation text. Finally, using a pre-trained language model, based on the generated thematic framework and combined with specific simulated data examples, a complete evaluation report text containing targeted improvement suggestions is generated. This detailed evaluation report not only points out the advantages and disadvantages of autonomous driving systems in addressing climate change but also provides specific directions and suggestions for improvement, offering important reference for the continuous optimization and upgrading of autonomous driving systems.

[0030] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0032] Figure 1 A flowchart illustrating the functional implementation of an AI-based autonomous driving evaluation system;

[0033] Figure 2 This is a schematic diagram of the structure of an AI-based autonomous driving evaluation system. Detailed Implementation

[0034] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0035] Example 1

[0036] An autonomous driving technology company needs to test the vehicle's obstacle avoidance capabilities in heavy rain to verify the reliability of sensors under precipitation interference and the rationality of autonomous driving decisions.

[0037] Real-time meteorological data is accessed: when rainfall in a certain area reaches 50 mm / h (exceeding the rainstorm threshold), and combined with historical rainstorm data (such as the set of extreme rainfall parameters H for the same season in previous years), a simulated rainstorm scenario is constructed through a dynamic climate parameter generation algorithm.

[0038] S=α×H+(1-α)×R

[0039] The historical data weighting coefficient α = 0.4, and the real-time data R includes parameters such as rainfall and visibility.

[0040] After identifying the rainstorm scene, the camera's image quality dynamic adjustment algorithm is triggered: Assuming the original sharpness is C old =80%, rainfall impact coefficient β = 0.3, real-time rainfall V = 50 mm / h, maximum threshold V max =60mm / h, then the sharpness after adjustment

[0041] In a simulated heavy rain scenario, a stationary obstacle suddenly appears ahead. The system adjusts its steering action using a reinforcement learning algorithm. The state space includes the vehicle speed (20 m / s) and the distance to the obstacle detected by the sensors (30 meters). The action space selects to steer and avoid the obstacle. The action value function is updated using the Q-learning formula. If the obstacle avoidance is successful and the lane is not deviated from, the immediate reward is r = +10, the learning rate is η = 0.1, and the discount factor is γ = 0.9.

[0042] Calculate the safety risk value using a multi-dimensional risk assessment model: R s =ω1×D+ω2×A+ω3×L Let the weights be ω1=0.5 (distance), ω2=0.3 (acceleration), ω3=0.2 (lane deviation), and the measured distance D=10 meters and acceleration A=-5 m / s². 2 (Deceleration), deviation degree L = 0.2 meters, calculate R s=0.5×10+0.3×(-5)+0.2×0.2=3.54 (the lower the value, the safer).

[0043] Based on a traffic rule semantic understanding algorithm, determine whether a steering and avoidance decision conforms to the rule "reducing speed and maintaining a safe distance in heavy rain." Assuming two rules are involved, with matching degrees of 0.8 and 0.9 respectively, the probability of conformity is:

[0044] Example 2

[0045] An automaker conducted autonomous driving tests in a typical foggy area during winter. The goal was to verify the vehicle's ability to detect distant obstacles, the reliability of sensor data, and the system's emergency response mechanism in extremely dense fog environments with visibility of less than 50 meters. The test scenario was set as follows: 6 a.m. (poor lighting conditions), temperature 5°C, air humidity of 95%, and dense fog continuously covering the test section, simulating the challenges that real winter fog and haze weather poses to autonomous driving.

[0046] Accessing regional meteorological monitoring network data, identifying current environmental parameters, visibility is 40 meters, average droplet size is 10 micrometers, and droplet concentration per unit volume is relatively high.

[0047] By combining historical data of dense fog in the region over the past five years (such as fog duration and spatial distribution characteristics), a spatiotemporal interpolation algorithm (based on Kriging interpolation optimization) was used to supplement the meteorological parameters of the uncovered monitoring points in the test section, generating a high-precision dense fog distribution simulation scenario and clarifying the differences in fog droplet density at different locations.

[0048] A "dense fog weather" trigger signal is sent to the system to initiate the sensor adaptive adjustment process.

[0049] Upon receiving a signal from a dense fog scene, the system automatically identifies the sensor type that needs adjustment: LiDAR (which is most affected by fog droplet scattering).

[0050] Dynamically increase laser emission power to the safe limit to enhance signal strength and counteract droplet attenuation;

[0051] Adjust the LiDAR scanning mode from "wide-area scanning" to "focused scanning" to prioritize enhancing detection accuracy within 200 meters ahead. Enable the built-in anti-interference filtering algorithm to filter out noise data caused by fog droplet scattering, improve point cloud purity, and simultaneously send sensor status update information (such as the adjusted detection distance threshold and data refresh rate) to the autonomous driving simulation module.

[0052] After loading the dense fog scenario, the simulated vehicle travels at a speed of 40 km / h (lower than the conventional speed limit, meeting the safety requirements for low visibility) in a two-way two-lane road.

[0053] The system constructs an environmental model in real time using lidar point cloud data. When a blurry point cloud (suspected to be the rear of a truck) is detected 150 meters ahead, a multi-sensor fusion verification mechanism is triggered. Millimeter-wave radar scans the area simultaneously and confirms the presence of metal reflection signals. The camera activates a super-resolution algorithm to attempt to capture the object's outline through gaps in the fog (limited by visibility, only a blurry image is obtained). Based on the fused data, the autonomous driving system determines it to be a "stationary obstacle" and triggers an emergency braking decision. The motor controller outputs maximum braking force, the anti-lock braking system (ABS) is activated simultaneously, and the vehicle attitude control system adjusts the suspension damping to suppress vehicle pitch during emergency braking and maintain driving stability.

[0054] Comparing the obstacle distance (145 meters) output by the lidar with the actual location (150 meters), the calculation error rate was 3.3%, which is considered acceptable in dense fog. Analysis of the point cloud data integrity revealed missing points at the top and edges of the obstacle due to reduced penetration, requiring further optimization through algorithm interpolation. The vehicle took 4.2 seconds to come to a complete stop from 40 km / h, with a braking distance of 38 meters, which is less than the theoretical safe distance (45 meters, calculated based on the road friction coefficient). Therefore, the braking performance was deemed satisfactory. The system was confirmed to complete the "perception-decision-execution" chain within 0.2 seconds of detecting an obstacle, meeting the requirement of "emergency response time < 0.5 seconds" in autonomous driving safety standards.

[0055] LiDAR has basic obstacle detection capabilities in extremely dense fog environments, but the accuracy of point clouds decreases significantly at long distances. It needs to be combined with millimeter-wave radar to improve reliability. The emergency braking strategy is reasonable, but it is recommended to optimize the multi-sensor fusion algorithm to reduce the risk of misjudgment in low visibility conditions.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automatic driving evaluation system based on artificial intelligence, characterized by, The system comprises the following components: a weather model coupling module, a sensor parameter adjustment module, an autonomous driving simulation module, and an evaluation and analysis module. The weather model coupling module is used for establishing a data connection with an external weather model, acquiring weather data in real time, including but not limited to temperature, humidity, wind speed, rainfall, snowfall, and visibility information, based on the weather data, constructing a simulated extreme weather scenario, and using a dynamic climate parameter generation algorithm based on fusion analysis of historical weather data and real-time weather data to generate a simulated extreme weather scenario parameter S through a formula wherein H is a parameter set corresponding to an extreme weather type in historical weather data, R is a parameter set of real-time weather data, is a historical data weight coefficient. The sensor parameter adjustment module automatically adjusts the performance parameters of the vehicle sensors according to different weather types when the weather model coupling module generates extreme weather scenarios. The autonomous driving simulation module runs the autonomous driving system model in the simulated extreme weather scenarios based on the adjusted sensor performance parameters, simulates the vehicle's driving process in that environment, including acceleration, deceleration, steering, and avoidance operations, as well as the autonomous driving system's perception, decision-making, and control processes. The evaluation and analysis module uses artificial intelligence algorithms to analyze the running data of the autonomous driving simulation module, evaluates the vehicle's driving safety, decision-making rationality, sensor data accuracy, and passenger comfort based on the set evaluation indicators, quantitatively evaluates the performance of the autonomous driving system in simulated extreme weather scenarios, generates a detailed evaluation report, and points out the strengths and weaknesses of the autonomous driving system in response to climate change. 2.The AI-based automatic driving evaluation system according to claim 1, characterized in that, When adjusting the camera imaging quality parameters, the sensor parameter adjustment module uses a dynamic adjustment algorithm based on visual blur, and for heavy rain scenarios, the camera image clarity adjustment formula is: where C old is the camera image definition before adjustment, C new is the definition after adjustment, β is the rainfall influence coefficient, V is the real-time rainfall, and V max is the set maximum rainfall threshold. 3.The AI-based automatic driving evaluation system according to claim 1, characterized in that, The algorithm used in the evaluation and analysis module to evaluate vehicle driving safety uses a multi-dimensional risk assessment model, and its calculation formula is: wherein R s is a vehicle driving safety risk value, D is a real-time distance from a preceding vehicle or an obstacle, A is a vehicle acceleration, L is a lane deviation degree, are weights of corresponding parameters, respectively. 4.The AI-based automatic driving evaluation system according to claim 1, characterized in that, The evaluation analysis module adopts a decision evaluation algorithm based on semantic understanding of traffic rules when evaluating the rationality of automatic driving system decisions. First, the traffic rules are semantically analyzed and a rule knowledge base is constructed. For each decision of the automatic driving system, the probability p that the decision conforms to the traffic rules is calculated by the formula p = 1 - Σ (1 - mi) / N r where N is the total number of traffic rules involved in the decision, mi is the matching degree of the i th rule. ​ 5.The AI-based automatic driving evaluation system according to claim 1, wherein When evaluating passenger comfort of the autonomous driving system, the evaluation and analysis module uses interactive experience comfort evaluation indicators: Decision transparency: the autonomous driving system informs passengers of the decision logic in real time through the instrument panel or voice, with a coverage rate of ≥90%; Path planning rationality: avoid frequent lane changes or choose bumpy sections, generate a comfortable path model through historical navigation data training, and calculate the deviation of the actual path from the comfortable path; Emergency response softness: when avoiding obstacles or braking, adjust the operation force in advance through the preview algorithm to avoid sudden braking or sharp steering, which can be quantitatively evaluated through passenger body displacement sensor data. 6.The AI-based automatic driving evaluation system according to claim 1, wherein The automatic driving simulation module adopts a dynamic environment adaptation algorithm based on reinforcement learning during simulation of vehicle driving, defines a state space as the current position, speed, sensor data and surrounding environment information of the vehicle, defines an action space as acceleration, deceleration and steering operation of the vehicle, and updates an action value function Q(s,a) through a formula Q(s,a)=Q(s,a)+a(r+γmaxa′Q(s′,a′)-Q(s,a)) wherein a is a learning rate, γ is a discount factor, r is an instant reward obtained by a current action, and a reward mechanism is set according to whether a collision occurs or a lane is deviated. 7.The AI-based automatic driving evaluation system according to claim 1, characterized in that, The sensor parameter adjustment module adopts a penetration rate adjustment algorithm based on particle scattering theory when adjusting the laser radar performance parameter, and in a dense fog scene, the penetration rate P of the laser radar is calculated as t The formula is Wherein The scattering cross section of the fog droplet to the laser, N is the number of fog droplets per unit volume, and d is the laser propagation distance. 8.The AI-based automatic driving evaluation system according to claim 1, characterized in that, When generating the evaluation report, the evaluation and analysis module uses a report generation algorithm based on natural language generation, which specifically includes: first, perform semantic conversion processing on the quantitative evaluation results, build a mapping relationship between different evaluation dimensions and semantic descriptions, then determine a weight distribution mechanism based on the importance of each evaluation dimension, perform weighted fusion processing on the results of each evaluation dimension according to the determined weights, generate a theme framework for the evaluation text, and finally use a pre-trained language model to generate a complete evaluation report text containing targeted improvement suggestions based on the generated theme framework and combined with specific simulation data cases. 9.The AI-based automatic driving evaluation system according to claim 1, wherein The weather model coupling module also performs space-time interpolation processing on the data after obtaining the weather data, adopts an optimization algorithm based on Kriging interpolation, and specifically includes: for positions in the target region where no weather monitoring points are set, a spatial interpolation model based on known monitoring point data is constructed, the influence weight of each known monitoring point on the estimated value of the weather parameter of the target position is determined by analyzing the spatial autocorrelation of the weather parameter, the weather parameter values of each known monitoring point are combined and calculated by using the determined weight for weighting, the estimated value of the weather parameter of the target position is obtained, and the interpolation result is cross-validated. By comparing the estimation errors of different interpolation methods on the validation points, the interpolation model is optimized and adjusted.

Images