To satisfy the preference of each driver, the development of a Lane-Keeping Assistance (LKA) system that can adapt to individual drivers has become a research hotspot in recent years. However, existing studies have mostly relied on the assumption that the LKA characteristic aligned with the driver's preference is consistent with this driver's naturalistic driving characteristic. Nevertheless, this assumption may not always hold true, causing limitations to the effectiveness of this method. This paper proposes a novel method for a Driver-Adaptive Lane-Keeping Assistance (DALKA) system based on drivers' real preferences. First, metrics are extracted from collected naturalistic driving data using action point theory to describe drivers' naturalistic driving characteristics. Then, the subjective and objective evaluation method is introduced to obtain the real preference of each test driver for the LKA system. Finally, machine learning methods are employed to train a model that relates naturalistic driving characteristics to the drivers' real preferences, and the model-predicted preferences are integrated into the DALKA system. The developed DALKA system is then subjectively evaluated by the drivers. The results show that our DALKA system, developed using this method, can enhance or maintain the subjective evaluations of the LKA system for most drivers.
Keywords: lane-keeping assistance system; driver adaption; subjective and objective evaluation; naturalistic driving characteristic; machine learning
Advanced Driver Assistance Systems (ADAS) are designed to enhance both driving safety and comfort. Lane-Keeping Assistance (LKA) is one type of ADAS that prevents the hazards resulting from unintended lane departure. However, during the design process of ADAS, insufficient consideration is given to the differences in preferences among drivers. Existing literature has shown that drivers of different genders, ages, and driving experiences have different levels of acceptance regarding ADAS [[
One approach to addressing this issue is to offer mode selection for drivers. For instance, in the case of Adaptive Cruise Control (ACC), various time headway modes—such as short, normal, and long—could be made available, allowing drivers to choose the mode that best suits their preferences via the human–machine interface. Although this approach can help customize ADAS for individual drivers, there are also some potential challenges to consider. For instance, when a driver lacks enough ADAS experience, he may be unsure which mode would best satisfy his preference. On the other hand, the number of ADAS modes provided may be severely limited, which can restrict driver choice and make it difficult to find the most suitable mode. As a result, driver-adaptive ADAS that can automatically satisfy the preferences of different drivers has become a research hotspot in recent years.
The current primary approach for developing driver-adaptive ADAS is to learn and mimic the naturalistic driving characteristics of the current driver, aiming to make system characteristics satisfying the driver's preference. Naturalistic driving characteristics are the behavior and performance exhibited during a driver's manual driving process, i.e., when not using ADAS [[
However, whether the characteristics of the LKA system that are most preferred by the driver should align with his own naturalistic driving characteristics has become a key question. Some literature found that when a driver uses ADAS, the system characteristic he prefers may not be completely consistent with his own naturalistic driving characteristic. Some literature [[
In this paper, we focus on the LKA system and present a novel method for developing a Driver-Adaptive Lane-Keeping Assistance (DALKA) system. This method can be employed to initialize the driver preference model in the study [[
- Extracting metrics for describing naturalistic driving characteristics based on action point theory (hereafter, these metrics will be referred to as "naturalistic driving characteristic metrics");
- Introducing subjective and objective evaluation methods to obtain the test drivers' real preferences to the LKA system, making model training possible;
- Instead of having the LKA system directly mimic the driver's naturalistic driving characteristics, employing machine learning models to train a model using the driver's individual driving characteristics and their real preferred LKA system characteristics and integrating the model-predicted drivers' real preferences into the LKA system.
The remaining content of this paper is organized as follows. Section 2 introduces the development method of DALKA. Section 3 describes the experimental platform and process. Section 4 presents the drivers' real preferences, which were used to train the model, and the predictive performance of the model. Section 5 explains how the predicted preferences are integrated into the LKA system, along with the results of the validation experiments. Section 6 gives a summary of the entire paper and potential issues for further research.
The implementation roadmap of the proposed DALKA system is illustrated in Figure 1. In this implementation path, we follow the approach of configuring the LKA system parameters based on the analysis of naturalistic driving data. However, to better align the system characteristics with drivers' real preferences, we introduce the "Driver Preference Prediction Model (DPPM)" in the implementation roadmap. In the DALKA system described in this paper, we do not conduct research on the environment perception module. The key focus of this study will be on the naturalistic driving data analysis module, DPPMs, and the LKA decision and control module.
The development roadmap for the DPPMs is illustrated in Figure 2. To train the DPPMs, it is necessary to obtain the test drivers' real preferences for the LKA system during the model training phase. The real preferences, along with the driver's naturalistic driving characteristic metrics, are used as training samples to train the DPPM.
Naturalistic driving characteristics are the driving behaviors and performance during the driver's manual driving process (as stated before). It can provide an intuitive insight into the driving behavior of an individual driver [[
In this study, we collected the lateral offset, steering wheel angle, steering wheel torque, yaw rate, and their first and second derivatives with respect to time as basic variables during the naturalistic driving process. Statistical metrics of these basic variables were computed for all of the driving data to serve as metrics of naturalistic driving. These metrics are categorized into three aspects: basic metrics, steering returning metrics, and frequency-domain metrics. Basic metrics include the mean, standard deviation, 5th percentile, and 95th percentile of basic variables. Steering returning metrics, based on the analysis in ref. [[
The action point theory was first proposed in the study of longitudinal car-following processes. In contrast to modeling the driver, the action point theory is based on the direct analysis of driving processes and driver control behaviors. Action points have a clear physical meaning, making it more straightforward to apply in the analysis of driving characteristics [[
We apply action point theory in longitudinal driving to the lateral naturalistic driving process, extracting action points for the lane-keeping process based on the steering wheel angle and lateral offset. The specific extraction process is the same as the method described in ref. [[
The three action points during the lane-keeping process in naturalistic driving are illustrated in Figure 3, specifically:
- Lane-Keeping Steering Starting Point, LKSSP:
The moment when the driver initiates steering to bring the vehicle back to the center of the lane, typically when perceiving a risk of deviating out of the lane;
- Lane-Keeping Lateral Maximum Deviation Point, LKMDP:
The moment following LKSSP when the lateral offset of the vehicle reaches its peak. At this moment, the vehicle' tendency to deviate toward outside of the lane is stopped, and the driver no longer perceives a risk of lane departure;
- Lane-Keeping Steering Ending Point, LKSEP:
Lane-Keeping Steering Ending Point (LKSEP): The moment after LKMDP when the lateral offset returns to zero or when the velocity relative to the lane (referred to hereafter as "lane-relative velocity") becomes zero. At this point, the driver steers the vehicle back to the lane center, marking the conclusion of one lane-keeping process.
Graph: Figure 3 A piece of data of lane-keeping process and action points: (a) Lateral offset data; (b) Steering wheel angle data.
Graph: sensors-24-01666-g003b.tif
Based on these action points, we segmented naturalistic driving data to extract specific processes that better reflect lateral driving characteristics. The process between LKSSP and LKMDP is defined as the Risk-Perception Process. During this process, due to the continuous trend of the vehicle deviating from the lane, the driver focuses on perceiving the risk of lane departure. The process between LKSSP and LKSEP is defined as Returning Process. In Returning Process, the driver steers the wheel to correct the vehicle's position back to the center of the lane. We also use lateral offset, steering wheel angle, steering wheel angular velocity, yaw rate, and their first and second derivatives as basic variables. The mean, standard deviation, 5th percentile, and 95th percentile of these variables are calculated as naturalistic driving characteristic metrics for Risk-Perception Process and Returning Process specifically.
Additionally, in ref. [[
In traditional design method of subjective and objective evaluation tests, whether based on system models or actual vehicles, diverse system characteristics for subjective evaluation (referred to as "evaluation samples") are generated by altering internal system parameters. However, this approach is constrained by model or mechanical structure limitations, resulting in a limited scope covered by these characteristics. Therefore, we adopted the method used in ref. [[
In order to comprehensively describe the characteristics of the LKA system, we divide the working process of the LKA system into different sub-processes. When the vehicle gradually deviates from the lane and reaches a certain distance from the lane boundary, the LKA system intervenes based on certain intervention rules. It applies torque to the steering wheel to correct the vehicle back to the center of the lane.
Therefore, as shown in Figure 4, the LKA system's working process can be divided into the following phases:
- Intervention timing: This refers to the situation at the moment when LKA system initiates its intervention
- Intervention process: This refers to the process from the moment when the LKA system initiates its intervention
Graph: Figure 4 Illustrations of LKA intervention timing and LKA intervention process.
The LKA intervention timing determines under what conditions the LKA system should start to intervene in the vehicle's pose. The most common LKA intervention strategies are those based on the Distance to Lane Crossing (DLC) threshold [[
In the LKA intervention process, the system initially corrects the vehicle's heading to make it parallel to the lane. At this point, the vehicle has no tendency to deviate further from the lane, and the lateral offset reaches maximum. Subsequently, the system controls the vehicle back to the center of the lane. Therefore, there are three key points in the LKA intervention path: the starting point
However, representing the LKA intervention path with coordinates of these control points may not be intuitive. Therefore, we transformed the coordinates of these points shown in Figure 5a into the variables shown in Figure 5b, which have clearer physical meanings, as shown in Equation (
The variables in Equation (
(
We can keep the total longitudinal distance
(
(
As shown in Figure 5b, in Equation (
Finally, we can eliminate the Bezier curve control arm lengths
We employed the uniform design method in experimental design to achieve an even distribution of metrics across various samples. This method eliminates the necessity for numerous repetitive experiments and demonstrates a certain robustness to variations in the model [[
Relevant research has previously proposed subjective evaluation questions related to driver perception in LKA intervention timing [[
Sample design metrics are only used for constructing evaluation samples. To establish a subjective and objective evaluation model, objective metrics still need to be extracted. The extracted objective metrics are presented in Table A1.
Utilizing natural driving metrics of drivers to predict their preferences for LKA system is fundamentally a regression problem. Random Forest (RF) is a Bagging-style ensemble learning method based on decision trees or regression trees. Given the difficulty of obtaining extensive experimental data through subjective evaluation tests, among various machine learning methods, RF stands out for its advantages in controlling model overfitting and requiring a smaller amount of data. Additionally, the method's importance ranking based on node impurity provides excellent support for model analysis. Therefore, we choose RF method to train the model in predicting driver preferences, referred to as the Driver Preference Prediction Model (DPPM). The modeling approach of Random Forest can be referenced from [[
We focused on the research of lateral natural driving characteristics and the LKA system. In order to eliminate the influence of different drivers' longitudinal speed control abilities on their steering control during naturalistic driving and the perception of the LKA system, we ensured that drivers did not need to control the longitudinal speed during the experiments. A constant speed of 80 km/h was set for the experiments.
The procedure of the lane-keeping data-collection test is outlined in Table A2. The subjective evaluation tests were divided into three tests for each working process of the LKA system. The procedures for each test of LKA working processes are shown in Table A3.
Our experiments were conducted on a fixed-base driving simulator. It consists of three main components: a steering feedback simulation device consisting of a Steering-Force-Feedback Actuator (FFA) system, a rapid prototyping controller for vehicle dynamics, an EPS model, and LKA controller computations, as well as a computer with a screen for generating virtual reality environments and simulating traffic flow.
The overall architecture and a physical illustration of the driving simulator are shown in Figure 6.
We recruited test drivers for naturalistic driving data-collection tests and subjective evaluation tests. These drivers had a certain experience and understanding of the LKA system. We primarily selected researchers with more than 3 years of driving experience engaged in relevant research projects and engineers from automotive companies. Driver information is shown in Table 4.
In this section, the results of drivers' real preferences with regard to LKA intervention timing and the LKA intervention process will be presented. Firstly, we establish models comparing subjective evaluations and objective metrics. Subsequently, the obtained models are analyzed to identify the key metrics that influence drivers' subjective evaluations. These metrics are applied to the subsequent LKA decision and control module. Finally, drivers' real preferred values for these metrics can be obtained based on optimal subjective ratings.
For LKA intervention timing, a linear model can effectively represent the relationship between a driver's subjective evaluation and objective metrics
(
In the equation,
By setting
(
From Equation (
(
Equation (
(
By setting
Combining Table 4, we can explore the relationship between age and drivers' preferences for LKA intervention timing, as shown in Figure 7. It can be observed that although the relationship between age and preference is not very clear, drivers aged 30 and above tend to prefer a larger (i.e., safer)
Regarding subjective evaluation questions of the LKA intervention process
During the training of the Random Forest, in the process of building each individual base regression tree, the impurity of each input (i.e., the objective metrics in this paper) is calculated. The objective metric with the lowest impurity at each node is selected for partitioning, resulting in the creation of new subsets for further splitting. Therefore, recording the impurity of nodes during the training process can serve as a basis for assessing the importance of each objective metric, allowing for their importance ranking and potential feature reduction [[
(
where
Based on the node impurity of objective metrics, the most important objective evaluation metrics can be identified. For
The extraction of important objective evaluation metrics based on node impurity cannot avoid internal correlations among these metrics, leading to potential information redundancy. The correlations among these metrics were analyzed. The correlation coefficients between
We utilized a grid search to optimize the value of important objective metrics to obtain preferences. The objective metrics values for the LKA intervention process preferences of eight drivers are shown in Table 6.
Combining Table 4, we can also explore the relationship between age and drivers' preferences for the LKA intervention process, as shown in Figure 8. There is almost no clear relationship between age and drivers' preference for
We used 80% of the data from the dataset as the training set and the remaining 20% as the test set. The predicted values and actual values for
The predicted values and actual values for
Although the initial demonstration of the model's predictive performance in Section 4.2 through MAE provides insights, there is still a lack of established indices for determining an appropriate level of accuracy. In this study, a greater deviation between the metric's value predicted by DPPM and the actual value from drivers could result in lower subjective ratings for the DALKA system. This deviation may potentially extend beyond the acceptable range shown in Table 3. To address this, we introduced two indices: the tolerance
(
According to Table 3, drivers are considered within an acceptable range when their subjective ratings fall within
As shown in Figure 13, for the prediction of
As shown in Figure 14, for the prediction of
For subjective and objective evaluation models of the LKA intervention process, trained using RF models, predictions for various inputs are obtained by traversing the input space. This process allows us to determine the input ranges corresponding to outputs within the
As shown in Figure 15, for the prediction of
As shown in Figure 16, for the prediction of
The LKA decision and control module consist of state decision module, path planning and control module, and output torque decision module, as shown in Figure 17.
We have extracted key metrics influencing drivers' preferences:
The decision logic is illustrated in Figure 18.
In the state decision module, a new variable is introduced, which is the steering assistance torque gain coefficient
Initially, the system receives the LKA system switch signal from the human–machine control panel. If the driver deactivates the LKA system, the system enters the off state, setting
- When the system confirms that the driver has activated the LKA system, it receives the status "If at least one lane line can be effectively detected" from the environment-perception module. If the status is "No," indicating insufficient conditions for activating the LKA system, the system again enters the off state with
- If the environment-perception module confirms effective lane line detection, it evaluates the risk of the vehicle deviating from the lane by checking if the current DLC satisfies Equation (
10 ): (10 )11 ): (11 )10 ) is not met, the LKA system remains standby with - If Equation (
10 ) is satisfied, it is necessary to determine whether the driver has the intention of actively steering. We adopted the method proposed in refs. [[34 ]] to judge the driver's intention to steer actively based on the steering wheel torque threshold12 ). If Equation (12 ) is not satisfied,
(
Regarding the path planning and control module, the path-planning method uses the same approach described in Section 2.3.1 when constructing the characteristics of the LKA intervention process. Regarding path-tracking control, numerous scholars have conducted research. Common methods include Linear Quadratic Regulator (LQR) control [[
Regarding the logic in the output torque decision module, assuming the current state is at step
(
If Equation (
(
If Equation (
(
(
Furthermore, the slope-constraint module restricts the rate of change of
To validate the effectiveness of the DALKA system, an additional subjective evaluation test was conducted by inviting 12 drivers who had not participated in the previous subjective and objective evaluation experiments. A comprehensive subjective evaluation was used to assess the overall performance of the integrated DALKA system. Ratings were given on a scale of
When using the DALKA LKA system, drivers gave an average subjective rating of 4.56, compared to 4.40 when using the fixed-characteristic LKA system. Regarding the acceptance of the drivers, it can be found that when using the LKA system with averagely preferred characteristic, 10 out of 12 drivers (83%) gave subjective ratings within the acceptable range. After experiencing the DALKA system, six drivers showed an improvement in subjective evaluations, three drivers maintained their subjective evaluations, and three drivers experienced a slight decline. However, the subjective evaluations of these three drivers remained within the acceptable range. In summary, subjective evaluations for the DALKA system from all 12 drivers (100%) fell within the acceptable range.
It can be observed that the DALKA system we developed demonstrates more pronounced adaptive effects for those drivers whose preference deviate significantly from the average preference. However, for drivers whose preferences align closely with the average preference, the DALKA system may lead to a decrease in subjective evaluation. Nonetheless, as these drivers already give high subjective evaluations for the average preference characteristics, it does not result in their evaluations falling outside the acceptable range.
Fixed, singular LKA system characteristics struggle to satisfy various preferences of different drivers. The DALKA system, which mimics drivers' individual driving characteristics, addresses the issue of not meeting their real preferences.
The methodology presented in this paper, based on subjective and objective evaluations, provides a novel method of DALKA system development. Firstly, driver preferences to various LKA system processes are obtained through subjective and objective evaluation tests. Secondly, naturalistic driving characteristics are analyzed using action point theory to effectively describe individual lateral driving characteristics. Finally, DPPMs are built using the Random Forest method to predict LKA system characteristics preferred by drivers. The results show that, for
However, several limitations need to be addressed. The DALKA system development and related experiments in this paper are based on straight-road conditions. Further research is needed to explore other conditions such as curves. Additionally, while the DALKA system designed based on the DPPM model improves the acceptance of the LKA system for most drivers, there is a small group of drivers with reduced evaluation. For these drivers, it is necessary to explore more factors influencing driver preferences regarding the LKA system.
Graph: Figure 1 The implementation roadmap of the DALKA system.
Graph: Figure 2 The development roadmap for the DPPMs.
Graph: Figure 5 Path of LKA intervention process: (a) Key points and Bezier curve control points; (b) Objective metrics of path.
Graph: sensors-24-01666-g005b.tif
Graph: Figure 6 The driving simulator: (a) The overall architecture; (b) The physical illustration.
Graph: Figure 7 The relationship between drivers' ages and their preference for LKA intervention timing: (a) offsetVB ; (b) TLCVB.
Graph: Figure 8 The relationship between drivers' age and their preferences for LKA intervention process: (a) ωr-mean ; (b) DLCmin.
Graph: Figure 9 The predicted values and actual values for offsetVB : (a) Training set; (b) Testing set.
Graph: Figure 10 The predicted values and actual values for TLCVB : (a) Training set; (b) Testing set.
Graph: Figure 11 The predicted values and actual values for ωr-mean : (a) Training set; (b) Testing set.
Graph: Figure 12 The predicted values and actual values for DLCmin : (a) Training set; (b) Testing set.
Graph: Figure 13 DPPM's predicted values for offsetVB on the test set compared to the tolerance.
Graph: Figure 14 DPPM's predicted values for TLCVB on the test set compared to the tolerance.
Graph: Figure 15 DPPM's predicted values for ωr-mean on the test set compared to the tolerance.
Graph: Figure 16 DPPM's predicted values for DLCmin on the test set compared to the tolerance.
Graph: Figure 17 LKA decision and control module.
Graph: Figure 18 The decision logic for LKA system.
Graph: Figure 19 Comparison of drivers' subjective evaluations of fixed-characteristic LKA system and DALKA system.
Table 1 Value of metrics under different evaluation samples for LKA intervention timing.
No. 1 0.0 0.15 2 0.1 0.10 3 0.2 0.05 4 0.3 0.20 5 0.4 0.45 6 0.5 0.35 7 0.6 0.25 8 0.7 0.50 9 0.8 0.40 10 0.9 0.30
Table 2 Value of the metrics under different evaluation samples for LKA intervention process.
No. 1 0.20 90 0.3 2 0.35 85 0.7 3 0.50 80 0.2 4 0.15 75 0.6 5 0.30 70 0.1 6 0.45 65 0.5 7 0.10 60 0.0 8 0.25 55 0.4 9 0.40 50 0.8
Table 3 The subjective evaluation questions, scoring range, and optimal scores for LKA system.
Category Subjective Evaluation Question Scoring Range Acceptable Range Optimal Score LKA intervention timing : Is the intervention timing acceptable? [−4,4] [−1,1] 0 LKA intervention process : Is the process of vehicle returning to road center acceptable? [−4,4] [−1,1] 0 : During the intervention, is the minimum distance to lane line acceptable? [−4,4] [−1,1] 0
Table 4 Driver information.
No. of Driver Job Age Driving Experience 1 Researcher 25 4 2 Researcher 24 5 3 Researcher 24 4 4 Researcher 26 4 5 Researcher 25 5 6 Engineer 38 10 7 Engineer 30 3 8 Other 25 3 9 Other 35 8 10 Engineer 25 4 11 Researcher 24 3
Table 5 Driver's preferences for LKA intervention timing.
No. of Driver 1 0.31 0.68 2 0.66 0.62 3 0.32 0.77 4 0.42 0.70 5 0.39 1.08 6 0.74 0.30 7 0.66 0.65 8 0.35 0.39 9 0.89 0.39 10 0.26 0.60
Table 6 Driver's preferences for LKA intervention process.
No. of Driver 1 0.45 0.51 2 0.54 0.51 3 0.03 0.36 4 0.11 0.47 5 0.27 0.56 7 0.34 0.27 9 0.34 0.19 11 0.36 0.35
Table 7 Drivers' subjective evaluations of fixed-characteristic LKA system and DALKA system.
No. of Driver Subjective Ratings of Fixed-Characteristic LKA System Does the Driver Find the Fixed-Characteristic LKA System Acceptable? Subjective Ratings of DALKA System Does the Driver Find the DALKA System Acceptable? 1 3.67 No 4.08 Yes 2 3.83 No 4.58 Yes 3 4.17 Yes 4.83 Yes 4 4.17 Yes 4.75 Yes 5 4.67 Yes 4.83 Yes 6 4.83 Yes 5.00 Yes 7 4.50 Yes 4.50 Yes 8 5.00 Yes 5.00 Yes 9 4.33 Yes 4.33 Yes 10 4.67 Yes 4.50 Yes 11 4.58 Yes 4.33 Yes 12 4.33 Yes 4.00 Yes Average 4.40 83% 4.56 100%
Conceptualization, J.C., H.C. and X.L.; methodology, J.C.; software, J.C., X.L. and B.Z.; validation, J.C., X.L. and B.Z.; formal analysis, J.C.; investigation, J.C., X.L. and W.R.; resources, J.C., X.L. and B.Z.; data curation, J.C., X.L. and B.Z.; writing—original draft preparation, J.C.; writing—review and editing, J.C., H.C. and W.R. All authors have read and agreed to the published version of the manuscript.
Not applicable.
Informed consent was obtained from all subjects involved in the study.
Data are contained within the article.
The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analysis, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
The author is sincerely grateful to the reviewers for their insightful commen-dation and suggestions. Thanks to Nishimura Yosuke, Kei Kitahara from JTEKT Corporation and Youyu Yin from JTEKT Research and Development Center (WUXI), for their help and work in methodology investigation, data analysis and paper review. Also, thanks to Xiaolin He, Quyi Liu and Taokai Xia for helping to prepare the simulator experiments.
The extracted objective metrics used for subjective and objective model training of the LKA system are shown in Table A1.
Table A1 Objective metrics of LKA system.
No. Description of Metrics Symbol Unit 1 DLC when LKA initiates intervention m 2 Lane-relative velocity when initiating intervention m/s 3 TLC when LKA initiates intervention s 4 Maximum steering wheel torque change rate N·m/s 5 Maximum steering wheel torque N·m 6 Average steering wheel torque N·m 7 Maximum steering wheel rotation speed deg/s 8 Maximum steering wheel rotation angle deg 9 Average steering wheel rotation angle deg 10 Minimum TLC s 11 Minimum DLC m 12 Maximum DLC m 13 Average DLC m 15 Maximum yaw rate deg/s 16 Average yaw rate deg/s 17 Maximum lane-relative velocity m/s 18 Average lane-relative velocity m/s 19 Intervention duration s 20 Maximum steering wheel torque N·m 21 Average steering wheel torque N·m·s
The procedure for the lane-keeping data-collection test is outlined in Table A2.
Table A2 Procedures of naturalistic driving data-collection tests.
Test Category Test Procedure Naturalistic driving data-collection test ① Have the driver operate the driving simulator for at least 10 min to familiarize themselves with the test environment. Inform them in advance about the location of the lane boundaries to minimize the perceptual differences between the simulated and real environments. ② Ask the driver to simulate their real driving process as closely as possible, but keep the vehicle within the center lane with continuous traffic flow on both sides of the lane. ③ Data collection is ended after 1 h.
The procedures for each test of the LKA working processes are shown in Table A3.
Table A3 Procedures of subjective evaluation tests for LKA system.
Test Category Test Procedure Test of LKA intervention timing ① The vehicle is controlled along the center of the lane. The LKA system does not initiate intervention as the vehicle remains within the lane center. ② Choose one evaluation sample for LKA intervention timing shown in Table 1. By applying crosswinds in virtual environment, make the vehicle deviate from the lane with preset . ③ Initiate the LKA system intervention when the vehicle deviates to a certain degree, controlling the vehicle to return to the center of the lane. Subsequently, end the LKA system intervention and return to the state of procedure ①. ④ Repeat procedures ① to ③ multiple times, allowing the driver to fully experience the LKA intervention timing. ⑤ Let the driver give subjective ratings to the evaluation questions in Table 3 based on his current experience of LKA intervention timing. ⑥ Select another LKA intervention timing sample shown in Table 1, and repeat procedures ① to ⑤ until subjective ratings have been collected for all evaluation samples. ⑦ Randomly select several evaluation samples for test driver and ask him to give subjective ratings repletely, ensuring consistent ratings for same evaluation sample. Repeat this procedure until the driver's ratings stabilize. Test of LKA intervention process ① The vehicle is controlled along the center of the lane. The LKA system does not initiate intervention as the vehicle remains within the lane center. ② Choose one evaluation sample for LKA intervention process shown in Table 2. By applying crosswinds in virtual environment, make the vehicle deviate from the lane with preset . ③ Initiate the LKA system intervention when the vehicle deviates to a certain degree, controlling the vehicle to return to the center of the lane. Subsequently, end the LKA system intervention and return to the state of procedure ①. ④ Repeat procedures ① to ③ multiple times, allowing the driver to fully experience the LKA intervention process; ⑤ Let the driver give subjective ratings of the evaluation questions in Table 3 based on his current experience of LKA intervention process. ⑥ Select another LKA intervention process sample shown in Table 2, and repeat procedures ① to ⑤ until subjective ratings have been collected for all evaluation samples. ⑦ Randomly select several evaluation samples for test driver and ask him to give subjective ratings repletely, ensuring consistent ratings of same evaluation sample. Repeat this procedure until the driver's ratings stabilize.
By Jiachen Chen; Hui Chen; Xiaoming Lan; Bin Zhong and Wei Ran
Reported by Author; Author; Author; Author; Author