Automatic driving method, automatic driving device and vehicle

Abstract

The invention relates to an automatic driving method, an automatic driving device and a vehicle. The method comprises the following steps: predicting a driving parameter sequence when the vehicle executes a currently generated candidate trajectory; the driving parameter sequence comprises specified driving parameters of the vehicle in each predicted time step; according to the specified driving parameters under each prediction time step, determining a comfort index under the prediction time step by using a pre-trained comfort evaluation model; according to the comfort index under each prediction time step, determining the comfort cost of the candidate track, and determining the estimated cost of the candidate track under each preset cost item; selecting a target trajectory from the plurality of candidate trajectories according to the comfort cost of the candidate trajectory and the estimated cost under each preset cost item; wherein the target trajectory is a trajectory taking comfort cost and estimated cost into consideration at the same time; and controlling the vehicle to run according to the target trajectory. By adopting the method, the problem of low comfort in the automatic driving process can be solved.

Description

Technical Field

[0001] This application relates to the field of autonomous driving, and in particular to autonomous driving methods, autonomous driving devices, and vehicles. Background Technology

[0002] The ultimate goal of autonomous driving technology is to completely liberate humans from driving tasks. In the process of implementing autonomous driving technology, traditional research and testing have almost entirely focused on safety, reliability, and basic driving efficiency. Therefore, in autonomous driving mode, the core mission of the vehicle's decision-making system is to avoid collisions, obey traffic rules, and ensure the vehicle arrives at its destination safely and efficiently.

[0003] As autonomous driving technology matures and moves towards large-scale commercial use, the riding experience has become a key factor influencing user acceptance and commercial success. Even if safety is ensured technically, if the ride is bumpy, abrupt, or causes discomfort such as tension or motion sickness, such autonomous driving, while meeting basic safety requirements, ignores the passenger's need for comfort as an individual with subjective feelings.

[0004] There is currently no effective solution to the problem of low comfort during autonomous driving in related technologies. Summary of the Invention

[0005] In view of this, an autonomous driving method, an autonomous driving device, and a vehicle are provided to improve the comfort of autonomous driving.

[0006] Firstly, this embodiment provides an autonomous driving method, the method comprising:

[0007] For each of the multiple candidate trajectories generated so far, predict the sequence of driving parameters for the vehicle when executing that candidate trajectory; wherein, the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step;

[0008] Based on the specified driving parameters at each predicted time step, the comfort index at that predicted time step is determined using a pre-trained comfort evaluation model.

[0009] Based on the comfort index at each predicted time step, determine the comfort cost of the candidate trajectory, and determine the estimated cost of the candidate trajectory under each preset cost item.

[0010] Based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item, a target trajectory is selected from the multiple candidate trajectories; wherein, the target trajectory is a trajectory that simultaneously considers the comfort cost and the estimated cost;

[0011] Control the vehicle to travel along the target trajectory.

[0012] In some embodiments, the training process of the pre-trained comfort evaluation model includes:

[0013] Acquire the set of historical driving parameters and historical feedback data at historical driving times; the historical feedback data includes user status parameters and / or user feedback comfort categories;

[0014] The target comfort index for each historical moment is determined based on historical feedback data from each historical driving period; wherein, the target comfort index is used to characterize the user's riding comfort level;

[0015] Based on the historical driving parameter set and historical feedback data at each historical driving time, specified driving parameters related to ride comfort are selected from the historical driving parameter set, and an initial model is constructed based on the specified driving parameters; wherein, the initial model is used to characterize the mapping relationship between the specified driving parameters and the comfort index;

[0016] Training samples are established based on the specified driving parameters and target comfort index at each historical time, and the initial model is trained using the training samples to obtain the pre-trained comfort evaluation model.

[0017] In some embodiments, the method further includes:

[0018] During the process of controlling the vehicle to travel along the target trajectory, when the specified conditions are met at the current moment, the current driving parameters and the user feedback data at the current moment are obtained; wherein, the current driving parameters include at least the specified driving parameters; the user feedback data includes the comfort category or user status parameters.

[0019] Based on the user feedback data, determine the current comfort index at the current moment, and construct the correspondence between the current driving parameters and the current comfort index.

[0020] The comfort evaluation model is updated using the correspondence.

[0021] In some embodiments, selecting the target trajectory from the plurality of candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under various preset cost items includes:

[0022] Determine the first weight of the comfort cost and the second weight of each preset cost item;

[0023] The comfort cost and the estimated cost are weighted according to the first weight and the second weight, and the weighted result is determined as the comprehensive score of the candidate trajectory.

[0024] The target trajectory is selected from the multiple candidate trajectories based on the comprehensive score of each candidate trajectory.

[0025] In some embodiments, determining the first weight of the comfort cost and the second weight of each preset cost item includes:

[0026] Obtain the preset first and second weights;

[0027] The first weight and / or the second weight are adjusted based on the vehicle's current driving parameters and / or the user's current preferences.

[0028] In some embodiments, the comfort evaluation model includes multiple first sub-models corresponding to different users; each first sub-model is obtained based on user data of the same user; the step of determining the comfort index for each predicted time step using a pre-trained comfort evaluation model based on specified driving parameters for each predicted time step includes:

[0029] Find the first target sub-model that matches the current user from among the plurality of first sub-models;

[0030] The specified driving parameters for each predicted time step are input into the first target sub-model, so that the first target sub-model outputs the comfort index for that predicted time step.

[0031] In some embodiments, the comfort evaluation model includes multiple second sub-models corresponding to different scenarios, each second sub-model being derived from user data within the same scenario; the step of determining the comfort index for each predicted time step using a pre-trained comfort evaluation model based on specified driving parameters for that predicted time step includes:

[0032] The current situation is determined based on the vehicle's current driving parameters and / or user feedback data at the current moment;

[0033] Find the second target sub-model that matches the current situation from among the plurality of second sub-models;

[0034] The driving parameters at each predicted time step are input into the second target sub-model, so that the second target sub-model outputs the comfort index at that predicted time step.

[0035] In some embodiments, the method further includes:

[0036] During the process of controlling the vehicle to travel along the target trajectory, if the comfort index corresponding to the collected specified feedback data is less than a specified threshold, the driving parameters of the target trajectory are adjusted according to preset optimization rules to improve the ride comfort of the vehicle.

[0037] Secondly, this embodiment provides an autonomous driving device, including:

[0038] The driving simulation module is used to predict the sequence of driving parameters of a vehicle when executing a candidate trajectory among multiple candidate trajectories generated at the moment; wherein, the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step;

[0039] The prediction module is used to determine the comfort index for each prediction time step based on the specified driving parameters at each prediction time step using a pre-trained comfort evaluation model.

[0040] The cost estimation module is used to determine the comfort cost of the candidate trajectory based on the comfort index at each predicted time step, and to determine the estimated cost of the candidate trajectory under each preset cost item.

[0041] The trajectory selection module is used to select a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under various preset cost items; wherein, the target trajectory is a trajectory that simultaneously considers the comfort cost and the estimated cost;

[0042] The control module is used to control the vehicle to travel along the target trajectory.

[0043] Thirdly, this embodiment provides a vehicle having a memory and a processor, the memory storing a computer program, and the processor being configured to run the computer program to implement the autonomous driving method described in the first aspect above.

[0044] The aforementioned autonomous driving method, autonomous driving device, and vehicle, for each of the multiple candidate trajectories generated, predict the sequence of driving parameters the vehicle will use when executing that candidate trajectory. Based on the specified driving parameters at each predicted time step, a pre-trained comfort evaluation model is used to determine the comfort index at that predicted time step. Then, based on the comfort indices at each predicted time step, the comfort cost of that candidate trajectory is determined, along with the estimated cost of that candidate trajectory under each preset cost item. Based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item, a target trajectory is selected from the multiple candidate trajectories. This allows the vehicle to be controlled to drive along the target trajectory. In this way, the pre-trained comfort evaluation model maps driving parameters to corresponding comfort indices, thus forming a comfort cost reflecting the overall riding experience of the trajectory. Furthermore, by incorporating the comfort cost into the candidate trajectory decision-making process, trajectory selection is no longer limited to simple rule judgments but rather optimizes comfort holistically from the perspective of temporal continuity. Therefore, the selected target trajectory can effectively reduce unnecessary rapid acceleration, deceleration and sharp turns, significantly improve the ride comfort during vehicle operation and enhance the user experience. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating an embodiment of the autonomous driving method of this application;

[0046] Figure 2 This is a flowchart illustrating Embodiment 2 of the autonomous driving method of this application;

[0047] Figure 3 This is a flowchart illustrating Embodiment 3 of the autonomous driving method of this application;

[0048] Figure 4 This is a flowchart illustrating Embodiment 4 of the autonomous driving method of this application;

[0049] Figure 5 This is a flowchart illustrating Embodiment 5 of the autonomous driving method of this application;

[0050] Figure 6 This is a structural block diagram of an embodiment of the autonomous driving device provided in this application;

[0051] Figure 7 This is a hardware structure diagram of a vehicle shown as an exemplary embodiment of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0053] In one embodiment, such as Figure 1 As shown, an autonomous driving method is provided, which is applied to... Figure 1 Taking the terminal in the example, the following is an explanation:

[0054] Step S102: For each of the multiple candidate trajectories generated so far, predict the sequence of driving parameters of the vehicle when executing the candidate trajectory; wherein, the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step.

[0055] The candidate trajectory is the achievable vehicle trajectory during the journey to the destination. The vehicle's driving parameter sequence is a set of specified driving parameters for the vehicle at multiple predicted time steps. These specified driving parameters are physical parameters directly related to vehicle movement and significantly impacting user comfort. Optionally, the specified driving parameters may include: dynamic data, and / or, external environmental data during vehicle movement. Dynamic data may include longitudinal acceleration, lateral acceleration, jerk, etc.; external environmental data may include external road conditions, obstacles, traffic lights, etc.

[0056] Optionally, candidate trajectories are generated using either of the following methods: a trajectory generation method based on decoupled lateral and longitudinal sampling, or a trajectory generation method based on grid search. The decoupled lateral and longitudinal sampling method operates in the Frenet coordinate system, achieving lane-level planning through lateral planning (sampling the destination and generating a path) and speed-level planning through longitudinal planning (sampling driving behavior and generating a speed curve). The grid-based search method discretizes the current vehicle's state space into a grid, and... The algorithm dynamically searches for a geometric path from the starting point to the ending point in a gridded space. After generating the geometric path, it is synthesized with the corresponding vehicle speed and time to obtain the spatiotemporal trajectory (x(t), y(t)) in Cartesian coordinates, i.e., the candidate trajectory. The specific implementation principles and processes of these two generation methods can be found in related technical descriptions, and will not be elaborated upon here.

[0057] After generating candidate trajectories using methods based on lateral and longitudinal decoupling sampling and / or grid search, it is further necessary to determine whether the trajectories meet smoothness and feasibility constraints. Specifically, determining whether the smoothness constraint is met includes: judging whether one or more of the driving parameters of the candidate trajectory, such as position, velocity, acceleration, and jerk, are continuous and smooth; if not, the candidate trajectory is deleted. To improve the smoothness of the candidate trajectory, higher-order polynomial interpolation can be performed on the driving parameters of the candidate trajectory during generation. Further, determining whether the feasibility constraint is met includes: judging whether the trajectory is allowed by the vehicle's dynamics and actuator physical limits (such as maximum curvature / steering angle, maximum acceleration); if not, the candidate trajectory is deleted.

[0058] Optionally, for each candidate trajectory (x(t), y(t)), the vehicle dynamics model can be used to predict the vehicle's driving parameters at each prediction time step when executing the candidate trajectory. Then, a sequence of driving parameters can be obtained by combining the specified driving parameters at each time step. For details on the specific implementation principle and process of outputting the predicted driving parameter sequence of the vehicle when executing the candidate trajectory based on the vehicle dynamics model, please refer to the descriptions in related technologies; they will not be repeated here.

[0059] Step S104: Based on the specified driving parameters at each predicted time step, determine the comfort index at that predicted time step using a pre-trained comfort evaluation model.

[0060] The comfort index quantifies the degree of discomfort experienced by passengers while riding in a vehicle. A smaller comfort index value indicates poorer ride comfort, meaning higher levels of discomfort; conversely, a larger value indicates better ride comfort, meaning lower levels of discomfort. A pre-trained comfort evaluation model indicates the mapping relationship between specified vehicle driving parameters and the comfort index; its input is the specified driving parameters, and its output is the comfort index.

[0061] Optionally, the specified driving parameters include at least one or more of the following: vehicle position (x,y), velocity v, acceleration a, jerk j, curvature acceleration, and centripetal acceleration.

[0062] Optionally, the specified driving parameters for each prediction time step are input into a pre-trained comfort evaluation model. The comfort evaluation model quantifies and outputs a comfort index within a preset numerical range based on the ride comfort corresponding to the specified driving parameters. The preset numerical range can be set to 1 to 100. Optionally, after obtaining the ride comfort corresponding to the specified driving parameters, the comfort index corresponding to each specified driving parameter can be normalized to unify the evaluation scale and output the comfort index for that prediction time step. The normalization range can be set to 0 to 1. It should be understood that the aforementioned preset numerical range and normalization range can also be set to other values ​​as needed.

[0063] The comfort evaluation model is trained based on the correspondence between specified driving parameters and comfort indices. For example, the pre-trained comfort evaluation model can be a general model. The general model can be a model pre-trained based on the correlation between specified driving parameters and comfort levels under different users and / or different situations, and it is adapted to different users and / or different situations.

[0064] Step S106: Determine the comfort cost of the candidate trajectory based on the comfort index at each predicted time step, and determine the estimated cost of the candidate trajectory under each preset cost item.

[0065] The comfort cost is used to measure the comfort level of a candidate trajectory. The comfort cost can be determined based on at least one of the following feature parameters: the minimum comfort index in the candidate trajectory, the duration for which the comfort index is less than a specified threshold, and the rate of change of the comfort index. Optionally, the comfort cost can be obtained by weighted summation of the feature parameters according to preset weights. Alternatively, one of the above feature parameters can be directly used as the comfort cost; or, the above feature parameter can be normalized to obtain the comfort cost. For example, the integral of the comfort index over the entire candidate trajectory can be used as the comfort cost.

[0066] For example, when a vehicle executes a candidate trajectory, the driving parameters of the vehicle at each predicted time step are input into the comfort evaluation model to calculate the comfort index corresponding to each predicted time step, thus obtaining a "comfort index-time" curve. The aforementioned characteristic parameters can be obtained from the "comfort index-time" curve. Specifically, the maximum value of the comfort index in the candidate trajectory, i.e., peak discomfort, is obtained from the "comfort index-time" curve; the duration of discomfort is obtained from the duration of the comfort index exceeding a specified threshold; and the trend of discomfort change is obtained from the rate of change of the comfort index.

[0067] In one possible implementation, the preset cost items include safety cost and / or efficiency cost. For example, the safety cost is used to quantify the level of safety when driving based on a candidate trajectory; the safety cost can be determined based on one or more safety-related driving parameters such as the minimum distance between the vehicle and obstacles and the maximum vehicle speed in the candidate trajectory. In specific implementations, the specific calculation method refers to the description in related technologies, for example, normalizing the safety-related driving parameters and then weighting and summing them to form a safety cost that can characterize the safety level of the trajectory. The efficiency cost is used to quantify the efficiency of driving to the destination based on the candidate trajectory; the efficiency cost can be determined based on one or more driving parameters related to the efficiency of the vehicle reaching the destination, such as the vehicle's travel time and energy consumption in the candidate trajectory. The specific calculation method can refer to the description in related technologies, for example, normalizing the efficiency-related driving parameters and then weighting and summing them to form an efficiency cost that can characterize the efficiency of the trajectory driving.

[0068] Step S108: Select a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item; wherein, the target trajectory is a trajectory that takes into account both comfort cost and estimated cost.

[0069] Optionally, the static cost of candidate trajectories can be estimated based on comfort cost and estimated cost: comfort cost is used as the core cost item, and a cost function is constructed by combining it with preset cost items; specifically, the cost function can be based on demand, treating comfort and each estimated cost as independent cost items, and integrating each cost item according to preset weights to calculate the value of the cost function. The value of the cost function is used as the comprehensive score of the candidate trajectory; where, the lower the comprehensive score, the better the riding experience corresponding to the candidate trajectory; conversely, the higher the comprehensive score, the worse the riding experience corresponding to the candidate trajectory; thus, the candidate trajectory with the lowest comprehensive score can be selected as the target trajectory. Further, the preset cost items include a safety cost item; after determining the estimated cost and comfort cost corresponding to the preset cost items of each candidate trajectory and constructing the cost function, the safety cost of the safety cost item can be set as a hard constraint for constructing the cost function, that is, when the safety cost exceeds a preset upper limit threshold, the comprehensive score of the candidate trajectory is directly assigned a value much larger than the normal candidate trajectory score range, thereby significantly increasing the comprehensive score so that the candidate trajectory is directly excluded during the optimization process. Subsequently, the target trajectory with the lowest comprehensive score is selected from the set of safe candidate trajectories.

[0070] Optionally, not only can the static cost of the candidate trajectory be estimated, but the dynamic changes in the comfort index of the candidate trajectory at the predicted time step can also be predicted. This allows for a globally optimal decision based on the dynamic changes. For example, the corresponding optimization indicators can be obtained based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item. The optimization direction is to reduce the overall score, and a multi-objective optimization method is used to solve the candidate trajectory. If there exists a candidate trajectory that achieves optimality across all optimization indicators, the candidate trajectory with the lowest overall score is selected as the target trajectory. If there is no candidate trajectory that achieves optimality across all optimization indicators, the Pareto optimal method can be implemented. Without compromising other optimization indicators, any optimization indicator can be improved to find a Pareto optimal solution, and finally, the candidate corresponding to the Pareto optimal solution is selected as the target trajectory. The specific optimization process of the multi-objective optimization method or the Pareto optimal method can be found in the descriptions in related technologies, and will not be elaborated here.

[0071] Step S110: Control the vehicle to travel along the target trajectory.

[0072] Optionally, the vehicle's actuators are controlled according to the driving parameters indicated by the target trajectory, so that the vehicle travels along the target trajectory under the drive of the actuators.

[0073] In the aforementioned autonomous driving method, the driving parameter sequence of the vehicle when executing the candidate trajectory is predicted, and the comfort index of the candidate trajectory at the predicted time step is used by a pre-trained comfort evaluation model. In this way, the user's riding comfort level during vehicle operation can be accurately captured based on the quantified comfort index, and the comfort cost of the candidate trajectory can be determined based on the comfort index. Then, based on the comfort cost of the candidate trajectory and the estimated cost under the preset cost item, the target trajectory that takes into account both comfort cost and estimated cost can be selected from each candidate trajectory. In this way, the user's riding comfort level is included as a factor in the selection of the target trajectory, which improves the comfort of the vehicle when driving along the target trajectory and solves the problem of low comfort in autonomous driving.

[0074] In one embodiment, Figure 2 This is a flowchart illustrating Embodiment 2 of the autonomous driving method in this application, as shown below. Figure 2 As shown, based on the above embodiments, the training process of the pre-trained comfort evaluation model includes:

[0075] Step S201: Obtain the set of historical driving parameters and historical feedback data at the historical driving time; the historical feedback data includes the user's status parameters and / or the user's feedback comfort category.

[0076] Historical feedback data is generated based on user-perceived ride comfort levels while the vehicle is operating according to historical driving parameters. The target comfort index is a quantified representation of ride comfort levels derived from historical feedback data. User state parameters are obtained through an implicit feedback mechanism, while comfort categories are obtained through an explicit feedback mechanism, as detailed below:

[0077] User status parameters are parameters that objectively reflect a user's stress and discomfort, as well as physiological signals that are difficult to conceal, obtained through in-vehicle sensor data collection. These status parameters can be collected by at least one of the following sensors: cameras, microphones, sensors built into the steering wheel or seat, etc.

[0078] For example, after acquiring images using an in-cabin camera, the user's state parameters are analyzed. These state parameters may include the user's micro-expressions, body posture, and head movement stability. Specifically, analyzing micro-expressions can reveal features of the user in uncomfortable states such as frowning or clenching their lips, while analyzing body posture can reveal features of the user in uncomfortable states such as suddenly leaning forward or gripping the handrail tightly.

[0079] For example, after the sound signal is collected by the microphone, recognition is performed based on the sound signal, and the state parameters include preset words related to the recognized level of discomfort in the ride.

[0080] For example, state parameters measured by sensors built into the steering wheel or seat include: skin conductance, heart rate and its variability, etc. It should be understood that data collected by devices such as cameras and microphones must strictly comply with privacy regulations; for example, the collected signals must be analyzed locally and without facial recognition.

[0081] User-reported comfort categories are data obtained by mapping user-initiated comfort ratings to predefined comfort categories using a classification model or rule engine. Comfort categories can be collected through at least one of the following methods: user interaction with physical or tactile feedback buttons within the vehicle; data input through in-vehicle interfaces such as the central control screen or e-ink screen; or data input through mobile terminals or applications associated with the vehicle.

[0082] For example, a simple physical or tactile feedback button with a "comfort rating" can be set on the armrest or door; users can press it instantly when they feel obvious discomfort, and the corresponding comfort category can be identified based on the button that is triggered.

[0083] For example, input through the in-vehicle interface can be voice commands. For instance, through Automatic Speech Recognition (ASR) and Natural Language Processing (NLP) technologies, the comfort evaluation contained in the natural language commands such as "The car is too bumpy" or "The brakes are a bit too strong" input by the user through the interface can be converted into structured comfort labels, that is, to obtain comfort categories.

[0084] For example, the data input through vehicle-connected input devices such as the central control screen and e-ink screen includes: after the trip, a brief survey can be conducted via the touchscreen, such as: "Are you satisfied with the smoothness of this trip?"; or a detailed feedback entry can be provided in the vehicle-connected mobile APP, allowing users to retrospectively evaluate specific road sections or driving behaviors after the trip, thereby obtaining user responses; based on preset mapping rules or classification models, user responses are converted into corresponding comfort categories.

[0085] Step S202: Determine the target comfort index for each historical moment based on the historical feedback data for each historical driving period; wherein, the target comfort index is used to characterize the user's riding comfort level.

[0086] The higher the user comfort level reflected by the state parameters, the lower the target comfort index obtained after mapping; conversely, the lower the level, the lower the target comfort index. Optionally, after obtaining the user's state parameters, a preset mapping rule for indicating the relationship between the state parameters and the comfort index is obtained, and the user state parameters are converted into a comfort index based on the preset mapping rule; or, a classification model for indicating the relationship between the state parameters and user comfort is obtained, a user comfort category corresponding to the state parameters is obtained based on the classification model, and the user comfort category is converted into a corresponding comfort index according to the specified numerical rules.

[0087] Different comfort categories correspond to different target comfort indices. Optionally, after obtaining the comfort categories from user feedback, the comfort categories can be converted into quantified comfort indices according to the set numerical rules.

[0088] Step S203: Based on the historical driving parameter set and historical feedback data at each historical driving time, select the specified driving parameters related to the degree of ride comfort from the historical driving parameter set, and construct an initial model based on the specified driving parameters; wherein, the initial model is used to characterize the mapping relationship between the specified driving parameters and the comfort index.

[0089] The higher the correlation between driving parameters and ride comfort, the greater the change in the target comfort index when the driving parameter value changes; conversely, the lower the correlation, the smaller the change in the target comfort index when the driving parameter value changes. Driving parameters with a correlation greater than a specified level are designated as the specified driving parameters.

[0090] The correlation between a driving parameter and its target comfort index can be obtained by analyzing historical feedback data at various historical driving times for the same driving parameter. For example, based on the average change in the target comfort index corresponding to a unit change in the driving parameter at multiple historical driving times, the correlation value between the driving parameter and the target comfort index is quantified. If the correlation value is greater than a specified threshold, the correlation is determined to be greater than a specified degree. Driving parameters with a correlation greater than the specified degree are then set as designated driving parameters.

[0091] The model architecture of the initial model built based on the specified driving parameters can be a linear regression architecture, a fully connected neural network architecture, a convolutional neural network architecture, etc., and there are no restrictions here.

[0092] Step S204: Establish training samples based on the specified driving parameters and target comfort index at each historical time, and use the training samples to train the initial model to obtain the pre-trained comfort evaluation model.

[0093] Among these methods, online supervised learning and / or reinforcement learning can be used to train the initial model based on training samples constructed from historical feedback data and corresponding specified driving parameters.

[0094] Taking supervised learning as an example, establishing training samples includes: obtaining sample labels to indicate the target comfort index based on historical feedback data; obtaining the context of the training samples based on historical time points and specified driving parameters at those historical time points; and obtaining the features of the training samples based on the specified driving parameters. Subsequently, methods such as online gradient descent or Bayesian update are used to fine-tune the weights of each specified driving parameter in the model based on the training samples.

[0095] The sample labels can be configured as "comfortable" or "uncomfortable." The comfort evaluation model converts the sample labels into corresponding comfort indices; for example, "comfortable" is converted to a comfort index of "0," and "uncomfortable" is converted to a comfort index of "1." Alternatively, the sample labels can be configured as the comfort index itself. It should be understood that the specific settings of the sample labels can be adjusted according to needs and are not restricted here.

[0096] For example, when the comfort evaluation model specifies driving parameters including jerk and acceleration; comfort index = w1 × |Jerk| + w2 × |lateral acceleration|; and w1 and w2 are the weights of jerk and acceleration, respectively, training the initial model involves adjusting w1 and / or w2. For instance, when lateral acceleration is collected at a certain moment, and the corresponding user feedback data label is "uncomfortable," training samples can be constructed based on this lateral acceleration and the corresponding feedback data, and supervised learning can be performed based on these training samples. During training, the model adaptively adjusts the parameters, thereby increasing the weight term w2 corresponding to lateral acceleration. Methods for training models using supervised learning can be found in descriptions of related technologies and will not be repeated here.

[0097] Taking reinforcement learning as an example, the planning of ride comfort is considered a reinforcement learning problem; each decision to select the target trajectory from candidate trajectories is considered an "action," and the comfort index obtained from user feedback data is used as the "reward." Through online reinforcement learning, based on the "action," the "reward," and the specified driving parameters associated with the "action," the model parameters are gradually adjusted to maximize the long-term accumulated "comfort reward," continuously optimizing the decision to select the target trajectory from candidate trajectories, thus adjusting the comfort evaluation model. The method for training the model using reinforcement learning can be found in the descriptions in related technologies, and will not be elaborated here.

[0098] Optionally, the comfort evaluation model trained based on the historical driving parameter set and its corresponding historical feedback data can be any of the following models:

[0099] Based on the set of historical driving parameters and corresponding historical feedback data generated by the same user during the use of autonomous driving function, a personalized model corresponding to that user is trained.

[0100] Based on the historical driving parameter set of a fixed vehicle or multiple vehicles of a fixed model during the autonomous driving process, and the historical feedback data generated by different users in different vehicles during the autonomous driving process, a personalized model corresponding to the fixed vehicle or a fixed model is trained.

[0101] The federated learning approach trains a global model based on the set of historical driving parameters generated by multiple different users during the use of autonomous driving functions and their corresponding historical feedback data.

[0102] In particular, by training or updating the comfort evaluation model based on feedback data and driving parameters from multiple different users, the limitations of traditional testing in covering all corner cases (such as special road seams and specific overtaking scenarios) can be overcome. Through feedback from a large number of real users, comfort issues in these "long tail" scenarios can be continuously discovered and optimized.

[0103] Furthermore, after constructing the comfort evaluation model, during the autonomous driving process based on this model, the historical driving parameter set can be updated based on the collected driving parameters during vehicle operation, and the historical feedback parameters can be updated based on the collected feedback data. Based on the updated historical driving parameter set and historical feedback parameters, designated driving parameters related to ride comfort are re-selected. For example, new designated driving parameters related to ride comfort can be selected from the historical driving parameter set, or if the correlation between the original designated driving parameters and ride comfort is reduced based on the historical driving parameter set, the designated driving parameters can be adjusted, such as adding new designated driving parameters and / or deleting existing ones. Based on the adjusted designated driving parameters and the target comfort index at the corresponding time, new training samples are established, and the initial model is retrained using the new training samples to obtain the pre-trained comfort evaluation model. The method for retraining the initial model is the same as the method described above for training the initial model based on the historical driving parameter set and historical feedback data, and will not be repeated here.

[0104] For ease of understanding, for example, the analysis found that after selecting a new specified driving parameter related to the degree of ride comfort from the historical driving parameter set, the vibration parameter of a certain frequency can be added to the specified driving parameter, and a new training sample containing the vibration parameter can be established. The initial model can then be retrained using the new training sample, so that the next generation of comfort evaluation model can include the driving parameter of the new dimension of "vibration comfort".

[0105] In this embodiment, feedback data is collected through a low-interference, high-reliability channel, and a comfort evaluation model is trained using historical user feedback data and historical driving parameters at historical driving times. This allows the comfort index output by the comfort evaluation model to accurately capture the user's real feelings during vehicle operation, thereby improving the effectiveness of the target trajectory determined from the candidate trajectory in enhancing the comfort of autonomous driving.

[0106] In one embodiment, Figure 3 A flowchart of an embodiment three of an autonomous driving method is also provided, including:

[0107] Step S301: During the process of controlling the vehicle to travel along the target trajectory, when the specified conditions are met at the current moment, acquire the current driving parameters and the user feedback data at the current moment; wherein, the current driving parameters include at least the specified driving parameters; the user feedback data includes the user's feedback comfort category or the user's state parameters. Optionally, when user actively outputs feedback data and / or the current time reaches a preset periodic collection time point, determine that the specified conditions are met at the current moment.

[0108] Step S302: Based on user feedback data, determine the current comfort index at the current moment, and construct the correspondence between the current driving parameters and the current comfort index; use the correspondence to update the comfort evaluation model.

[0109] The updated comfort evaluation model can be a model trained on multiple user data sets, or it can be a model trained on the current user data set; there is no restriction on this. User data includes feedback data generated by the user and the specified driving parameters corresponding to the time the feedback data was generated.

[0110] When a comfort evaluation model is used to weight specified driving parameters using their corresponding weights to obtain a comfort index, the weights corresponding to the specified driving parameters can be updated using the correspondence. It should be understood that the comfort evaluation model can also be a classification model, a neural network, or other architectures; this is not a limitation.

[0111] For example, a comfort evaluation model can be composed of a weighted combination of physical quantities from several vehicle driving parameters: longitudinal dynamic parameters, lateral dynamic parameters, and comfort-related composite parameters. The weights corresponding to the longitudinal dynamic parameters, lateral dynamic parameters, and composite parameters are updated based on updated training samples.

[0112] Since jerk is the rate of change of acceleration, rapid acceleration and braking both result in significant jerk, a major source of motion sickness and discomfort. Sustained excessive positive or negative acceleration can also cause discomfort. Therefore, longitudinal dynamic parameters can include jerk (Jerk) and acceleration (a). Optionally, the comfort index is positively correlated with jerk: comfort index ∝ |Jerk|. Optionally, the comfort index ∝ |a-a_comfy|², where a_comfy is a near-zero acceleration threshold that provides comfort to occupants.

[0113] Since centripetal acceleration or lateral acceleration is related to a vehicle's cornering behavior, excessive centripetal or lateral acceleration indicates that the vehicle is cornering at too high a speed or with too small a turning radius. In this case, the vehicle generates a large lateral force, causing occupants to experience lateral thrust, leading to passenger discomfort. Lateral jerk is related to steering wheel rotation speed; excessive lateral acceleration indicates that the steering wheel is turning too quickly. This results in a sudden change in lateral force, also causing passenger discomfort. Therefore, lateral dynamics parameters can include: centripetal or lateral acceleration, and lateral jerk. Centripetal or lateral acceleration is used to indicate the vehicle's cornering behavior. Therefore, the comfort index ∝ |v² / R|, where R is the turning radius, v is the vehicle's current speed, and |v² / R| is the lateral acceleration. Thus, the comfort index ∝ lateral jerk.

[0114] Composite parameters are parameters other than longitudinal and lateral dynamic parameters involved in models and / or existing standards used to evaluate ride comfort. For example, the ISO 2631-1 standard considers the impact of vibrations at different frequencies on human comfort; therefore, composite parameters may also include vibration frequencies generated during vehicle movement. Optionally, multiple vibration frequencies generated during vehicle movement are assigned corresponding weights, and each vibration frequency is weighted and filtered to obtain a comfort index corresponding to the composite parameters.

[0115] The method for determining the current comfort index based on user feedback data is the same as the method for obtaining the target comfort index corresponding to historical feedback data in step S302, and will not be described in detail here.

[0116] Furthermore, after constructing training samples based on user feedback data and current driving parameters, online supervised learning and / or reinforcement learning methods can be used to learn the correspondence between the current driving parameters and the current comfort index from the training samples, and update the parameters in the comfort evaluation model. Supervised learning and / or reinforcement learning can be found in the descriptions of related technologies, and will not be elaborated upon here.

[0117] In this embodiment, the comfort evaluation model is dynamically adjusted and calibrated using the current driving parameters and user feedback data at the current moment, so that the prediction results of the comfort evaluation model are more in line with real human feelings and the accuracy of the comfort index output by the comfort evaluation model is improved.

[0118] In one embodiment, selecting a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item includes: determining a first weight for the comfort cost and a second weight for each preset cost item; weighting the comfort cost and the estimated cost according to the first weight and the second weight, and determining the weighted result as the comprehensive score of the candidate trajectory; and selecting the target trajectory from multiple candidate trajectories based on the comprehensive score of each candidate trajectory.

[0119] For example, a cost function is constructed based on comfort cost and its first weight, and each preset cost item and its second weight. The total cost of the cost function is used as a comprehensive score. The trajectory with the lowest comprehensive score is selected from multiple candidate trajectories as the target trajectory. The second weight includes efficiency weight and safety weight. The total cost is calculated as follows:

[0120] Total cost = w_safety × C_safety + w_efficiency × C_efficiency + w_comfort × C_discomfort.

[0121] Where C_efficiency is the efficiency cost, and w_efficiency is the efficiency weight. C_safety is the safety cost, and w_safety is the safety weight. C_discomfort is the comfort cost, and w_discomfort is the first weight. For example, C_discomfort can be the integral of the comfort index over the entire candidate trajectory, or it can be the maximum value of the comfort index over the entire candidate trajectory.

[0122] In this embodiment, a comfort cost based on a quantified comfort index is introduced. The comfort cost and the estimated cost are combined with a first weight and a second weight to obtain a comprehensive score. This allows the system to actively avoid trajectories that cause discomfort during the process of determining the target trajectory based on each candidate trajectory, thereby achieving human-centered adaptive planning and improving the user's comfort level during autonomous driving.

[0123] Furthermore, in one embodiment, determining a first weight for comfort costs and a second weight for each preset cost item includes: obtaining preset first and second weights; and adjusting the first and / or second weights based on the vehicle's current driving parameters and / or the user's current preferences.

[0124] The current driving status of the vehicle can be determined based on its driving parameters, and the corresponding first weight can be obtained based on the driving status.

[0125] Optionally, each preset cost item includes a safety cost item. If the vehicle's driving parameters indicate an emergency situation, the safety weight of the safety cost item is increased. For example, if the change in any of the vehicle's speed, acceleration, or jerk exceeds a specified threshold, and it is determined that another vehicle has suddenly cut in, the current vehicle's driving state is confirmed to be in an emergency, and the safety weight of the safety cost item is increased to a specified value, thereby generating an emergency collision avoidance trajectory.

[0126] Optionally, if the vehicle's driving parameters indicate that it is in a normal cruising state, the comfort weight of the comfort cost can be increased. For example, if the vehicle's speed is within the permitted speed range for highways, and the lane position is relatively stable with no frequent lane changes as determined by lateral acceleration, it can be determined that the vehicle is currently in a highway driving state. Combining longitudinal acceleration and jerk, the vehicle's speed change is determined to be smooth, indicating that the vehicle is in a stable following state. Increasing the comfort weight of the comfort cost to a specified value can achieve the effect of smoothing acceleration and jerk, thus improving comfort.

[0127] User preferences are determined based on at least one of the following: the vehicle driving mode determined in response to the current user instruction, historical user instructions used to determine the vehicle driving mode, and historical feedback data. Different vehicle driving modes correspond to different first weights and / or different second weights.

[0128] Optionally, when passengers board the vehicle, they can actively select the vehicle driving mode for their trip via the in-vehicle interface, such as "Smooth Mode," "Standard Mode," "Efficient Mode," "Sport Mode," or "Comfort Mode." Different first and / or second weights are configured for different driving modes. For example, the first weight in "Smooth Mode" is higher than the first weight in "Efficient Mode"; the second weights for "Smooth Mode" and "Efficient Mode" can be the same or different. The method for adjusting weights according to the driving mode is simple, computationally inefficient, and requires no complex perception mechanisms.

[0129] Optionally, user preferences can be statistically derived based on historical user commands and user feedback data used to determine vehicle driving modes. For example, if the frequency or number of times a historical user command indicates a certain vehicle driving mode exceeds a specified value, that vehicle driving mode is considered the user's preferred driving mode, and the preference is adjusted accordingly. Alternatively, if the comfort index corresponding to the user's historical feedback data is less than a specified threshold, and the proportion of historical feedback data with a comfort index less than the specified threshold exceeds a certain percentage threshold, it can be determined that the user prefers a driving mode with high comfort, and the first weight is increased accordingly.

[0130] In this embodiment, the weights of each cost item used to generate a comprehensive evaluation are adaptively adjusted based on the vehicle's current driving parameters and / or the user's current preferences, thereby achieving personalized autonomous driving adjustment and improving the comfort of autonomous driving.

[0131] In one embodiment, the comfort evaluation model includes multiple first sub-models corresponding to different users; each first sub-model is obtained based on user data of the same user. The first sub-model corresponding to the same user can be trained based on that user's user data; alternatively, a base model can be trained first based on user data from multiple users, and then updated based on the current user's user data to obtain the corresponding first sub-model. The user data includes feedback parameters and corresponding specified driving parameters.

[0132] Figure 4 A flowchart illustrating a fourth embodiment of an autonomous driving method is provided. Based on the above embodiments, the method determines the comfort index for each predicted time step using a pre-trained comfort evaluation model, according to specified driving parameters at each predicted time step. This includes:

[0133] Step S401: Find the first target sub-model that matches the current user from multiple first sub-models.

[0134] Optionally, user profiles are created for different users, including independent comfort evaluation model configuration files for each user. During autonomous driving, the current user can log in via Bluetooth or facial recognition on their mobile phone, obtain the corresponding user profile, and configure the comfort evaluation model (first objective sub-model) according to the configuration file in the user profile, thereby predicting the comfort index based on the first objective sub-model.

[0135] Step S402: Input the specified driving parameters for each predicted time step into the first target sub-model, so that the first target sub-model outputs the comfort index for that predicted time step. The first target sub-model outputs the corresponding comfort index in the same way as the comfort evaluation model; the specific implementation process can be found in the previous description and will not be repeated here.

[0136] In this embodiment, different comfort evaluation models are obtained based on different user data, which allows multiple comfort evaluation models to gradually learn the preferences of different users; and based on user preferences, accurate prediction of the comfort index is achieved, thereby improving the user's riding experience.

[0137] In one embodiment, the comfort evaluation model includes multiple second sub-models corresponding to different scenarios, each second sub-model being obtained based on user data in the same scenario; wherein, each second sub-model is trained based on the same user data in the same scenario, or it can be trained based on different user data in the same scenario. Figure 5 A flowchart of a fifth embodiment of an autonomous driving method is provided. Based on the above embodiments, the comfort index for each predicted time step is determined using a pre-trained comfort evaluation model according to the specified driving parameters at each predicted time step, including:

[0138] Step S501: Determine the current situation based on the vehicle's current driving parameters and / or user feedback data at the current moment.

[0139] The current driving state of the vehicle can be obtained from the vehicle's current driving parameters, and the current situation can be determined.

[0140] For example, based on the vehicle's current driving parameters, such as determining that the current time is a weekday based on the time information collected by the vehicle terminal, determining that the vehicle exhibits characteristics of low-speed driving, brief stops and smooth starts based on driving speed, longitudinal acceleration and jerk, and determining that the driving road is a commuter road such as an urban arterial road or expressway based on road environment data, it is determined that the current vehicle is in a commuter driving state on a weekday morning, and the driving scenario is determined to be commuting.

[0141] For example, based on the vehicle's current driving parameters, such as determining that it is currently a weekend period based on the time information collected by the vehicle terminal, and combining the driving speed, longitudinal acceleration and jerk data, it is determined that the vehicle is driving steadily at a medium-low speed; at the same time, based on the road environment data, it is determined that the driving section is a non-commuting leisure section such as suburban or scenic roads, and the current vehicle is determined to be in a weekend outing driving state, and the driving scenario is determined to be a leisure trip.

[0142] The current context can be determined based on the user status obtained from the user feedback data at the current moment.

[0143] For example, the user's state can be determined based on feedback data collected by in-cabin sensors. For instance, based on images containing the user's face and posture, it can be determined that the user is in a state of rest. After determining that the user's physiological rhythm characteristics conform to a resting state based on physiological signals such as heart rate, respiratory rate, and electromyography, it can be determined that the user is in a state of rest with their eyes closed, and the current situation can be determined as rest.

[0144] Step S502: Find the second target sub-model that matches the current situation from multiple second sub-models.

[0145] Optionally, from multiple second sub-models, a model trained on user data in the current context is selected and used as the second target sub-model. For example, after determining the driving scenario as commuting, a second sub-model trained on user data during commuting can be obtained; since users prioritize efficiency and have a higher tolerance for minor discomfort during commuting, the first weight of comfort cost is relatively low in this second sub-model. After determining the driving scenario as leisure, a second sub-model trained on user data during leisure trips is obtained; in this second sub-model, users value comfort more, and the first weight of comfort cost is relatively high. For example, a second sub-model corresponding to the rest scenario is obtained; this second sub-model pre-configures a high first weight for comfort cost to prioritize ensuring cabin comfort needs during user rest scenarios.

[0146] Step S503: Input the driving parameters for each predicted time step into the second target sub-model, so that the second target sub-model outputs the comfort index for that predicted time step. The second target sub-model outputs the corresponding comfort index in the same way as the comfort evaluation model; the specific implementation process can be found in the previous description and will not be repeated here.

[0147] In this embodiment, the current situation is determined based on the vehicle's current driving parameters and / or user feedback data at the current moment, and the corresponding second sub-model is adaptively selected based on the situation. By switching the second sub-model, an accurate comfort index that meets the user's needs is output, so that automated and personalized autonomous driving adjustment can be achieved based on the comfort index.

[0148] In one embodiment, while controlling the vehicle to travel along a target trajectory, if the comfort index corresponding to the collected specified feedback data is less than a specified threshold, the driving parameters of the target trajectory are adjusted according to a preset optimization rule to improve the ride comfort of the vehicle.

[0149] The designated feedback data consists of at least one type of feedback data that is pre-screened from multiple types and is considered the most accurate and crucial for determining passenger comfort. The designated threshold is the minimum critical criterion for identifying whether the designated feedback data indicates that the user is in a state of high discomfort. If the comfort index is below the designated threshold, the user who generated the designated feedback data is in a state of high discomfort.

[0150] Optionally, the specified feedback data is a user image detected by a camera; the user is identified based on the image, and the comfort index corresponding to the specified feedback data is determined to be less than a specified threshold based on the identification result. For example, if the identification result shows that the user continuously closes their eyes (the frequency of eye closure is greater than a specified value) and tilts their head back within a specified time, it can be determined that the comfort index corresponding to the specified feedback data is less than the specified threshold, and the user may experience motion sickness or discomfort.

[0151] Optionally, the specified feedback data is user speech detected by a microphone. Speech recognition is performed on the user speech, and based on the recognition result, it is determined that the comfort index corresponding to the specified feedback data is less than a specified threshold. For example, if the duration and / or frequency of the user's output of preset keywords is greater than a corresponding threshold, it is determined that the comfort index corresponding to the specified feedback data is less than the specified threshold. The preset keywords can be specified complaint-like language or other words; there are no restrictions here.

[0152] Preset optimization rules are used to improve passenger comfort, and at least one driving parameter can be adjusted based on these rules. Optionally, the current cruising speed can be reduced by 10% and the following distance increased by 50% using preset optimization rules; alternatively, preset optimization rules can also adjust driving parameters such as cruising speed and following distance to specified values. Furthermore, preset optimization rules can also adjust the vehicle's driving mode, such as adjusting the vehicle's driving mode to a preset "safety mode" for a specified period of time, so that the vehicle operates based on the driving parameters under this driving model.

[0153] In this embodiment, adjusting the target trajectory directly based on feedback data can improve the response speed of comfort adjustment, and it is also highly robust and has a low implementation cost.

[0154] In one embodiment, after adjusting the comfort evaluation model based on feedback data and corresponding driving parameters, the adjusted comfort evaluation model can be applied in the next trajectory planning cycle to determine a new target trajectory. Alternatively, after adjusting the comfort evaluation model based on feedback data and corresponding driving parameters, the adjusted comfort evaluation model can be validated first: deploy the adjusted comfort evaluation model in some vehicles or some trips, and compare the adjusted comfort evaluation model with the unadjusted comfort evaluation model. By statistically analyzing the comfort index corresponding to the feedback data, the effectiveness of the adjusted comfort evaluation model in improving the comfort index can be evaluated, thereby determining whether the ride comfort has improved based on the comfort index. Specifically, after the model is updated, it is determined whether the frequency of receiving feedback data with a comfort index lower than a specified value in similar situations has decreased. If so, the optimization is effective; if not, the optimization is ineffective; thus verifying the effectiveness of the optimization. If the adjusted comfort evaluation model is superior to the unadjusted comfort evaluation model, the adjusted comfort evaluation model is deployed.

[0155] In one embodiment, such as Figure 6 As shown, an autonomous driving device 600 is provided, comprising:

[0156] The driving simulation module 601 is used to predict the sequence of driving parameters of the vehicle when executing each of the multiple candidate trajectories generated so far; wherein the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step.

[0157] The prediction module 602 is used to determine the comfort index for each prediction time step based on the specified driving parameters at each prediction time step using a pre-trained comfort evaluation model.

[0158] The cost estimation module 603 is used to determine the comfort cost of the candidate trajectory based on the comfort index at each predicted time step, and to determine the estimated cost of the candidate trajectory under each preset cost item.

[0159] The trajectory selection module 604 is used to select a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item; wherein, the target trajectory is a trajectory that takes into account both comfort cost and estimated cost.

[0160] The control module 605 is used to control the vehicle to travel along the target trajectory.

[0161] In some embodiments, the autonomous driving device 600 further includes a model training module for training a pre-trained comfort evaluation model. The model training module is used to: acquire a set of historical driving parameters and historical feedback data for each historical driving time; the historical feedback data includes user state parameters and / or the comfort category of user feedback; determine a target comfort index for each historical driving time based on the historical feedback data for each historical driving time; wherein the target comfort index characterizes the user's ride comfort level; select specified driving parameters related to ride comfort level from the set of historical driving parameters and historical feedback data for each historical driving time, and construct an initial model based on the specified driving parameters; wherein the initial model characterizes the mapping relationship between the specified driving parameters and the comfort index; establish training samples based on the specified driving parameters and the target comfort index for each historical driving time, and train the initial model using the training samples to obtain the pre-trained comfort evaluation model.

[0162] In some embodiments, the model training module is further configured to: during the process of controlling the vehicle to travel along the target trajectory, when a specified condition is met at the current moment, acquire the current driving parameters and the user feedback data at the current moment; wherein, the current driving parameters include at least the specified driving parameters; the user feedback data includes the comfort category or the user's state parameters; determine the current comfort index at the current moment based on the user feedback data, and construct a correspondence between the current driving parameters and the current comfort index; and update the comfort evaluation model using the correspondence.

[0163] In some embodiments, the trajectory selection module 604 selects a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item, including: determining a first weight for the comfort cost and a second weight for each preset cost item; performing weighted processing on the comfort cost and the estimated cost according to the first weight and the second weight, and determining the weighted processing result as the comprehensive score of the candidate trajectory; and selecting the target trajectory from multiple candidate trajectories based on the comprehensive score of each candidate trajectory.

[0164] Optionally, the trajectory selection module 604 determines a first weight for comfort costs and a second weight for each preset cost item, including: obtaining preset first and second weights; and adjusting the first and / or second weights based on the vehicle's current driving parameters and / or the user's current preferences.

[0165] In some embodiments, the comfort evaluation model includes multiple first sub-models corresponding to different users; each first sub-model is obtained based on user data of the same user; the prediction module 602 determines the comfort index for each prediction time step using a pre-trained comfort evaluation model based on specified driving parameters for each prediction time step, including: searching for a first target sub-model matching the current user from multiple first sub-models; inputting the specified driving parameters for each prediction time step into the first target sub-model so that the first target sub-model outputs the comfort index for the prediction time step.

[0166] In some embodiments, the comfort evaluation model includes multiple second sub-models corresponding to different scenarios, each second sub-model being derived from user data under the same scenario; the prediction module 602 determines the comfort index for each prediction time step using a pre-trained comfort evaluation model based on specified driving parameters for each prediction time step, including: determining the current scenario based on the vehicle's current driving parameters and / or user feedback data at the current moment; searching for a second target sub-model that matches the current scenario from among the multiple second sub-models; and inputting the driving parameters for each prediction time step into the second target sub-model so that the second target sub-model outputs the comfort index for that prediction time step.

[0167] In some embodiments, the control module 605 is further configured to, during the process of controlling the vehicle to travel along the target trajectory, adjust the driving parameters of the target trajectory according to a preset optimization rule when the comfort index corresponding to the collected specified feedback data is less than a specified threshold, so as to improve the ride comfort of the vehicle.

[0168] The modules in the aforementioned autonomous driving device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0169] In one embodiment, Figure 7 A hardware internal structure diagram of a vehicle is provided as shown. The vehicle includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The vehicle's processor provides computational and control capabilities. The vehicle's memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The vehicle's database stores data such as comfort evaluation models, candidate trajectories, and target trajectories. The vehicle's I / O interfaces are used for information exchange between the processor and external devices. The vehicle's communication interface is used for communication with external terminals via a network connection. When executed by the processor, the vehicle implements an autonomous driving method.

[0170] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the vehicle to which the present application is applied. The specific vehicle hardware may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0171] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0172] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0173] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0174] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0175] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0176] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0177] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An autonomous driving method, characterized in that, The method includes: For each of the multiple candidate trajectories generated so far, predict the sequence of driving parameters for the vehicle when executing that candidate trajectory; wherein, the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step; Based on the specified driving parameters at each predicted time step, the comfort index at that predicted time step is determined using a pre-trained comfort evaluation model. Based on the comfort index at each predicted time step, determine the comfort cost of the candidate trajectory, and determine the estimated cost of the candidate trajectory under each preset cost item. Based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item, a target trajectory is selected from the multiple candidate trajectories; wherein, the target trajectory is a trajectory that simultaneously considers the comfort cost and the estimated cost; Control the vehicle to travel along the target trajectory.

2. The method according to claim 1, characterized in that, The training process of the pre-trained comfort evaluation model includes: Acquire the set of historical driving parameters and historical feedback data at historical driving times; the historical feedback data includes user status parameters and / or user feedback comfort categories; The target comfort index for each historical moment is determined based on historical feedback data from each historical driving period; wherein, the target comfort index is used to characterize the user's riding comfort level; Based on the historical driving parameter set and historical feedback data at each historical driving time, specified driving parameters related to ride comfort are selected from the historical driving parameter set, and an initial model is constructed based on the specified driving parameters; wherein, the initial model is used to characterize the mapping relationship between the specified driving parameters and the comfort index; Training samples are established based on the specified driving parameters and target comfort index at each historical time, and the initial model is trained using the training samples to obtain the pre-trained comfort evaluation model.

3. The method according to claim 1 or 2, characterized in that, The method further includes: During the process of controlling the vehicle to travel along the target trajectory, when the specified conditions are met at the current moment, the current driving parameters and the user feedback data at the current moment are obtained; wherein, the current driving parameters include at least the specified driving parameters; the user feedback data includes the comfort category or user status parameters. Based on the user feedback data, determine the current comfort index at the current moment, and construct the correspondence between the current driving parameters and the current comfort index. The comfort evaluation model is updated using the correspondence.

4. The method according to claim 1, characterized in that, The step of selecting a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under each preset cost item includes: Determine the first weight of the comfort cost and the second weight of each preset cost item; The comfort cost and the estimated cost are weighted according to the first weight and the second weight, and the weighted result is determined as the comprehensive score of the candidate trajectory. The target trajectory is selected from the multiple candidate trajectories based on the comprehensive score of each candidate trajectory.

5. The method according to claim 4, characterized in that, The determination of the first weight of the comfort cost and the second weight of each preset cost item includes: Obtain the preset first and second weights; The first weight and / or the second weight are adjusted based on the vehicle's current driving parameters and / or the user's current preferences.

6. The method according to claim 1, characterized in that, The comfort evaluation model includes multiple first sub-models corresponding to different users; each first sub-model is obtained based on user data of the same user; the step of determining the comfort index for each predicted time step using the pre-trained comfort evaluation model based on specified driving parameters for each predicted time step includes: Find the first target sub-model that matches the current user from among the plurality of first sub-models; The specified driving parameters for each predicted time step are input into the first target sub-model, so that the first target sub-model outputs the comfort index for that predicted time step.

7. The method according to claim 1, characterized in that, The comfort evaluation model includes multiple second sub-models corresponding to different scenarios, each second sub-model being derived from user data within the same scenario; the step of determining the comfort index for each predicted time step using the pre-trained comfort evaluation model based on specified driving parameters at that predicted time step includes: The current situation is determined based on the vehicle's current driving parameters and / or user feedback data at the current moment; Find the second target sub-model that matches the current situation from among the plurality of second sub-models; The driving parameters at each predicted time step are input into the second target sub-model, so that the second target sub-model outputs the comfort index at that predicted time step.

8. The method according to claim 1, characterized in that, The method further includes: During the process of controlling the vehicle to travel along the target trajectory, if the comfort index corresponding to the collected specified feedback data is less than a specified threshold, the driving parameters of the target trajectory are adjusted according to preset optimization rules to improve the ride comfort of the vehicle.

9. An automatic driving device, characterized in that, include: The driving simulation module is used to predict the sequence of driving parameters of a vehicle when executing a candidate trajectory among multiple candidate trajectories generated at the moment; wherein, the sequence of driving parameters includes the specified driving parameters of the vehicle at each prediction time step; The prediction module is used to determine the comfort index for each prediction time step based on the specified driving parameters at each prediction time step using a pre-trained comfort evaluation model. The cost estimation module is used to determine the comfort cost of the candidate trajectory based on the comfort index at each predicted time step, and to determine the estimated cost of the candidate trajectory under each preset cost item. The trajectory selection module is used to select a target trajectory from multiple candidate trajectories based on the comfort cost of the candidate trajectory and the estimated cost of the candidate trajectory under various preset cost items; wherein, the target trajectory is a trajectory that simultaneously considers the comfort cost and the estimated cost; The control module is used to control the vehicle to travel along the target trajectory.

10. A vehicle having a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the steps of the method according to any one of claims 1 to 8.

Images