An intelligent driving method and related equipment based on brain-like perception

Abstract

The present application is applicable to the field of intelligent vehicle research technology, and provides an intelligent driving method based on brain-like perception and related equipment. It includes obtaining a bird's-eye view and a measurement vector of the target vehicle respectively, and performing semantic division on the bird's-eye view to obtain a multi-channel grayscale image; collecting the stimulus response of the human brain to the multi-channel grayscale image; constructing a brain-like perception network and a brain-like decision network respectively; aligning the brain-like perception network and the brain-like decision network with the human brain according to the stimulus response; inputting the multi-channel grayscale image and the measurement vector into the aligned brain-like perception network to obtain the environmental characteristics of the target vehicle; inputting the environmental characteristics into the aligned brain-like decision network to obtain the driving decision of the target vehicle; the driving decision includes multiple driving actions of the target vehicle; calculating the evaluation score corresponding to each driving action respectively, and controlling the driving of the target vehicle according to the driving action corresponding to the highest evaluation score, so as to realize intelligent driving. The present application can improve the interpretability of the intelligent driving method.

Description

Technical Field

[0001] The present application belongs to the technical field of intelligent vehicle research, and in particular, relates to an intelligent driving method and related equipment based on brain-like perception. Background Art

[0002] Intelligent driving technology refers to the ability of smart cars to perceive the road environment through on-board sensor systems, automatically plan driving routes and control the vehicle to reach the predetermined destination. Its technical system generally includes three major parts: environmental perception, decision-making planning and vehicle control, involving multiple research fields such as computers, mathematics, mechanics, control science, and psychology.

[0003] The current intelligent car autonomous driving technology still has technical difficulties that need to be overcome. The technical challenges mainly include active safety perception and decision-making of multi-modal fusion under complex dynamic road conditions. Urban dynamic road conditions are characterized by high dynamics, non-stationarity, and strong uncertainty. Efficient and accurate perception of driving scenes often involves multi-sensor data on board. A stable and reliable intelligent driving system must accurately perceive different targets around the intelligent car (such as vehicles, pedestrians, etc.) in order to make reasonable decisions to avoid potential traffic risks and ensure the driving safety of the intelligent car. However, the "black box" model based on deep learning has the problem of insufficient interpretability, which greatly limits the credibility of various perception and decision-making methods and their promotion and application in actual engineering. Summary of the invention

[0004] The present application provides an intelligent driving method and related equipment based on brain-like perception, which can solve the problem of insufficient explainability of traditional intelligent driving methods.

[0005] In a first aspect, the present application provides an intelligent driving method based on brain-like perception, comprising:

[0006] A bird's-eye view of the target vehicle and a measurement vector for describing the state of the target vehicle are respectively obtained, and the bird's-eye view is semantically divided to obtain a multi-channel grayscale image; each channel in the multi-channel grayscale image includes at least one grayscale image, and the grayscale images corresponding to different channels are different from each other;

[0007] Collect the stimulus response of the human brain to multi-channel grayscale images;

[0008] A brain-like perception network for extracting environmental features of a target vehicle and a brain-like decision network for generating driving decisions of the target vehicle are respectively constructed; wherein the brain-like perception network includes a brain-like visual perception module for extracting static features of a perceived object, a brain-like motion perception module for extracting dynamic features of a perceived object, and a brain-like multimodal perception module for extracting other features of a perceived object; the brain-like decision network includes a decision generation module for generating driving decisions and a decision evaluation module for evaluating the generated driving decisions, and the driving decisions include multiple driving actions of the target vehicle;

[0009] According to the stimulus response, the brain-like perception network and the brain-like decision network are aligned with the human brain to obtain an aligned brain-like perception network and an aligned brain-like decision network;

[0010] The multi-channel grayscale image and the measurement vector are input into the aligned brain-like perception network to obtain the environmental characteristics of the target vehicle;

[0011] Input the environmental features into the aligned brain-like decision network to obtain the driving decision of the target vehicle; the driving decision includes multiple driving actions of the target vehicle;

[0012] The evaluation scores corresponding to each driving action are calculated respectively, and the driving action corresponding to the highest evaluation score is used to control the driving of the target vehicle to achieve intelligent driving.

[0013] Optionally, the measurement vectors include a steering wheel rotation vector, a throttle vector, a brake vector, a gear position vector, a lateral velocity vector, and a longitudinal velocity vector;

[0014] The multi-channel grayscale image includes a driving area grayscale image, a driving route grayscale image, a lane boundary grayscale image, a vehicle grayscale image, a pedestrian grayscale image, and a traffic light grayscale image.

[0015] Optionally, the brain-like visual perception module corresponds to the ventral pathway of the human brain, the brain-like motion perception module corresponds to the dorsal pathway of the human brain, and the brain-like multimodal perception module corresponds to the superior colliculus of the human brain;

[0016] The decision generation module corresponds to the orbitofrontal cortex and dorsal prefrontal cortex in the human brain, and the decision evaluation module corresponds to the ventral prefrontal cortex in the human brain.

[0017] Optionally, according to the stimulus response, the brain-like perception network and the brain-like decision network are aligned with the human brain to obtain an aligned brain-like perception network and an aligned brain-like decision network, including:

[0018] Step I, by calculating the formula

[0019]

[0020] Get the brain-like network output Yi Stimulus response X i The similarity r between them, i represents the index of the data point, and n represents the total number of samples; wherein the brain-like network is either a brain-like perception network or a brain-like decision network;

[0021] Step II, collect data on human brain activity, using functional magnetic resonance imaging neuroimaging technology. These data provide information about the activation of brain regions. Preprocess and analyze the data obtained from the human brain, including denoising, spatial alignment, identification of brain regions, etc., extract features from the human brain data to understand which brain regions are related to perception; collect behavioral data and corresponding brain activity data when the human brain participates in decision-making tasks, preprocess and analyze the human brain decision data to understand the neural basis of decision-making, and extract features from the human brain decision data to understand which brain regions are related to the decision-making process;

[0022] Step III, by calculating the formula

[0023]

[0024]

[0025]

[0026]

[0027]

[0028] Get the updated network parameters θ of the brain-like decision network k+1 , and get the updated brain-like decision network; where L ppo represents the gradient loss of the clipping policy, that is, the gradient loss of the adjustment policy using the generalized advantage estimate, corresponding to the gradient of the dopamine reward signal in the human brain, Represents the current brain-like decision network The generated driving decision τ, E represents the expected value corresponding to the driving decision, which is the weighted average of the driving decisions of different strategy networks, ω is the weight of the corresponding driving decision, and L ent represents the maximum entropy loss, which is used to describe the exploratory behavior of the reward system in the human brain, λ ent represents the weight parameter used to control the maximum entropy loss, H represents the entropy of the strategy, L exp represents the exploration loss, corresponding to the mechanism of regulating exploratory behavior in the human brain, λ exp represents the hyperparameter used to adjust the exploration loss weight, T represents the trajectory length corresponding to the driving decision, and N represents the last N used to control the exploration loss. z time step, (k) represents the last N driving decisions zThe return value at time step k within the time step, (k)∈{0,1}, P z represents the predefined exploration prior, i.e., a priori knowledge or prior preference for exploration behavior in reinforcement learning tasks, which is used to guide the behavior of the target vehicle. RL The semantic component vector representing the bird's-eye view, m RL represents the measurement vector;

[0029] Step IV, return to execute step II until the similarity is greater than a preset similarity threshold, and obtain the aligned brain-like perception network and the aligned brain-like decision network.

[0030] Optionally, the evaluation score is calculated as:

[0031] S=r1μ+r2λ

[0032] Where r1 and r2 represent activation similarity and output behavior similarity respectively, and μ and λ are the corresponding weights.

[0033] In a second aspect, the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned intelligent driving method when executing the computer program.

[0034] In a third aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned intelligent driving method is implemented.

[0035] The above solution of the present application has the following beneficial effects:

[0036] The intelligent driving method based on brain-like perception provided in the present application constructs a brain-like perception network for extracting the environmental characteristics of the target vehicle and a brain-like decision network for generating driving decisions for the target vehicle, and aligns the brain-like perception network and the brain-like decision network with the human brain based on the stimulus response of the human brain. This can make the subsequently generated driving strategy more in line with the thinking logic of the human brain, which is conducive to enhancing the interpretability of the intelligent driving method.

[0037] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 A flowchart of an intelligent driving method based on brain-like perception provided in one embodiment of the present application;

[0040] Figure 2 A multi-channel grayscale image provided by an embodiment of the present application;

[0041] Figure 3 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0042] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0043] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0044] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0045] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0046] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0047] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0048] In response to the problem of insufficient explainability of traditional intelligent driving methods, the present application provides an intelligent driving method based on brain-like perception. The method constructs a brain-like perception network for extracting the environmental characteristics of the target vehicle and a brain-like decision network for generating driving decisions for the target vehicle, and aligns the brain-like perception network and the brain-like decision network with the human brain based on the stimulus response of the human brain. This can make the subsequently generated driving strategy more in line with the thinking logic of the human brain, which is conducive to enhancing the explainability of the intelligent driving method.

[0049] The following is an exemplary description of the intelligent driving method based on brain-like perception provided in this application.

[0050] like Figure 1 As shown, the intelligent driving method based on brain-like perception includes the following steps:

[0051] Step 11, respectively obtain a bird's-eye view of the target vehicle and a measurement vector for describing the state of the target vehicle, and perform semantic segmentation on the bird's-eye view to obtain a multi-channel grayscale image.

[0052] Each channel in the multi-channel grayscale image includes at least one grayscale image, and the grayscale images corresponding to different channels are different.

[0053] In an embodiment of the present application, the bird's-eye view of the target vehicle may be collected by a camera pre-installed on the vehicle.

[0054] The measurement vectors include a steering wheel rotation vector, an accelerator vector, a brake vector, a gear position vector, a lateral velocity vector, and a longitudinal velocity vector.

[0055] The steering wheel rotation vector is used to describe the degree of steering wheel rotation and is used to indicate the steering direction and angle of the vehicle.

[0056] The throttle vector is used to describe the throttle opening, indicating the degree to which the vehicle accelerates or decelerates.

[0057] The brake vector is used to describe the degree to which the brake pedal is depressed and is used to control the vehicle's deceleration or stopping.

[0058] The gear vector is used to describe the gear status of the vehicle, such as forward gear, reverse gear, neutral gear, etc.

[0059] The lateral velocity vector is used to describe the vehicle's velocity in the lateral direction (horizontal direction) and is used to understand the lateral movement of the vehicle on the road.

[0060] The longitudinal velocity vector is used to describe the vehicle's velocity in the longitudinal direction (vertical direction) and is used to understand the vehicle's longitudinal motion on the road, including acceleration and deceleration.

[0061] For example, in one embodiment of the present application, the vehicle camera is turned on to obtain the raw data of the surrounding environment of the target vehicle, the image data after the raw data is normalized is output, the sensor data is converted to the vehicle coordinate system, and the data in the vehicle coordinate system is projected to the top view to form a bird's-eye view, and then the bird's-eye view is semantically segmented, and the size is adjusted (the size of each image is unified) and then converted into a grayscale image. The obtained multi-channel grayscale image can be specifically referred to Figure 2 .

[0062] The on-board camera collects the raw data array, and actually takes the first three color channels, as shown below: [119112108...5050255] [132134146...3838255] [172180194...108115255] [194204213...98106255] ...

[0067] The steps of BEV semantic segmentation are as follows:

[0068] Step 1: Get a BEV image dataset with semantic labels. Each pixel should be labeled with its semantic category, such as road, vehicle, pedestrian, etc.

[0069] Step 11, select a deep learning model suitable for the semantic segmentation task to preprocess the BEV image, including resizing, normalization, etc. Make sure that the image input to the model matches the input size and format expected by the model;

[0070] In step III, the semantic labels are encoded for comparison with the output of the model, and the deep learning model is trained using the prepared dataset.

[0071] In step IV, the trained model is used to perform inference on new BEV images to obtain BEV semantic segmentation results.

[0072] In an embodiment of the present application, the multi-channel grayscale image includes a driving area grayscale image, a driving route grayscale image, a lane boundary grayscale image, a vehicle grayscale image, a pedestrian grayscale image, and a traffic light grayscale image.

[0073] The driving area grayscale map is used to describe the area where the driving vehicle is located, which may include roads, sidewalks, intersections, etc., and is used for the model to perceive the entire driving environment.

[0074] The driving route grayscale image is used to describe the vehicle's current driving route, which may include lane center lines, navigation paths, etc., to guide the vehicle's driving direction.

[0075] The lane boundary grayscale image is used to describe the boundary of the lane where the driving vehicle is located, helping the vehicle to stay in the correct lane.

[0076] The vehicle grayscale image is used to describe the information of other surrounding vehicles, including location, speed, etc., to avoid collisions and ensure safe driving.

[0077] The pedestrian grayscale image is used to describe the information of surrounding pedestrians, including location, walking direction, etc., to avoid collisions with pedestrians.

[0078] Traffic light grayscale images are used to describe the status of traffic lights at intersections or roads and to control the movement and parking of vehicles.

[0079] Step 12, collecting the stimulus response of the human brain to the multi-channel grayscale image.

[0080] Among them, the brain-like visual perception module is responsible for processing visual data from cameras or sensors during vehicle driving, and identifying roads, vehicles, pedestrians, signs, etc.; the brain-like motion perception module perceives motion and action, including vehicle motion, dynamic behavior of other road users, etc., by processing time series data (such as video frame sequences); the brain-like multimodal perception module integrates information from different perception modalities, including vision, radar, lidar, ultrasound, etc., to provide more comprehensive environmental perception; the brain-like decision-making module is responsible for analyzing, processing, and judging the information transmitted by the perception module, and then making corresponding driving decisions.

[0081] Specifically, it includes steps 12.1 to 12.4:

[0082] Step 12.1, collecting raw functional magnetic resonance imaging data of subjects viewing multi-channel grayscale image stimulation.

[0083] Step 12.2, use neuroimaging data processing software to perform three-dimensional reconstruction, registration, head motion correction and other preprocessing on the original cortical activation data, including voxel interpolation, anatomical structure segmentation, etc., to register the subject's brain data with the standard or other subjects' brain data, and then further preprocess to remove noise, spatial smoothing, etc., to obtain the blood oxygen level-dependent response signal of the visual brain area of ​​interest that changes over time when the subject views different image stimuli, which reflects the brain's neural activity to the image stimulus

[0084] Step 12.3, using a deconvolution method to obtain the amplitude of the blood oxygen level dependent response signal, and using it as the response value of the brain visual cortex voxel to the multi-channel grayscale image stimulation.

[0085] Step 12.4, obtain the response vector of the brain visual cortex voxels to each image stimulus, thereby forming the final functional magnetic resonance imaging (fMRI) data set. The correspondence between fMRI and stimulus response is based on measuring the changes in blood oxygen levels in the brain under different stimulus conditions. By analyzing these changes, we can understand the functional response of the brain to different stimuli.

[0086] Step 13, respectively constructing a brain-like perception network for extracting environmental features of the target vehicle and a brain-like decision network for generating driving decisions of the target vehicle.

[0087] Among them, the brain-like perception network includes a brain-like visual perception module for extracting static features of perceived objects, a brain-like motion perception module for extracting dynamic features of perceived objects, and a brain-like multimodal perception module for extracting other features of perceived objects. The brain-like decision-making network includes a decision generation module for generating driving decisions and a decision evaluation module for evaluating the generated driving decisions. The driving decisions include multiple driving actions of the target vehicle.

[0088] In the embodiments of the present application, the brain-like visual perception module corresponds to the ventral pathway of the human brain. The ventral pathway of the human brain is located in the V4 area of ​​the human cerebral cortex. This area is responsible for processing and interpreting color information in visual scenes in the brain. It has different responses to light of different wavelengths, which helps the human body perceive the color and color contrast of objects; it is also involved in the perception of the shape of objects, and its selectivity for specific shapes and edges helps the human body recognize and identify objects; the neurons in the area are selective for specific visual features and are more sensitive to certain types of stimuli, which helps the human body focus on and process specific information in a complex visual environment.

[0089] The brain-like motion perception module corresponds to the dorsal pathway in the human brain. The dorsal pathway of the brain is particularly good at spatial positioning and motion perception when processing visual information. It is responsible for tracking the movement, speed and direction of objects in the field of vision. It participates in the human body's spatial perception and guidance of objects in the environment. It enables the human body to locate, navigate and perform specific actions in space. It is highly sensitive to fast-moving and changing visual information, and can quickly capture dynamic changes in the surrounding environment, allowing the human body to respond in a timely manner.

[0090] The brain-like multimodal perception module corresponds to the superior colliculus in the human brain. The superior colliculus is a structure located in the brainstem that is responsible for processing visual and auditory information. It is the intersection of the visual and auditory pathways, allowing the information of these two perceptual modes to influence each other. The neuronal population of multimodal neurons located in the brain area is sensitive to information from different sensory pathways. These neurons can respond to visual and auditory stimuli at the same time, thereby achieving comprehensive processing of multiple sensory information.

[0091] The decision generation module corresponds to the orbitofrontal cortex and dorsal prefrontal cortex in the human brain. The orbitofrontal cortex is one of the prefrontal regions of the cerebral cortex, located in the front of the frontal lobe. The orbitofrontal cortex plays a key role in executive functions, including planning, decision-making, execution, and control of behavior. The orbitofrontal cortex is interconnected with the motor cortex and other executive systems to collaboratively complete complex motor and behavioral tasks, and is involved in monitoring behavioral results and environmental feedback, and then adjusting and correcting based on feedback information; the dorsal prefrontal cortex is part of the prefrontal lobe of the brain, located on the side of the prefrontal lobe. It is involved in planning, organizing, and executing complex cognitive tasks, including flexible thinking, problem solving, and decision-making, and helps evaluate and select behaviors based on previous behavioral results. Design a brain-like decision module based on the orbitofrontal cortex and dorsal prefrontal cortex to simulate the unique brain decision-making methods of the orbitofrontal cortex and dorsal prefrontal cortex. The ventral prefrontal cortex is part of the prefrontal lobe of the brain, located at the base of the prefrontal lobe. It is involved in the perception, expression, and regulation of emotions, helps integrate emotional experiences and regulate emotional responses, and is closely related to the processing of rewards and punishments. It helps individuals evaluate the rewards and punishments of different behaviors, thereby influencing decision-making and behavioral choices.

[0092] The decision-making evaluation module corresponds to the ventral prefrontal cortex in the human brain. The ventral prefrontal cortex is part of the prefrontal lobe of the brain, located at the base of the prefrontal lobe. It is involved in the perception, expression and regulation of emotions, helps integrate emotional experiences and regulate emotional responses, and is closely related to the processing of rewards and punishments. It helps individuals evaluate the rewards and punishments of different behaviors, thereby affecting decision-making and behavioral choices.

[0093] Step 14: align the brain-like perception network and the brain-like decision-making network with the human brain according to the stimulus response to obtain an aligned brain-like perception network and an aligned brain-like decision-making network.

[0094] Specifically, the following steps are included:

[0095] Step I, by calculating the formula

[0096]

[0097] Get the brain-like network output Y i Stimulus response X i The similarity r between them, i represents the index of the data point, and n represents the total number of samples; among them, the output of the brain-like perception network and the output of the brain-like decision-making network are linearly related to the stimulus response, and they are consistent in direction and change in the same direction with the change of the input. Therefore, the brain-like network is either the brain-like perception network or the brain-like decision-making network;

[0098] Step II, collect data on human brain activity, using neuroimaging techniques such as functional magnetic resonance imaging (fMRI), which provide information about the activation of brain regions, preprocess and analyze the data obtained from the human brain, including denoising, spatial alignment, identification of brain regions, etc., extract features from the human brain data to understand which brain regions are related to perception; collect behavioral data and corresponding brain activity data when the human brain participates in decision-making tasks, preprocess and analyze the human brain decision data to understand the neural basis of decision-making, and extract features from the human brain decision data to understand which brain regions are related to the decision-making process;

[0099] Step III, brain-like perception network In autonomous driving, the update strategy of the perception network is implemented through a reinforcement learning framework. The strategy includes defining state representation, determining action sets, designing reward mechanisms, establishing decision-making strategies, selecting appropriate learning algorithms, and setting network optimization goals. Through a balance between exploration and utilization, the perception network continuously iterates by observing the environment, performing actions, obtaining rewards, and updating network parameters to maximize the cumulative reward. This process is designed to enable the perception network to extract key features from sensor data, accurately classify objects in the environment, and improve the autonomous driving system's understanding of the surrounding environment and decision-making level. The human learning process is similar to the update process of the brain-like perception network, and it also perceives the external environment through this learning method.

[0100] The brain-like decision network uses the calculation formula

[0101]

[0102]

[0103]

[0104]

[0105]

[0106] Get the updated network parameters θ of the brain-like decision network k+1 , and get the updated brain-like decision network; where L ppo represents the gradient loss of the clipping policy, that is, the gradient loss of the adjustment policy using the generalized advantage estimate, corresponding to the gradient of the dopamine reward signal in the human brain, Represents the current brain-like decision network The generated driving decision τ, E represents the expected value corresponding to the driving decision, which is the weighted average of the driving decisions of different strategy networks, ω is the weight of the corresponding driving decision, and L ent represents the maximum entropy loss, which is used to describe the exploratory behavior of the reward system in the human brain, λ ent represents the weight parameter used to control the maximum entropy loss, H represents the entropy of the strategy, L exp represents the exploration loss, corresponding to the mechanism of regulating exploratory behavior in the human brain, λ exp represents the hyperparameter used to adjust the exploration loss weight, T represents the trajectory length corresponding to the driving decision, N represents the last N2 time steps used to control the exploration loss, (k) represents the return value at time step k within the last N2 time steps of the driving decision, (k)∈{0,1}, P z represents the predefined exploration prior, i.e., a priori knowledge or prior preference for exploration behavior in reinforcement learning tasks, which is used to guide the behavior of the target vehicle. RL The semantic component vector representing the bird's-eye view, m RL represents the measurement vector;

[0107] Step IV, return to execute step II until the similarity is greater than a preset similarity threshold, and obtain the aligned brain-like perception network and the aligned brain-like decision network.

[0108] Step 15: Input the multi-channel grayscale image and the measurement vector into the aligned brain-like perception network to obtain the environmental characteristics of the target vehicle. The details are as follows:

[0109] Y=OutputLayer(F merged (Merge(F image (X), F measurement (X))))

[0110] X is the input data, F image Extract features for multi-channel grayscale images, F measuremext Extract features F for the measurement vector mergedis the fused feature, and Y is the output target vehicle environment feature.

[0111] Step 16: Input the environmental features into the aligned brain-like decision network to obtain the driving decision of the target vehicle.

[0112] Specifically, the above driving decision includes multiple driving actions of the target vehicle. Exemplarily, in one embodiment of the present application, the action space corresponding to the brain-like decision network includes steering and acceleration, positive acceleration corresponds to throttle, and negative acceleration corresponds to braking. The driving action can be described by Beta distribution, where M is a random variable that obeys Beta distribution, and α and β are two shape parameters of the distribution, as follows:

[0113] B(a,β) is defined as

[0114]

[0115] Step 17, respectively calculate the evaluation score corresponding to each driving action, and control the driving of the target vehicle according to the driving action corresponding to the highest evaluation score to achieve intelligent driving.

[0116] In the embodiment of the present application, the calculation formula of the evaluation score is:

[0117] S=r1μ+r2λ

[0118] Where r1 and r2 represent activation similarity and output behavior similarity respectively, and μ and λ are the corresponding weights.

[0119] For example, a set of driving actions is obtained, such as: {(driving action at the next moment: braking, evaluation score: 87 points), (driving action at the next moment: accelerator, evaluation score: 55 points), (driving action at the next moment: steering, evaluation score: 66 points)}, then "braking" is selected as the driving action at the next moment to control the target vehicle to brake and decelerate.

[0120] It can be seen from the above steps that the intelligent driving method based on brain-like perception provided by the present application, by constructing a brain-like perception network for extracting the environmental characteristics of the target vehicle and a brain-like decision network for generating driving decisions of the target vehicle, and based on the stimulus response of the human brain, aligning the brain-like perception network and the brain-like decision network with the human brain, can make the subsequently generated driving strategy more in line with the thinking logic of the human brain, which is conducive to enhancing the interpretability of the intelligent driving method.

[0121] like Figure 3 As shown, an embodiment of the present application provides a terminal device, such as Figure 3 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0122] Specifically, when the processor D100 executes the computer program D102, a brain-like perception network for extracting the environmental features of the target vehicle and a brain-like decision network for generating the driving decision of the target vehicle are respectively constructed; according to the stimulus response, the brain-like perception network and the brain-like decision network are aligned with the human brain to obtain the aligned brain-like perception network and the aligned brain-like decision network; the multi-channel grayscale image and the measurement vector are input into the aligned brain-like perception network to obtain the environmental features of the target vehicle; the environmental features are input into the aligned brain-like decision network to obtain the driving decision of the target vehicle; the driving decision includes multiple driving actions of the target vehicle; the evaluation score corresponding to each driving action is calculated respectively, and the driving action corresponding to the highest evaluation score is controlled to realize intelligent driving. Among them, by constructing a brain-like perception network for extracting the environmental features of the target vehicle and a brain-like decision network for generating the driving decision of the target vehicle, and aligning the brain-like perception network and the brain-like decision network with the human brain based on the stimulus response of the human brain, the driving strategy generated subsequently can be more in line with the thinking logic of the human brain, which is conducive to enhancing the interpretability of the intelligent driving method.

[0123] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0124] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0125] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0126] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0127] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying the computer program code to the terminal device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0128] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0129] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0130] In the embodiments provided in the present application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0132] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An intelligent driving method based on brain-like perception, characterized in that: include: A bird's-eye view of the target vehicle and a measurement vector for describing the state of the target vehicle are respectively obtained, and the bird's-eye view is semantically divided to obtain a multi-channel grayscale image; each channel in the multi-channel grayscale image includes at least one grayscale image, and the grayscale images corresponding to different channels are different from each other; Obtaining stimulus responses of different users to the multi-channel grayscale image; A brain-like perception network for extracting environmental features of the target vehicle and a brain-like decision network for generating driving decisions of the target vehicle are respectively constructed; wherein the brain-like perception network includes a brain-like visual perception module for simulating neural pathways in the ventral pathway, a brain-like motion perception module for simulating neural pathways in the dorsal pathway of the brain, and a brain-like multimodal perception module for simulating neural pathways in the superior colliculus pathway; the brain-like decision network includes a decision generation module for generating driving decisions and a decision evaluation module for evaluating the generated driving decisions, wherein the driving decisions include multiple driving actions of the target vehicle; According to the stimulus response, aligning parameters of the brain-like perception network and the brain-like decision-making network to obtain an aligned brain-like perception network and an aligned brain-like decision-making network; Inputting the multi-channel grayscale image and the measurement vector into the aligned brain-like perception network to obtain the environmental characteristics of the target vehicle; Inputting the environmental features into the aligned brain-like decision network to obtain a driving decision of the target vehicle; the driving decision includes a plurality of driving actions of the target vehicle; Calculate the evaluation score corresponding to each driving action respectively, and control the driving of the target vehicle according to the driving action corresponding to the highest evaluation score to achieve intelligent driving; Wherein, obtaining stimulus responses of different users to the multi-channel grayscale image includes: Collecting raw functional magnetic resonance imaging data of subjects viewing multi-channel grayscale image stimulation; The original cortical activation data is preprocessed using neuroimaging data processing software, including three-dimensional reconstruction, registration, and head motion correction, including voxel interpolation and anatomical structure segmentation. The subject's brain data is registered with the standard or other subjects' brain data, and further preprocessed to remove noise and perform spatial smoothing, etc., to obtain the blood oxygen level-dependent response signal of the visual brain area of ​​interest that changes over time when the subject views different image stimuli, which reflects the brain's neural activity to the image stimuli; The amplitude of the blood oxygen level dependent response signal is obtained by deconvolution, and is used as the response value of the brain visual cortex voxel to the multi-channel grayscale image stimulation; Obtain the response vector of the brain's visual cortex voxels to each image stimulus, thereby forming the final functional MRI data set. The correspondence between functional MRI and stimulus response is based on measuring changes in the brain's blood oxygen level under different stimulus conditions. By analyzing these changes, we can understand the brain's functional response to different stimuli. The step of aligning the brain-like perception network and the brain-like decision network with the human brain according to the stimulus response to obtain an aligned brain-like perception network and an aligned brain-like decision network comprises: Step I, by calculating the formula Get the brain-like network output Y i Stimulus response X i The similarity r between them; wherein the brain-like network is any one of the brain-like perception network and the brain-like decision network; Step II, collect data on human brain activity, using functional magnetic resonance imaging neuroimaging technology. These data provide information about the activation of brain regions. Preprocess and analyze the data obtained from the human brain, including denoising, spatial alignment, and identification of brain regions. Extract features from the human brain data to understand which brain regions are related to perception; collect behavioral data and corresponding brain activity data when the human brain participates in decision-making tasks, preprocess and analyze the human brain decision data to understand the neural basis of decision-making, and extract features from the human brain decision data to understand which brain regions are related to the decision-making process; Step III, by calculating the formula Get the updated network parameters θ of the brain-like decision network k+1 , and get the updated brain-like decision network; where L ppo represents the gradient loss of the clipping policy, that is, the gradient loss of the adjustment policy using the generalized advantage estimate, corresponding to the gradient of the dopamine reward signal in the human brain, Represents the current brain-like decision network The generated driving decision τ, E represents the expected value corresponding to the driving decision, which is the weighted average of the driving decisions of different strategy networks, ω is the weight of the corresponding driving decision, and L ent represents the maximum entropy loss, which is used to describe the exploratory behavior of the reward system in the human brain, λ ent represents the weight parameter used to control the maximum entropy loss, H represents the entropy of the strategy, L exp represents the exploration loss, corresponding to the mechanism of regulating exploratory behavior in the human brain, λ exp represents the hyperparameter used to adjust the exploration loss weight, T represents the trajectory length corresponding to the driving decision, and N represents the last N used to control the exploration loss. z time step, (k) represents the last N driving decisions z The return value at time step k within the time step, (k)∈{0,1}, R z represents a predefined exploration prior, i.e., a priori knowledge or prior preference for exploration behavior in a reinforcement learning task, which is used to guide the behavior of the target vehicle. RL Represents the semantic component vector of the bird's-eye view, m RL represents the measurement vector; Step IV, returning to step II until the similarity is greater than a preset similarity threshold, to obtain an aligned brain-like perception network and an aligned brain-like decision network.

2. The intelligent driving method according to claim 1, characterized in that: The measurement vectors include a steering wheel rotation vector, a throttle vector, a brake vector, a gear position vector, a lateral velocity vector, and a longitudinal velocity vector; The multi-channel grayscale image includes a driving area grayscale image, a driving route grayscale image, a lane boundary grayscale image, a vehicle grayscale image, a pedestrian grayscale image, and a traffic light grayscale image.

3. The intelligent driving method according to claim 1, characterized in that: The brain-like visual perception module corresponds to the ventral pathway of the human brain, the brain-like motion perception module corresponds to the dorsal pathway of the human brain, and the brain-like multimodal perception module corresponds to the superior colliculus of the human brain; The decision generation module corresponds to the orbitofrontal cortex and dorsal prefrontal cortex in the human brain, and the decision evaluation module corresponds to the ventral prefrontal cortex in the human brain.

4. The intelligent driving method according to claim 1, characterized in that: The calculation formula of the evaluation score is: S=r1μ+r2λ Where r1 and г2 represent activation similarity and output behavior similarity respectively, and μ and λ are the corresponding weights.

5. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the intelligent driving method as described in any one of claims 1 to 4 is implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the intelligent driving method according to any one of claims 1 to 4 is implemented.

Images