Driving risk cognitive map construction method and system based on free energy

Abstract

The invention discloses a driving risk cognitive map construction method and system based on free energy, and the method comprises the steps: inputting a starting position and a target position of an intelligent driving vehicle into a trained risk entropy calculation model, so as to output a risk cognitive map of the intelligent driving vehicle; the training process of the risk entropy calculation model is as follows: acquiring environment state information of a vehicle to construct a training data set; constructing a perception and prediction model based on a free energy theory by using the environmental state information, wherein the free energy is the difference between the prediction error of the vehicle to the environmental state and a real perception value; information entropy is introduced into the perception and prediction model to define driving risk entropy, the risk degree of the vehicle in the environment is represented, and therefore the free energy of the current state of the vehicle is calculated; a driving risk entropy is calculated in a continuous time sequence manner, a high-risk area in the risk cognition map is dynamically updated in combination with a historical driving track of the vehicle, and the updated risk cognition map is provided for a decision module of the vehicle; constructing a loss function to adjust trainable parameters in the risk entropy calculation model; according to the risk cognition map construction method and system, dynamic risk assessment of the intelligent driving vehicle in a complex driving environment is effectively realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent driving vehicles, and particularly to a method and system for constructing a driving risk perception map based on free energy. Background Art

[0002] With the rapid development of autonomous driving technology, it has become a key challenge for intelligent vehicles to drive safely and reliably in complex traffic environments. Current autonomous driving systems rely on a variety of sensors, such as cameras, radars, and lidars, etc., to obtain real-time environmental data and complete autonomous driving tasks through path planning and decision-making algorithms. However, the traffic environment is full of uncertainties and complexities, including weather changes, different road conditions, and random behaviors of various traffic participants, etc. Facing these uncertainties, how to achieve effective perception and decision-making of driving risks has become an important research direction to ensure the safety and stability of autonomous driving systems.

[0003] One of the core tasks of an autonomous driving system is to perceive the state of the external environment in real time and predict potential risks that may occur in the future. To achieve this, the system must effectively process multi-source sensor data and conduct accurate risk assessments based on this data. This involves multiple factors such as the behavior prediction of traffic participants, the monitoring of road conditions, and the changes in dynamic environments. Currently, most autonomous driving systems rely on traditional path planning and obstacle avoidance algorithms, such as the A* algorithm, Dijkstra algorithm, RRT (Rapidly-Exploring Random Trees), etc. These algorithms mainly optimize paths based on the current position of the vehicle, the target position, and the distribution of obstacles in the static environment. However, their performance in the face of complex dynamic environments is not satisfactory. Especially for sudden situations or risks that cannot be predicted in advance, traditional planning algorithms often show slow responses or even failures. For example, when driving on a highway, if the vehicle in front suddenly brakes or changes lanes, the autonomous driving system needs to judge whether there is a collision risk within a very short time and take corresponding avoidance measures. In the urban driving environment, the uncertainties of the behaviors of pedestrians, non-motor vehicles, etc. also make it difficult for the vehicle to accurately predict risks during the decision-making process. In these scenarios, traditional algorithms based on static path planning often cannot cope with complex and changeable traffic conditions. Therefore, a new method is needed to quantify and evaluate the uncertainties and potential risks existing during driving, so as to provide more intelligent risk perception and decision-making support for autonomous driving systems.

[0004] Entropy is an important concept in information theory, used to measure the randomness or uncertainty of a system. In the context of autonomous driving, entropy can be used to quantify the uncertainty and potential risks in the environment. For example, on urban roads with heavy traffic, factors such as pedestrians, non-motor vehicles, and traffic signals in the environment increase the difficulty for the system to predict the environmental state. These complex traffic states can be characterized by the entropy value. The larger the entropy value, the weaker the system's control over the environment and the greater the potential risks.

[0005] Introducing entropy into the quantification model of driving risks can provide a tool for measuring risks for the autonomous driving system. By calculating the risk entropy of the vehicle in different driving scenarios, the autonomous driving system can dynamically adjust its driving strategy. For example, during high-speed driving, if the system detects a sudden increase in the entropy of the surrounding environmental information, it means that the environmental uncertainty has increased, and the system can decelerate in advance or plan a new path to avoid potential risks. Summary of the Invention

[0006] Based on the technical problems existing in the background art, the present invention proposes a method and system for constructing a driving risk cognitive map based on free energy, effectively realizing the dynamic risk assessment of intelligent driving vehicles in complex driving environments.

[0007] A method for constructing a driving risk cognitive map based on free energy proposed by the present invention inputs the starting position and target position of an intelligent driving vehicle into a trained risk entropy calculation model to output a risk cognitive map of the intelligent driving vehicle;

[0008] The training process of the risk entropy calculation model is as follows:

[0009] Obtain the environmental state information of the vehicle to construct a training data set;

[0010] Use the environmental state information to construct a perception and prediction model based on the free energy theory, where the free energy is the difference between the vehicle's prediction error of the environmental state and the true perception value;

[0011] Introduce entropy into the perception and prediction model to define the driving risk entropy, indicating the risk level of the vehicle in the environment, and thus calculate the free energy of the vehicle's current state;

[0012] By continuously calculating the driving risk entropy in chronological order, combined with the vehicle's historical driving trajectory, dynamically update the high-risk areas in the risk cognitive map, and provide the updated risk cognitive map to the vehicle's decision-making module;

[0013] Construct a loss function to adjust the trainable parameters in the risk entropy calculation model.

[0014] Further, obtaining the vehicle environment state information specifically includes: collecting environmental data including traffic participants, road conditions, obstacles, weather information, vehicle speed, acceleration, and position information through multiple sensor modules of the vehicle.

[0015] Further, the perception and prediction model includes a perception model and a prediction model;

[0016] The perception model of the free energy for the environmental state is expressed as follows:

[0017] F(s t ) = E(s t |m t ) + H(s t |s t-1 );

[0018] Among them, F(s t ) is the free energy of the vehicle at time t, E(s t |m t ) is the expectation of the state s t under the condition of the given prediction model m t , H(s t |s t-1 ) is the conditional entropy of the state s t under the condition of the given state s t-1 at the previous moment, reflecting the uncertainty of the state, s t is the state of the vehicle at time t, s t-1 is the state of the vehicle at time t - 1, and m t is the prediction model of the vehicle;

[0019] The vehicle perceives the environment through the perception model and makes decisions based on the prediction model.

[0020] Further, the calculation formula of the driving risk entropy S r is as follows:

[0021]

[0022] Among them, p(x i ) is the probability of the state x i appearing, x i is the state of the i-th risk factor, and N is the number of risk factors.

[0023] Further, calculate the free energy of the vehicle state based on the driving risk entropy, and the formula is as follows:

[0024] F(s t ) = -logp(s t |m t );

[0025] st = w × S r ;

[0026] where s t is the state of the vehicle at time t, m t is the prediction model of the vehicle, and w is the weight coefficient.

[0027] Furthermore, in the dynamic update of high-risk areas in the risk perception map, the vehicle approximately estimates the true state in the environment through a risk entropy calculation model:

[0028] F(s t ) = D KL [q(s t ) ∥ p(s h | m t )] + Ε q [log p(o | s t )];

[0029] where F(s t ) is the free energy of the vehicle at time t, D KL is the divergence function, used to measure the difference between the predicted distribution q(s t ) and the true distribution p(s t | m t ), m t is the prediction model of the vehicle; Ε q [logo(o | s t )] is the expected log-likelihood of the observed data o, reflecting the observation result based on the current state s t .

[0030] Furthermore, in the continuous sequential calculation of driving risk entropy, specifically:

[0031] In a dynamic driving environment, the vehicle calculates the real-time driving risk entropy by continuously updating the probability distribution of the state:

[0032]

[0033] where p(s t | o t ) is the updated probability of the vehicle state s t based on the current observation o t ; p(o t | s t ) is the likelihood estimate of the current observation o t given the state s t ; p(s t-1 ) is the prior probability of the state at the previous moment, and s t-1 is the state of the vehicle at time t - 1.

[0034] A free-energy-based driving risk perception map construction system inputs the starting position and target position of an intelligent driving vehicle into a trained risk entropy calculation model to output the risk perception map of the intelligent driving vehicle.

[0035] The training process of the risk entropy calculation model includes a data construction module, a perception and prediction model construction module, a risk entropy introduction module, a dynamic update module, and a loss construction module:

[0036] The data construction module is used to obtain the environmental state information of the vehicle to construct a training data set.

[0037] The perception and prediction model construction module is used to construct a perception and prediction model based on the free energy theory by using the environmental state information. The free energy is the difference between the prediction error of the vehicle for the environmental state and the true perception value.

[0038] The risk entropy introduction module is used to introduce the information entropy in the perception and prediction model to define the driving risk entropy, which represents the risk degree of the vehicle in the environment, so as to calculate the free energy of the vehicle's current state.

[0039] The dynamic update module is used to dynamically update the high-risk areas in the risk perception map by continuously calculating the driving risk entropy in chronological order, combining the vehicle's historical driving trajectory, and providing the updated risk perception map to the vehicle's decision-making module.

[0040] The loss construction module is used to construct a loss function to adjust the trainable parameters in the risk entropy calculation model.

[0041] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the driving risk perception map construction method as described above.

[0042] A computer-readable storage medium stores a number of classification programs, and the number of classification programs are used to be called by a processor and execute the driving risk perception map construction method as described above.

[0043] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk, or optical disc that can store program codes.

[0044] The advantages of a driving risk perception map construction method and system based on free energy provided by the present invention are as follows: By using the environmental data obtained by vehicle sensors, the driving risk entropy in the current environment can be calculated, and the free energy theory is used to adjust the prediction error of the system. At this time, the vehicle can adjust its driving strategy in real time according to the change of driving risk entropy, effectively improving the risk perception ability of the autonomous driving system and optimizing the safety and reliability of driving decisions. It is particularly suitable for the intelligent vehicle decision control system in complex traffic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of the present invention;

[0046] Figure 2 is a schematic diagram of the driving risk entropy calculation principle. This diagram shows how an intelligent driving vehicle evaluates risks by integrating the free energy theory and the calculation of information entropy, and makes real-time decisions based on this to ensure driving safety. The core of this mechanism is that by minimizing the prediction error and quantifying environmental uncertainty, the vehicle can adaptively adjust its driving strategy in complex and changing traffic environments.

[0047] Figure 3 is a block diagram of the driving risk entropy calculation and driving risk perception map construction for the intelligent driving vehicle in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] Next, the technical solutions of the present invention will be described in detail through specific embodiments. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0049] As Figures 1 to 3 shown, a driving risk perception map construction method based on free energy proposed by the present invention inputs the starting position and target position of an intelligent driving vehicle into a trained driving risk entropy calculation model to output a driving risk perception map of the intelligent driving vehicle;

[0050] The training process of the driving risk entropy calculation model is as follows:

[0051] Step 1: Obtain the environmental state information of the vehicle to construct a training data set;

[0052] Collect environmental data including traffic participants, road conditions, obstacles, weather information, vehicle speed, acceleration, position information, etc. through multiple sensor modules of the vehicle. The sensors include but are not limited to cameras, radars, lidars, ultrasonic sensors, and GPS modules.

[0053] Step 2: Construct a perception and prediction model based on the free energy theory using environmental state information, where the free energy is the difference between the vehicle's prediction error of the environmental state and the true perception value;

[0054] The vehicle predicts the spatio-temporal evolution of the behaviors of traffic participants and road conditions in the external environment through the perception and prediction model.

[0055] The perception and prediction model includes a perception model and a prediction model;

[0056] The perception model of the free energy for the environmental state is expressed as follows:

[0057] F(s t )=E(s t |m t )+H(s t |s t-1 );

[0058] Among them, F(s t ) is the free energy of the vehicle at time t, E(s t |m t ) is the expectation of state s t under the condition of the given prediction model m t , H(s t |s t-1 ) is the conditional entropy of state s t given the state s t-1 at the previous moment, reflecting the uncertainty of the state, s t is the state of the vehicle at time t, s t-1 is the state of the vehicle at time t - 1, and m t is the prediction model of the vehicle;

[0059] The vehicle perceives the environment through the perception model and makes decisions based on the prediction model. The free energy F is defined as:

[0060] F=E(q(s))-H(q(s));

[0061] Among them, E(q(s)) is the expected energy, representing the energy function of the current state of the intelligent driving vehicle, which is related to the perception and prediction errors; H(q(s)) represents the entropy, indicating the uncertainty of the intelligent driving vehicle about the environmental state s.

[0062] Step 3: Introduce information entropy into the perception and prediction model to define the driving risk entropy, representing the risk degree of the vehicle in the environment, so as to calculate the free energy of the current state of the vehicle;

[0063] The calculation formula of the driving risk entropy S r is as follows:

[0064]

[0065] where p(x i ) is the probability of the state x i appearing, x i is the state of the i-th risk factor, and N is the number of risk factors;

[0066] The free energy theory is used to achieve the prediction state of the trajectories of surrounding other vehicles and pedestrians by intelligent driving vehicles, and a steady state is achieved by reducing the prediction error p(s|m) of perception and the error of the current state s:

[0067] F(s t ) = -logp(s t |m t );

[0068] s t = w × S r ;

[0069] where s t is the state of the vehicle at time t, m t is the prediction model of the vehicle, and w is the weight coefficient.

[0070] This embodiment introduces a driving risk quantification model based on information entropy. By comprehensively analyzing various variables (such as other vehicles, pedestrians, weather, etc.) in the current environment, the risk entropy in the current driving scenario is calculated in real time. Information entropy is used to measure the perception ability of the system to the uncertainty of the external environment. The higher the entropy value, the greater the uncertainty of the environment. Through the dynamic evaluation of risk entropy, autonomous driving vehicles can adopt more conservative driving strategies in high-risk scenarios to ensure safety.

[0071] Step Four: Continuously calculate the driving risk entropy in chronological order, combine the historical driving trajectories of the vehicle, dynamically update the high-risk areas in the risk perception map, and provide the updated risk perception map to the decision-making module of the vehicle;

[0072] Use Kalman filtering or other prediction algorithms to predict and pre-label the possible risk areas in the future; since the vehicle does not know the true state in the environment, the vehicle can only approximately estimate it through perception and the prediction model:

[0073] F(s t ) = D KL [q(s t ) ∥ p(s t |m t )] + Ε q [logp(o|s t )];

[0074] where F(st ) is the free energy of the vehicle at time t, D KL is the divergence function, which is used to measure the difference between the predicted distribution q(s t ) and the true distribution p(s t |m t ), m t is the prediction model of the vehicle; Ε q [logp(o|s t )] is the expected log-likelihood of the observed data o, which reflects the observation result based on the current state s t .

[0075] In a dynamic driving environment, the vehicle needs to calculate the real-time driving risk entropy by continuously updating its probability distribution of the state. As Figure 2 shown, it shows how an intelligent driving vehicle evaluates risks by integrating the free energy theory and the calculation of information entropy, and makes real-time decisions accordingly to ensure driving safety. The core of this mechanism is that by minimizing the prediction error and quantifying the environmental uncertainty, the vehicle can adaptively adjust its driving strategy in a complex and changing traffic environment. By updating the probability distribution of the state to calculate the real-time driving risk entropy, the specific steps are as follows:

[0076]

[0077] Among them, p(s t |o t ) is the updated probability of the vehicle state s t based on the current observation o t ; p(o t |s t ) is the likelihood estimate of the current observation o t under the given state s t ; p(s t-1 ) is the prior probability of the state at the previous moment, and s t-1 is the state of the vehicle at time t-1.

[0078] In this embodiment, the map is dynamically updated, and high-risk areas are marked in the environment. By displaying the high-risk areas on the cognitive map, the system can avoid potential dangerous areas in advance, further improving the safe driving ability of the vehicle. The division standard of the high-risk areas can be adjusted according to actual applications, and this embodiment does not limit the location and threshold settings of the high-risk areas.

[0079] In addition, the uncertainty in the environment is quantified through information entropy calculation, and the risk entropy calculation model is continuously optimized in combination with the free energy theory, greatly improving the environmental perception ability of the autonomous driving system. The system can not only evaluate the current driving risk, but also predict the possible high-risk areas in the future through the construction of a risk perception map, thereby improving the accuracy and stability of overall decision-making.

[0080] Step Five: Construct a loss function to adjust the trainable parameters in the risk entropy calculation model.

[0081] In this embodiment, the driving risk entropy in the current environment can be calculated through the environmental data obtained by vehicle sensors, and the prediction error of the system is adjusted using the free energy theory. At this time, the vehicle can adjust its driving strategy in real time according to the change of the driving risk entropy to ensure the safe driving of the vehicle in a complex traffic environment.

[0082] In addition, for the risk entropy calculation and risk perception map construction proposed in this embodiment, the risk entropy can not only reflect the complexity of the current environment, but also predict future risks through temporal updates. For example, by analyzing the driving data over a period of time, the system can predict the possible high-risk areas in a certain area in the future and map this information to the perception map. The perception map is a dynamically updated risk map. By marking the high-risk areas on the map, the autonomous driving system can avoid entering these high-risk areas, thereby reducing the potential accident rate.

[0083] Therefore, in this embodiment, by collecting the environmental information of the vehicle during driving, a risk perception and prediction model is constructed using the free energy theory, and the uncertainty is quantified through the calculation of risk entropy. Based on the calculated driving risk entropy, the system can dynamically update the risk perception map and predict potential dangerous areas and driving trajectories. It effectively improves the risk perception ability of the autonomous driving system, optimizes the safety and reliability of driving decisions, and is particularly suitable for the intelligent vehicle decision control system in a complex traffic environment.

[0084] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its inventive concept, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A method for constructing a driving risk perception map based on free energy, which inputs the starting position and target position of an intelligent driving vehicle into a trained risk entropy calculation model to output a risk perception map of the intelligent driving vehicle; The training process of the risk entropy calculation model is as follows: Obtain the vehicle's environmental status information to construct a training data set; Using environmental state information to build a perception and prediction model based on free energy theory, the free energy is the difference between the vehicle's prediction error of the environmental state and the actual perception value; Information entropy is introduced into the perception and prediction model to define driving risk entropy, which represents the risk level of the vehicle in the environment, thereby calculating the free energy of the vehicle's current state; By continuously calculating the driving risk entropy in a time series and combining the historical driving trajectory of the vehicle, the high-risk areas in the risk perception map are dynamically updated, and the updated risk perception map is provided to the decision-making module of the vehicle; Construct a loss function to adjust the trainable parameters in the risk entropy calculation model.

2. The method for constructing a driving risk awareness map based on free energy according to claim 1, characterized in that: Acquiring vehicle environmental status information specifically includes: collecting environmental data including traffic participants, road conditions, obstacles, weather information, vehicle speed, acceleration, and location information through multiple sensor modules of the vehicle.

3. The method for constructing a driving risk awareness map based on free energy according to claim 1, characterized in that: The perception and prediction model includes the perception model and the prediction model; The free energy perception model of the environmental state is expressed as follows: F(s t )=E(s t |m t )+H(s t |s t-1 ); Among them, F(s t ) is the free energy of the vehicle at time t, E(s t |m t ) is the value of the given prediction model m t Under the condition of t Expectation, H(s t |s t-1 ) is state s t Given the state s at the previous moment t-1 The conditional entropy reflects the uncertainty of the state, s t is the state of the vehicle at time t, s t-1 is the state of the vehicle at time t-1, m t A prediction model for the vehicle; The vehicle perceives the environment through the perception model and makes decisions based on the prediction model.

4. The method for constructing a driving risk awareness map based on free energy according to claim 3 is characterized in that: Driving risk entropy S r The calculation formula is as follows: Among them, p(x i ) is the state x i The probability of occurrence, x i is the state of the ith risk factor, and N is the number of risk factors.

5. The method for constructing a driving risk awareness map based on free energy according to claim 4 is characterized in that: The free energy of the vehicle's state is calculated based on the driving risk entropy. The formula is as follows: F(s t )=-logp(s t |m t ); s t =w×S r ; Among them, s t is the state of the vehicle at time t, m t is the prediction model of the vehicle, and w is the weight coefficient.

6. The method for constructing a driving risk awareness map based on free energy according to claim 1, characterized in that: In the dynamic update of the high-risk areas in the risk perception map, the vehicle uses the risk entropy calculation model to approximate the real state of the environment: F(s t )=D KL [q(s t )∥p(s t |m t )]+Ε q [logp(o|s t )]; Among them, F(s t ) is the free energy of the vehicle at time t, D KL is a divergence function used to measure the predicted distribution q(s t ) and the true distribution p(s t |m t ), m t is the prediction model of the vehicle; q [logp(o|s t )] is the expected log-likelihood of the observed data o, reflecting the probability of t 's observation results.

7. The method for constructing a driving risk awareness map based on free energy according to claim 1, characterized in that: In the calculation of driving risk entropy in a continuous time series, specifically: In a dynamic driving environment, the vehicle calculates the real-time driving risk entropy by continuously updating the probability distribution of the state: Among them, p(s t |o t ) is based on the current observation o t Vehicle status s t The update probability of p(o t |s t ) is the current observation o t In a given state s t Likelihood estimate under t-1 ) is the prior probability of the state at the previous moment, s t-1 is the state of the vehicle at time t-1.

8. A free energy-based driving risk awareness map construction system, characterized in that: Inputting the starting position and target position of the intelligent driving vehicle into the trained risk entropy calculation model to output a risk perception map of the intelligent driving vehicle; The training process of the risk entropy calculation model includes data construction module, perception and prediction model construction module, risk entropy introduction module, dynamic update module and loss construction module: The data construction module is used to obtain the vehicle's environmental status information to construct a training data set; The perception and prediction model building module is used to build a perception and prediction model based on free energy theory using environmental state information, where the free energy is the difference between the vehicle's prediction error of the environmental state and the actual perception value; The risk entropy introduction module is used to introduce information entropy into the perception and prediction model to define driving risk entropy, which indicates the risk level of the vehicle in the environment, thereby calculating the free energy of the vehicle's current state; The dynamic update module is used to dynamically update the high-risk areas in the risk perception map by continuously calculating the driving risk entropy in a time series and combining the historical driving trajectory of the vehicle, and provide the updated risk perception map to the decision module of the vehicle; The loss building module is used to build a loss function to adjust the trainable parameters in the risk entropy calculation model.

9. A computer 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 method for constructing a driving risk awareness map according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of classification programs, and the plurality of classification programs are used to be called by a processor and execute the method for constructing a driving risk awareness map as described in any one of claims 1-7.

Images