Unstructured environment motion planning method, device and equipment

Abstract

The application discloses a non-structured environment motion planning method, device and equipment, relates to the fields of automation technology and vehicle engineering, and comprises the following steps: inputting a to-be-detected non-structured environment road surface terrain image into a neural network model to determine the probability of any terrain type to which each pixel in the to-be-detected non-structured environment road surface terrain image belongs; adopting a Bayesian regression model to determine the comprehensive passable risk under different road surfaces according to the probability of any terrain type to which each pixel belongs; adopting an entropy risk measurement method to convert the comprehensive passable risk under different road surfaces into a quantitative risk to generate a passability map; determining passability parameters of a cross-entropy-based motion planner according to the passability map; and utilizing the cross-entropy-based motion planner to perform motion planning on a vehicle in the to-be-detected non-structured environment according to the passability parameters, so that comprehensive probabilistic evaluation of terrain risk is realized to solve the uncertainty of terrain perception and the shortage of terrain adaptability.

Description

Technical Field

[0001] The present application relates to the fields of automation technology and vehicle engineering, and in particular to a method, device and apparatus for motion planning in an unstructured environment. Background Art

[0002] Motion control for unmanned ground vehicles (UGVs) faces significant challenges in unstructured environments, often characterized by complex terrain conditions such as slopes, sand, and gravel. The vehicle's dynamics are strongly influenced by terrain characteristics, potentially leading to slippage and rolling. However, existing motion planning methods struggle to effectively address the risks introduced by terrain parameter uncertainty, primarily in the following aspects:

[0003] (1) Uncertainty in terrain perception: Existing perception systems do not consider the probabilistic distribution characteristics of uncertainty risks under uniform terrain.

[0004] (2) Insufficient terrain adaptability: Traditional methods are unable to dynamically assess the risks of different terrains, making it difficult to achieve a balance between efficiency and safety in path planning. Summary of the Invention

[0005] The purpose of this application is to provide a method, device and equipment for motion planning in an unstructured environment, which can realize a comprehensive probabilistic assessment of terrain risks to address the uncertainty of terrain perception and the lack of terrain adaptability.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a method for motion planning in an unstructured environment, comprising:

[0008] An unstructured environment road terrain image to be tested is input into a neural network model to determine the probability that each pixel in the unstructured environment road terrain image belongs to any terrain type; the terrain types include sand, gravel, and grass; and the neural network model generates a terrain category probability map for each pixel.

[0009] A Bayesian regression model is used to determine the comprehensive drivability risk under different road conditions based on the probability of each pixel belonging to any terrain type.

[0010] The entropy risk measurement method is used to convert the comprehensive drivability risk under different road surfaces into quantitative risk and generate a drivability map.

[0011] Based on the feasibility map, feasibility parameters of a cross-entropy based motion planner are determined.

[0012] The vehicle in the to-be-tested unstructured environment is motion planned according to the passability parameter by using the cross-entropy-based motion planner.

[0013] In a second aspect, the present application provides an unstructured environment motion planning device, comprising:

[0014] A terrain type probability determination module inputs a to-be-tested unstructured environment road terrain image into a neural network model to determine a probability of each pixel in the to-be-tested unstructured environment road terrain image belonging to any terrain type; the terrain types include sand, gravel and grassland; and the neural network model generates a terrain category probability map for each pixel.

[0015] A comprehensive passability risk determination module determines a comprehensive passability risk under different road surfaces according to the probability of each pixel belonging to any terrain type by using a Bayesian regression model.

[0016] A passability map generation module converts the comprehensive passability risk under different road surfaces into a quantitative risk by using an entropy risk measurement method to generate a passability map.

[0017] A passability parameter determination module determines a passability parameter of a cross-entropy-based motion planner according to the passability map.

[0018] An unstructured environment vehicle motion planning module uses the cross-entropy-based motion planner to motion plan a vehicle in a to-be-tested unstructured environment according to the passability parameter.

[0019] In a third aspect, the present application provides a computer device, comprising a memory, a processor, a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the unstructured environment motion planning method in any one of the above aspects.

[0020] According to the specific embodiments provided in the present application, the present application has the following technical effects:

[0021] The present application provides a method, apparatus, and device for motion planning in an unstructured environment. These methods input a terrain image of a road surface in the unstructured environment to be tested into a neural network model to determine the probability that each pixel in the image belongs to any terrain type, including sand, gravel, and grass. The neural network model generates a terrain category probability map for each pixel. A Bayesian regression model is used to determine the comprehensive traversability risk of different road surfaces based on the probability that each pixel belongs to any terrain type. An entropy risk measurement method is used to convert the comprehensive traversability risk of different road surfaces into a quantitative risk, generating a traversability map. Traversability parameters are determined based on the traversability map. A cross-entropy-based motion planner performs motion planning for a vehicle in the unstructured environment based on the traversability parameters. To address the aforementioned issues, the present application proposes a motion planning method based on probabilistic risk. By integrating deep learning and Bayesian inference and incorporating an entropy risk measure (ERM) method, a comprehensive probabilistic assessment of terrain risk is achieved. This method addresses the uncertainty of terrain perception and the lack of terrain adaptability, thereby forming a more efficient and safer motion plan based on a cross-entropy optimization method. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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 creative work.

[0023] Figure 1 A schematic diagram of the flow of the unstructured environment motion planning method provided in an embodiment of the present application;

[0024] Figure 2 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0026] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0027] A motion planning method for unstructured environments mainly includes the following innovative contents: (1) a terrain classification and risk assessment method based on deep learning and Bayesian regression, which uses the UNet model to classify terrain at the pixel level and combines it with Bayesian regression to realize the probabilistic assessment of terrain risk; (2) an ERM-based risk quantification method, which supports dynamic obstacle avoidance and path optimization of vehicles in high-risk terrain by comprehensively evaluating uncertainties; (3) a motion planning framework based on cross-entropy optimization, which directly generates the optimal control sequence and significantly improves the efficiency and robustness of motion planning.

[0028] like Figure 1 As shown, the embodiment of the present application provides a method for motion planning in an unstructured environment, specifically comprising:

[0029] S1: Inputting an unstructured environment road terrain image to be tested into a neural network model to determine the probability that each pixel in the unstructured environment road terrain image belongs to any terrain type; the terrain types include sand, gravel and grass; the neural network model generates a terrain category probability map for each pixel.

[0030] S2: A Bayesian regression model is used to determine the comprehensive drivability risk under different road conditions based on the probability of each pixel belonging to any terrain type.

[0031] S3: The entropy risk measurement method is used to convert the comprehensive drivability risk under different road surfaces into quantitative risk and generate a drivability map.

[0032] S4: Determine a feasibility parameter of a cross-entropy-based motion planner according to the feasibility map.

[0033] S5: Utilizing the cross-entropy-based motion planner, motion planning is performed on the vehicle in the unstructured environment to be tested according to the accessibility parameters.

[0034] This application proposes a motion planning method based on probabilistic risk, which specifically includes three steps: Step 1 models and estimates the road surface categories in the unstructured planning area and the mathematical relationship between the traversability risk and the road slope under different road surface categories; Step 2 uses the Entropy Risk Measure (ERM) to quantify the comprehensive traversability risk under different road surfaces, form a traversability map, and provide the traversability parameters of the motion model for motion planning; Step 3 uses the cross-entropy method to solve the control strategy. This method optimizes the control instructions (including linear velocity and angular velocity) of the unmanned ground vehicle based on implicit traversability, avoids high-risk areas, and guides the unmanned ground vehicle to travel along a path that balances efficiency and safety.

[0035] Further, in an exemplary embodiment, step S1 can be replaced with the following steps.

[0036] S101: Utilize the formula to determine the probability of each pixel belonging to any terrain type in the image I. where softmax is the softmax function. is the probability of each pixel belonging to terrain type c given the image I. is a neural network model that is .

[0037] A method for motion planning in unstructured environments is proposed, which includes the following steps: first, use the UNet model to identify the road surface class attributes in unstructured environments, the UNet model is widely used in pixel-level image segmentation tasks, its input is the color image of the terrain, and the output is the pixel-level classification of the terrain class, this segmentation allows to estimate the probability of each pixel belonging to a specific terrain type (e.g. sand, gravel, grass).

[0038] Mathematically, let the input image be where H and W are the height and width of the image, respectively. The UNet model generates a terrain class probability map for each pixel i, where c is the terrain class and yi is the terrain label for pixel i.

[0039] Further, in an exemplary embodiment, step S2 can be replaced with the following steps.

[0040] S201: Use Bayesian regression to model the relationship between terrain slope angle and traversability risk for each terrain type, and predict the mean and variance of the traversability risk for each pixel.

[0041] After segmenting the terrain, focus on modeling the traversability risk of a specific terrain as a regression function of the terrain slope angle ϕ. Here, Bayesian regression is used to model the relationship between terrain slope angle and traversability risk for each terrain type.

[0042] Through training data i, ri), where i represents the slope angle and ri represents the traversal risk of each pixel i, Bayesian regression is used to predict the mean μ( ) and variance σ2( ) of the traversal risk:

[0043] .

[0044] S202: Based on the probability of each pixel belonging to any terrain type, determine the risk distribution of each terrain type according to the mean and variance, and determine the comprehensive passable risk under different road surfaces according to the probability of each pixel belonging to any terrain type.

[0045] S203: Use the Bayesian regression model to predict the mean and variance of the comprehensive passable risk.

[0046] The output risk distribution of any terrain type is determined by the formula ; wherein, is the probability of each pixel belonging to any terrain type; is the output risk distribution of any terrain type to which each pixel belongs; is the mean of any terrain type of the comprehensive passable risk; is the variance of any terrain type c of the comprehensive passable risk.

[0047] Finally, by aggregating the risk of each terrain category (weighting the probability of the terrain category), the comprehensive probabilistic traversal risk of the entire terrain is calculated, i.e. for a given slope angle ϕ, the comprehensive risk is calculated by the formula to determine the comprehensive passable risk of the entire terrain; wherein, is the probability distribution of the passable risk given the input image I; is the comprehensive passable risk of the entire terrain; is the probability of each pixel belonging to terrain type c given the input image I; is the probability of each pixel belonging to any terrain type; is the expectation of the probability of each pixel belonging to any terrain type.

[0048] S204: The properties of the Bayesian regression model allow the quantification of the uncertainty of the passable risk. The total uncertainty of the traversal risk of a given pixel can be expressed as:

[0049] The total uncertainty of the passable risk of each pixel is determined by the formula ; wherein, is the probability of the entire terrain, is the passable risk uncertainty of each pixel.

[0050] .

[0051] Further, in an exemplary embodiment, step S3 can be replaced by the following steps.

[0052] S301: Using the entropy risk measurement method, using the formula The comprehensive traversability risk under different road conditions is converted into quantitative risk to generate a traversability map; To find the entropy risk measure for a random variable X under the premise of a given risk aversion parameter, X represents , is the probability that each pixel belongs to any terrain type; is the risk aversion parameter; E represents the expectation of X; ERM is the guaranteed loss amount.

[0053] Here, we use the entropy risk metric to convert probabilistic risk into quantitative risk. Entropy risk measurement (ERM) is a unified risk measurement method based on the concept of entropy, used to assess risk in uncertain environments. It captures both the magnitude and probability of risk. Given a random variable X representing risk, the ERM is defined as follows:

[0054] .

[0055] Where λ > 0 is the risk aversion parameter; smaller λ indicates a more risk-averse decision maker. E represents the expectation (or mean) of the distribution of the random variable X. ERM can be understood as the "certainty equivalent value" of risk outcomes—the guaranteed loss a decision maker is willing to accept given a specific level of risk aversion.

[0056] Further, in an exemplary embodiment, step S4 may be replaced by the following steps.

[0057] S401: Utilize formula Determine the accessibility parameters; where, represents the control input to the cross-entropy based motion planner, , is the linear velocity, is the angular velocity, T is the transposed matrix, , is an m-dimensional real number space; , is a q-dimensional real number space; is the accessibility parameter; is the state vector, , is the position of the vehicle in the XY plane in the spatial coordinate system, is the heading angle of the vehicle. The spatial coordinate system is a coordinate system fixed to the ground, ensuring that the x and y directions are perpendicular. Affects linear velocity and angular velocity, is the feasibility parameter affecting the linear velocity, is the feasibility parameter that affects the angular velocity.

[0058] A cross-entropy-based motion planner is used to solve motion planning problems in unstructured environments, with the goal of navigating to a specified target location quickly and efficiently. ,in represents a two-dimensional real space. In unstructured environments, UGV motion control faces significant challenges due to the variability and uncertainty of terrain conditions. To address these challenges, the vehicle's dynamics are modeled as a discrete-time system whose state transitions are influenced by both control inputs and accessibility parameters.

[0059] Specifically, consider a UGV whose motion is affected by a passability parameter that depends on the underlying terrain. The state of the system at any time t Evolution according to the following dynamic formula:

[0060] .

[0061] in, represents the control input, represents the passability parameter (passability parameter + passability risk = 1); Represents n-dimensional real space. A single wheel model is used to describe the kinematic behavior of the vehicle. The state vector Contains the vehicle's position in the XY plane and its heading angle . Control input Including line speed and angular velocity Passability parameters Affects linear velocity and angular velocity, characterizing the effect of terrain on vehicle motion.

[0062] S402: Using formula , determines the dynamic model of the terrain on the vehicle; where Δ is the time step.

[0063] Based on the dynamic model of the terrain on the vehicle, the formula Minimize the cost function consisting of terminal cost and stage cost to determine the motion plan of the vehicle in the unstructured environment to be tested; among them, the terminal cost for ; Cost of the stage for ;in, is the default speed estimate; is the weight of penalizing the distance from the target position; is the indicator function; is the current state of the vehicle, is the vehicle state after time T; is the target point position; is the position of the vehicle after time T; is the vehicle's position after time t. Although the objective function primarily aims to minimize the time to the target, optimizing for traversability is implicitly embedded. As shown in the system dynamics equations, higher traversability results in lower speed penalty, allowing the vehicle to maintain a higher speed. Therefore, achieving faster target arrival naturally requires the motion planning algorithm to select areas with higher traversability.

[0064] The cross entropy control algorithm is used to optimize the above objective function: the cross entropy control algorithm is a stochastic optimization technique that minimizes the total cost by iteratively optimizing the distribution of the control sequence. The specific steps include the following five stages:

[0065] Sampling phase: In each iteration, a sample is drawn from a Gaussian distribution Sample N control sequences in , where μ and Σ are the mean and covariance of the control sequence distribution, respectively.

[0066] Evaluation phase: For each sampling sequence, simulate the system dynamics and evaluate the cost function Calculate its cost.

[0067] Selection stage: Select the K control sequences with the lowest cost as the elite set.

[0068] Update phase: Use the elite set to update the mean μ and covariance Σ of the Gaussian distribution to focus the search on the optimal control sequence. The goal is to minimize the divergence (Kullback-Leibler, KLD) between the current distribution and a target distribution that assigns high probability to the optimal solution. The mathematical expression is as follows:

[0069] .

[0070] in, is the Kullback-Leibler divergence, which is used to measure the degree of difference between the distributions q and p; is the conditional distribution q; is the conditional distribution p; is the expectation of the distribution q; is a random variable; is the assumed parameter; is the target parameter.

[0071] Termination stage: When the distribution converges or the maximum number of iterations is reached, the optimal control sequence is output.

[0072] Ultimately, the optimized control sequence enables efficient and reliable motion planning in dynamic and uncertain environments.

[0073] The present application provides a motion planning device for an unstructured environment, specifically comprising:

[0074] The terrain type probability determination module inputs the unstructured environment road terrain image to be tested into the neural network model to determine the probability that each pixel in the unstructured environment road terrain image belongs to any terrain type; the terrain types include sand, gravel and grass; the neural network model generates a terrain category probability map for each pixel.

[0075] The comprehensive drivability risk determination module uses a Bayesian regression model to determine the comprehensive drivability risk under different road conditions based on the probability that each pixel belongs to any terrain type.

[0076] The accessibility map generation module uses the entropy risk measurement method to convert the comprehensive accessibility risk under different road surfaces into quantitative risk and generate an accessibility map.

[0077] A feasibility parameter determination module determines a feasibility parameter of a cross-entropy-based motion planner according to the feasibility map.

[0078] The vehicle motion planning module in the unstructured environment utilizes the cross-entropy-based motion planner to perform motion planning for the vehicle in the unstructured environment to be tested according to the feasibility parameters.

[0079] The present application provides a computer device, which can be a server or a terminal, and its internal structure diagram can be as follows: Figure 2 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store video tag processing data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a video tag processing method is implemented.

[0080] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, 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.

[0081] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A motion planning method for unstructured environments, characterized in that: The unstructured environment motion planning method comprises: Inputting an unstructured environment road terrain image to be tested into a neural network model, determining the probability of each pixel in the unstructured environment road terrain image belonging to any terrain type; the terrain types include sand, gravel, and grass; the neural network model generates a terrain category probability map for each pixel; A Bayesian regression model is used to determine the comprehensive drivability risk of different road surfaces based on the probability of each pixel belonging to any terrain type, including: Using the entropy risk measurement method, using the formula The comprehensive traversability risk under different road conditions is converted into quantitative risk to generate a traversability map; To find the entropy risk measure value of random variable X under the premise of given risk aversion parameter, X is ; is the probability that each pixel belongs to any terrain type; is the risk aversion parameter; E is the expectation of X; ERM is the guaranteed loss amount; The entropy risk measurement method is used to convert the comprehensive drivability risk under different road surfaces into quantitative risk and generate a drivability map; Determining feasibility parameters of a cross-entropy-based motion planner according to the feasibility map; the feasibility parameters include feasibility parameters affecting linear velocity and feasibility parameters affecting angular velocity; The cross-entropy-based motion planner is used to perform motion planning for a vehicle in an unstructured environment to be tested according to the traversability parameters.

2. The unstructured environment motion planning method according to claim 1, characterized in that: Inputting the unstructured environment road terrain image to be tested into the neural network model and determining the probability that each pixel in the unstructured environment road terrain image belongs to any terrain type specifically includes: Using the formula Determine the probability that each pixel in the unstructured environment road terrain image to be tested belongs to any terrain type; wherein, is the probability that each pixel belongs to any terrain type; softmax is the softmax function; is the probability that each pixel belongs to terrain type c under the given unstructured environment road terrain image I to be tested; is the terrain type corresponding to the i-th image pixel; is a neural network model, which is Model.

3. The unstructured environment motion planning method according to claim 1, characterized in that: The Bayesian regression model is used to determine the comprehensive drivability risk of different road surfaces based on the probability of each pixel belonging to any terrain type, specifically including: A Bayesian regression model was used to construct the relationship between terrain slope angle and the traversability risk of each terrain type, and the mean and variance of the traversability risk of each pixel were predicted. Based on the probability that each pixel belongs to any terrain type, the risk distribution corresponding to each terrain type is determined according to the mean and the variance, and the comprehensive drivability risk under different road surfaces is determined according to the probability that each pixel belongs to any terrain type.

4. The unstructured environment motion planning method according to claim 3, characterized in that: The method of determining the risk distribution corresponding to each terrain type based on the probability of each pixel belonging to any terrain type and the mean and variance, and determining the comprehensive drivability risk under different road surfaces based on the probability of each pixel belonging to any terrain type, specifically includes: The Bayesian regression model is used to predict the mean of the comprehensive traversability risk and variance ; Using the formula Determine the output risk distribution for any terrain type; where, is the probability that each pixel belongs to any terrain type; Output risk distribution for each pixel belonging to any terrain type; is the average value of the comprehensive traversability risk for any terrain type; is the variance of any terrain type c of the comprehensive traversability risk; Using the formula Determine the overall traversability risk of the entire terrain; where: is the probability distribution of the passable risk given the input image I; The combined traversability risk for the entire terrain; is the probability that each pixel belongs to terrain type c given the input image I; is the probability that each pixel belongs to any terrain type; Find the expected probability that each pixel belongs to any terrain type.

5. The unstructured environment motion planning method according to claim 4, characterized in that: The method further includes determining the risk distribution corresponding to each terrain type based on the probability that each pixel belongs to any terrain type and the mean and variance, and determining the comprehensive drivability risk under different road surfaces based on the probability that each pixel belongs to any terrain type. Using the formula Determine the total uncertainty of the traversability risk for each pixel; where, is the uncertainty of the traversable risk for each pixel, is the probability of the entire terrain.

6. The unstructured environment motion planning method according to claim 1, characterized in that: Determining the feasibility parameters of the cross-entropy-based motion planner according to the feasibility map specifically includes: Using the formula Determine the accessibility parameters; where, represents the control input to the cross-entropy based motion planner, , is the linear velocity, is the angular velocity, T is the transposed matrix, , is an m-dimensional real number space; , is a q-dimensional real number space; is the accessibility parameter, ; is the state vector, , is the position of the vehicle in the XY plane in the spatial coordinate system, is the heading angle of the vehicle. The spatial coordinate system is a coordinate system fixed to the ground, ensuring that the x and y directions are perpendicular. Affects linear velocity and angular velocity, is the feasibility parameter affecting the linear velocity, is the feasibility parameter that affects the angular velocity.

7. The unstructured environment motion planning method according to claim 1, characterized in that: The cross-entropy-based motion planner is used to perform motion planning for a vehicle in an unstructured environment according to the accessibility parameters, specifically comprising: Using the formula , determine the dynamic model of the terrain on the vehicle; where Δ is the time step; Based on the dynamic model of the terrain on the vehicle, the formula Minimize the cost function consisting of terminal cost and stage cost to determine the motion plan of the vehicle in the unstructured environment to be tested; among them, the terminal cost for ; Cost of the stage for ;in, is the default speed estimate; is the weight of penalizing the distance from the target position; is the indicator function; is the current state of the vehicle, is the vehicle state after time T; is the target point position; is the position of the vehicle after time T; is the position of the vehicle after time t.

8. A motion planning device for an unstructured environment, characterized in that: The unstructured environment motion planning device comprises: A terrain type probability determination module inputs an unstructured environment road terrain image to be tested into a neural network model and determines the probability that each pixel in the unstructured environment road terrain image belongs to any terrain type; the terrain types include sand, gravel, and grass; the neural network model generates a terrain category probability map for each pixel; The comprehensive drivability risk determination module uses a Bayesian regression model to determine the comprehensive drivability risk of different road surfaces based on the probability of each pixel belonging to any terrain type; The accessibility map generation module uses the entropy risk measurement method to convert the comprehensive accessibility risk under different road conditions into quantitative risk and generate accessibility maps. Specifically, it includes: Using the entropy risk measurement method, using the formula The comprehensive traversability risk under different road conditions is converted into quantitative risk to generate a traversability map; To find the entropy risk measure value of random variable X under the premise of given risk aversion parameter, X is ; is the probability that each pixel belongs to any terrain type; is the risk aversion parameter; E is the expectation of X; ERM is the guaranteed loss amount; a feasibility parameter determination module, which determines feasibility parameters of a cross-entropy-based motion planner according to the feasibility map; The vehicle motion planning module in the unstructured environment utilizes the cross-entropy-based motion planner to perform motion planning for the vehicle in the unstructured environment to be tested according to the feasibility parameters.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the unstructured environment motion planning method according to any one of claims 1 to 7.

Images