Computing resource restricted environment-based incomplete constraint vehicle path planning method

Abstract

The invention discloses an incomplete constraint vehicle path planning method based on a computing resource constrained environment, and solves the problems of computing resource waste, easy local optimum and insufficient scene adaptation of an existing grid sampling method. The method is applied to a qualified area scene with a tolerance range, and comprises the following steps of: defining a fuzzy target domain and a linear search interval which are formed by an ideal center point and the tolerance range, constructing a cost evaluation function containing factors such as path length and gear shifting times, generating a tentative path by adopting interval binary search iteration, and reducing the interval according to a cost trend, so as to obtain a tentative path; after the optimal target point is locked, a Reed-Shepp path is generated; and the search precision is dynamically adjusted through the processor load rate. The method does not need full sampling, reduces calculation overhead, adapts to resource-limited environments such as a vehicle-mounted low-end chip and embedded teaching equipment, and is especially suitable for low-speed close-range path planning such as back-up parking and side parking in a driving training scene.

Description

Technical Field

[0001] This invention relates to the field of vehicle motion planning and intelligent driving assistance technology, specifically to a nonholonomic constrained vehicle path planning method based on a computationally limited environment. Background Technology

[0002] In the field of autonomous driving and intelligent driving assistance technology, vehicle path planning is a core technology to ensure driving safety and smooth operation. Especially in low-speed, close-range operation scenarios such as reversing into a parking space and parallel parking, the adaptability and efficiency of path planning directly affect the implementation effect of the technology. The demand for intelligent driver training equipment and deployment of low-end automotive chips further poses a dual challenge to path planning technology: "adaptation of computing resources" and "matching of scenario requirements".

[0003] I. Commonly Known Technical Content 1. Nonholonomic Constraints and Reeds-Shepp (RS) Curves: As a typical nonholonomic constraint system, vehicles have inherent constraints during their motion—they cannot move laterally, and path generation must strictly adhere to this kinematic characteristic. The Reeds-Shepp curve (RS curve) is a well-known geometric algorithm in this field. Its core function is to accurately calculate the shortest path between two defined poses while satisfying the vehicle's nonholonomic constraints. Due to its optimal path and kinematic adaptability, this algorithm has become the mainstream foundational algorithm for short-distance vehicle path planning. However, it should be noted that the original design of the RS curve is only for "fixed target points," meaning the input parameters must be clearly defined start and end coordinates, and it cannot directly adapt to target scenarios with tolerance ranges.

[0004] 2. The "Fuzzy Destination" Characteristics of Driver Training Scenarios: In driver training scenarios such as reversing into a parking space and parallel parking, the target of path planning differs fundamentally from traditional path planning—it is not an absolute coordinate point, but a "qualified area" (i.e., a fuzzy destination) with a certain tolerance range. For example, in reversing into a parking space training, as long as the vehicle is parked within a 20cm allowable range before and after the parking space, it is considered a qualified operation. This tolerance characteristic of the target is a core feature of driver training scenarios and a scenario requirement that path planning technology needs to be specifically adapted to.

[0005] II. The closest existing technical solution To resolve the compatibility issue between RS curves and the "fuzzy target domain" in driver training scenarios, the most commonly used technical solution in this field is the grid sampling method. Its core idea is to transform continuous qualified regions into discrete target points through "full exhaustive sampling," thereby adapting to the application requirements of RS curves. The specific implementation steps are as follows: 1. Discretize the target region: Divide the "qualified region" (fuzzy target domain) in the driver training scenario into a grid with a fixed interval (usually 1cm). Break the continuous tolerance region into N discrete target pose points (the number of N depends on the area of ​​the qualified region, usually hundreds or thousands). 2. Full path calculation: Based on the RS curve algorithm, the path from the driving starting point to each discrete target point is calculated one by one, forming N candidate paths. The calculation process of each path involves a large number of trigonometric function operations, which has a large computational cost. 3. Cost Comparison and Optimal Selection: Define a single path cost evaluation index (mainly path length), compare the costs of N candidate paths one by one, and select the path with the lowest cost as the final planning result.

[0006] III. Deficiencies of Existing Technology Considering the actual needs of driver training scenarios and the limited computing resources of deployment environments (low-end automotive chips, embedded teaching equipment), existing grid sampling methods and traditional path planning schemes have the following irreconcilable technical defects: 1. Severe waste of computing resources and significant runtime latency: The grid sampling method requires calculating RS curves for hundreds or thousands of discrete target points one by one. The calculation of RS curves involves a large number of high-overhead trigonometric function operations, resulting in a huge overall computational load. On hardware platforms with limited computing resources, such as low-end automotive chips and embedded teaching equipment, this solution will result in excessively long path planning time, which will lead to vehicle control latency and affect the real-time performance of operations and the training experience. 2. Falling into the "sampling density contradiction" can easily lead to a local optimum trap: The path optimality of the existing solution depends entirely on the sampling interval density. If the sampling interval is too sparse (greater than 1cm), the true optimal target point in the fuzzy target domain may be missed, resulting in the final planned path being a local optimum rather than a global optimum. If the sampling interval is reduced to pursue optimality (e.g., less than 1cm), the number of discrete target points will increase dramatically, further aggravating the consumption of computing resources and exceeding the computing power capacity of the limited hardware, forming an irreconcilable contradiction between "optimality and computing power". 3. Insufficient adaptability to specific scenarios and incompatible with driver training logic: Existing solutions focus solely on the single metric of "shortest path length" in cost assessment. However, the core requirements of driver training scenarios are "low operational difficulty, few adjustments, and adherence to human-like teaching logic" (such as instructors' concerns about "can the car enter the parking space in one go" and "minimum steering wheel adjustments"). Simply finding the shortest path often requires frequent steering wheel adjustments or gear shifts, which is out of sync with the teaching needs of driver training scenarios, resulting in insufficient practicality of the planned path.

[0007] In summary, existing path planning solutions cannot simultaneously meet the three core requirements of "adaptation to environments with limited computing resources", "global optimality of the path", and "matching with the teaching logic of driver training scenarios". The technical pain points are prominent and a brand-new technical solution is urgently needed to break through them. Summary of the Invention

[0008] The core objective of this invention is to address the shortcomings of existing grid sampling methods in computationally limited environments (such as low-end automotive chips and embedded teaching equipment) and fuzzy target domains (such as the "qualified area" in driver training scenarios). These shortcomings include wasted computational resources, significant runtime latency, susceptibility to local optima traps, and incompatibility with teaching / anthropomorphic habits. This invention provides a highly efficient and adaptable nonholonomic constrained vehicle path planning method. Specifically, this invention innovatively transforms the path planning problem into an interval search problem, abandoning the exhaustive sampling approach. It utilizes the continuous characteristics of vehicle kinematics within a small range and the monotonic or convex function properties of path cost within the tolerance interval, combined with a binary search algorithm to quickly converge to the optimal target point, significantly reducing the number of Reeds-Shepp (RS) curve calculations and thus lowering computational overhead. Furthermore, this invention constructs a cost evaluation function that includes multiple factors such as path length, number of gear shifts, steering wheel position, and centering, generating paths that conform to the teaching logic of "entering the parking space in one go" and "minimizing the number of adjustments" in driver training scenarios. A dynamic search termination mechanism based on processor load rate flexibly adjusts the iteration depth, achieving a dynamic balance between computational accuracy and resource consumption. Ultimately, while balancing path optimization and computational feasibility, this invention provides a reliable path planning solution for scenarios with limited computational resources and targets with tolerance ranges, particularly meeting the practical application needs of driver training businesses.

[0009] To achieve the above objectives, the technical solution of this invention is: a nonholonomic constrained vehicle path planning method based on a computationally limited environment, applied to scenarios where the target is a qualified area with a tolerance range, comprising the following steps: S1. Define the fuzzy target domain and search interval: Determine the starting point of the path planning and the fuzzy target domain corresponding to the qualified area; the fuzzy target domain is defined by the ideal center point and the tolerance range, and a linear search interval [D_min, D_max] is defined based on the fuzzy target domain; S2. Define the cost evaluation function: Define the cost function used to evaluate the merits of the Reeds-Shepp path; S3. Perform a binary search within the interval to locate the optimal target point: Within the linear search interval, use a binary search strategy to iteratively search for the coordinates of the target point that minimizes the cost function value, specifically including: S31. Initialize the search boundary: Let the left boundary L = D_min and the right boundary R = D_max; S32. Iterative Search: Loop through the following sub-steps until the preset termination condition is met: S321. Calculate the midpoint: Take the midpoint M = (L + R) / 2; S322. Path Generation and Cost Calculation: Based on the Reeds-Shepp curve algorithm, generate paths from the starting point to points L, M, and R respectively, and calculate the corresponding path costs C_L, C_M, and C_R; S323. Gradient Judgment and Interval Update: Compare the magnitudes of C_L, C_M, and C_R. If C_L < C_M, update the right boundary R of the search interval to M; if C_R < C_M, update the left boundary L of the search interval to M; if C_M is less than both C_L and C_R, shrink the search interval towards the midpoint M; S33. Output Result: Take the coordinates of the target point determined when the termination condition is met as the final optimal target point; S4. Generate and Send the Final Path: Based on the starting point and the optimal target point, generate the final Reeds-Shepp path and send it to the vehicle control layer; Among them, the method optimizes the determination method of the target point through the interval binary search to adapt to the environment with limited computing resources.

[0010] Furthermore, in step S1, the fuzzy target domain is the qualified area corresponding to reverse parking or parallel parking in the driving training teaching scenario.

[0011] Furthermore, in step S1, the initial value range of the linear search interval [D_min, D_max] is [0.5m, 1.0m].

[0012] Furthermore, in step S2, the evaluation factors of the cost function include at least one of path length, number of gear shifts, steering wheel rotation amplitude, and vehicle centering degree within the target area.

[0013] Furthermore, in step S32, the preset termination condition is: the width |R - L| of the current search interval is less than the preset precision threshold ε, or the number of iterations reaches the preset maximum iteration limit.

[0014] Furthermore, it further includes step S5. Dynamically Adjust the Search Precision: Dynamically adjust the precision threshold ε and / or the maximum iteration limit according to the real-time load rate of the processor running this method.

[0015] Furthermore, the environment with limited computing resources is an in-vehicle low-end chip or an embedded teaching device.

[0016] This invention utilizes a core technical solution for optimizing the target point determination method through interval binary search. By combining a multi-factor cost evaluation function and a dynamic search accuracy adjustment mechanism, it effectively addresses the technical shortcomings of existing grid sampling methods, achieving the following significant beneficial effects: This invention abandons the approach of exhaustive sampling and instead uses interval binary search, requiring only iterative calculation of a small number of trial paths (generating only 3 Reeds-Shepp paths per iteration) to quickly converge to the optimal target point. This significantly reduces the number of calculations for trigonometric function-intensive Reeds-Shepp curves, effectively lowering the overall computational load. This characteristic enables the method to run efficiently on resource-constrained platforms such as low-end automotive chips and embedded teaching equipment, avoiding the runtime latency issues of traditional solutions and ensuring the real-time performance and smooth operation of vehicle control.

[0017] Leveraging the continuous nature of vehicle kinematics within a small range and the monotonic or convex function characteristics of path cost within the tolerance interval, this invention employs an iterative interval-reducing binary search strategy. This approach allows for precise localization of the globally optimal or suboptimal target point within the fuzzy target domain without relying on high-density grid sampling. It avoids the omission of optimal solutions caused by overly sparse sampling while also mitigating the computational burden of high-density sampling, thus perfectly balancing path optimality and computational feasibility.

[0018] This invention constructs a multi-factor cost evaluation function that includes path length, number of gear shifts, steering wheel rotation amplitude, and centering degree, replacing the traditional single-factor path length evaluation standard. The generated path better meets the anthropomorphic teaching needs of driver training scenarios, such as "entering the parking space in one go," "minimizing the number of adjustments," and "low operational difficulty." It solves the problem of the traditional shortest path being disconnected from teaching logic, significantly improving the practicality and teaching effectiveness of path planning in driver training.

[0019] By employing a dynamic search accuracy adjustment mechanism based on the processor's real-time load rate, this invention can flexibly adapt to different computing power scenarios. When the processor load is high, the iteration depth is reduced by adjusting the accuracy threshold or the maximum number of iterations to prioritize operational efficiency; when the load is low, the iteration depth is increased to ensure path accuracy. This design enables the method to adapt to complex and ever-changing hardware computing power environments, further broadening the application scenarios and adaptability of the technology.

[0020] In summary, this invention achieves multiple improvements in computational efficiency, scenario adaptability, and environmental flexibility while ensuring path optimization. It provides an efficient and reliable solution for non-holonomic constrained vehicle path planning with limited computational resources and a tolerance range for the target, and has outstanding application value, especially in low-speed, close-range operations such as reversing into a parking space and parallel parking in driver training scenarios. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this embodiment. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the nonholonomic constraint vehicle path planning method of the present invention. Detailed Implementation

[0023] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this embodiment. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this embodiment as detailed in the appended claims.

[0024] like Figure 1 As shown, this embodiment provides a nonholonomic constrained vehicle path planning method based on a computationally limited environment, applied to a scenario where the target is a qualified area with a tolerance range, including the following steps: S1. Define the fuzzy target domain and search interval: Determine the starting point of the path planning and the fuzzy target domain corresponding to the qualified area; the fuzzy target domain is defined by the ideal center point and the tolerance range, and a linear search interval [D_min, D_max] is defined based on the fuzzy target domain; In step S1 of this embodiment, the qualified area is a standard parking space (5.0m long and 2.5m wide) for a reverse parking scenario, with a tolerance range of ±20cm in front and behind the parking space; the ideal center point P_center is the longitudinal midpoint in the parking space (coordinate example: 2.5m lateral offset and 4.0m longitudinal offset relative to the starting point); the linear search interval [D_min, D_max] corresponds to the longitudinal tolerance range, with values ​​of [0.5m, 1.0m], that is, the interval from 0.5m forward to 1.0m backward from the ideal center point.

[0025] S2. Define the cost evaluation function: Define a cost function to evaluate the merits of the Reeds-Shepp path; as one implementation, the cost function is calculated using a weighted summation method, specifically the formula: Cost = w1×L + w2×S + w3×A + w4×C, where: L is the path length (in meters), and the weight w1 = 0.3. S represents the number of gear shifts (unit: times), and the weight w2 = 0.2 (the fewer the number of gear shifts, the lower the cost). A is the maximum steering angle of the steering wheel (unit: °), and the weight w3 = 0.3 (the smaller the angle, the lower the operation difficulty and cost); C is the centering degree (value range 0 - 1, C = 1 when the vehicle stops at the center of the qualified area, and C gets smaller as it gets closer to the boundary), and the weight w4 = 0.2; Each weight can be dynamically adjusted according to the teaching focus of the driving training scenario, and the sum of the weights is 1.

[0026] S3. Perform interval binary search to lock the optimal target point: In the linear search interval, use the binary search strategy to iteratively search for the target point coordinates that minimize the cost function value, specifically including: S31. Initialize the search boundary: Let the left boundary L = D_min and the right boundary R = D_max; S32. Iterative search: Loop to execute the following sub - steps until the preset termination conditions are met: In this embodiment, in step S32, the preset precision threshold ε = 0.01m (that is, when the width of the search interval is less than 1cm, it is determined to converge); the preset maximum number of iteration limits = 10 times (to avoid excessive iteration time in extreme cases) S321. Calculate the mid - point: Take the mid - point M=(L + R) / 2; S322. Path generation and cost calculation: Based on the Reeds - Shepp curve algorithm, generate paths from the starting point to points L, M, and R respectively, and calculate the corresponding path costs C_L, C_M, and C_R; S323. Gradient judgment and interval update: Compare the magnitudes of C_L, C_M, and C_R. If C_L < C_M, update the right boundary R of the search interval to M; if C_R < C_M, update the left boundary L of the search interval to M; if C_M is less than both C_L and C_R, shrink the search interval towards the mid - point M. As an implementation method, the specific operation of'shrinking the search interval towards the mid - point M' is: update the left boundary L to (L + M) / 2 and the right boundary R to (M + R) / 2, so that the search interval focuses on the area near the mid - point with the minimum cost and accelerates convergence.

[0027] S33. Output the result: Take the target point coordinates determined when the termination conditions are met as the final optimal target point; S4. Generate and distribute the final path: Based on the starting point and the optimal target point, generate the final Reeds-Shepp path and distribute it to the vehicle control layer. As one implementation, after the final Reeds-Shepp path is generated, it is distributed to the vehicle control layer via CAN bus or serial communication. The control layer outputs control signals such as steering wheel angle, vehicle speed, and gear shifting commands according to the path information to enable the vehicle to travel along the planned path. The path data format adopts JSON format, which includes path point coordinates (x, y, heading angle), control parameters of each path segment, and other information to facilitate parsing by the control layer.

[0028] The method optimizes the determination of the target point through the interval binary search to adapt to environments with limited computing resources.

[0029] As one implementation method, in step S1 of this embodiment, the fuzzy target domain is the qualified area corresponding to reversing into a parking space or parallel parking in a driver training teaching scenario.

[0030] In one implementation, in step S1 of this embodiment, the initial value range of the linear search interval [D_min, D_max] is [0.5m, 1.0m].

[0031] In one implementation, in step S2 of this embodiment, the evaluation factors of the cost function include at least one of path length, number of gear shifts, steering wheel rotation amplitude, and vehicle centering degree within the target area.

[0032] In one implementation, in step S32 of this embodiment, the preset termination condition is: the width |RL| of the current search interval is less than the preset precision threshold ε, or the number of iterations reaches the preset maximum number of iterations limit.

[0033] In one implementation, this embodiment further includes step S5: dynamically adjusting the search accuracy: dynamically adjusting the accuracy threshold ε and / or the maximum iteration limit based on the real-time load rate of the processor running the method. In one implementation, the real-time load rate of the processor is obtained through the system interface of the vehicle-mounted chip or embedded device, and the preset load rate thresholds include a high load threshold (80%) and a low load threshold (50%). When the processor load rate is greater than 80%, it is judged as a high load. The accuracy threshold ε is adjusted to 0.02m, and the maximum number of iterations is adjusted to 8 times to prioritize running efficiency. When the processor load rate is less than 50%, it is judged as low load, and the accuracy threshold ε is adjusted to 0.005m, the maximum number of iterations is adjusted to 15, and the path accuracy is prioritized. Keep the initial parameters unchanged when the load rate is between 50% and 80%.

[0034] In one implementation, the computing resource-constrained environment described in this embodiment is a low-end automotive chip or an embedded teaching device.

[0035] This embodiment adopts an algorithm architecture that transforms the path planning problem into an interval search problem: the algorithm does not change the generation logic of the Reeds-Shepp (RS) curve itself, but achieves a balance between resource adaptation and path optimality by optimizing the determination method of input parameters (target points). It utilizes the continuous characteristics of vehicle kinematics within a small range, assumes that the path cost exhibits monotonic or convex function characteristics within the tolerance interval, and uses binary search to quickly converge to the optimal target point, thereby avoiding the waste of computational resources caused by exhaustive sampling.

[0036] For path planning of fuzzy destinations, this method does not directly calculate all possible paths within the qualified area. Instead, it finds the optimal path by iteratively adjusting the coordinates of the target point: within the allowable target tolerance range, boundary points and median points are selected, and the Reeds-Shepp curve cost from the starting point to the point is calculated respectively. Based on the distribution trend of the cost, the tolerance range is narrowed using a bisection strategy until the target point with the minimum cost is locked.

[0037] Meanwhile, this method is configured with a search termination mechanism that dynamically balances computational accuracy and resources: the iteration depth of the binary search is dynamically adjusted according to the current processor load rate, ensuring path accuracy when computing power is sufficient, and prioritizing running efficiency when computing power is tight.

[0038] like Figure 1 As shown, the corresponding explanations are as follows: The "Start" node serves as the entry point for the method's workflow. S1: Define the fuzzy target domain SearchSpace The sub-step "Determine the ideal center point P_center" corresponds to the core reference point definition operation of the fuzzy target domain in this method; Sub-step “Define linear search interval D_min, D_max”: corresponds to the boundary definition operation of the tolerance range of the fuzzy target domain; Example “[0.5m, 1.0m]”: A typical example of values ​​for a linear search interval; S2: Define the cost evaluation function CostFunction The formula “Cost=f(RS_Path)” corresponds to the cost evaluation logic of the Reeds-Shepp path in this method; Factors "path length, number of gear shifts, steering wheel position, and centering": are the core evaluation dimensions of the cost function; S3: Interval Binary Search (Destination Binary Search) The sub-step of "Initialization" that "sets L = D_min and R = D_max": The operation of assigning the initial boundaries of the corresponding search interval. The termination condition judgment "The termination condition is satisfied, |R - L| < ε or the iteration limit is exceeded": The basis for judging the convergence of the search, where "|R - L| < ε" means that the width of the search interval is less than the preset precision threshold ε, and "the iteration limit is exceeded" means that the number of searches reaches the preset upper limit. The branch of "No (continue the search)": "Take the midpoint M = L + R / 2": The operation of calculating the midpoint of the corresponding search interval. "Generate three trial paths Path(L), Path(M), Path(R)": The operation of generating Reeds-Shepp paths from the starting point to the left boundary point L, the midpoint M, and the right boundary point R. "Calculate the costs C_L, C_M, C_R": The operation of evaluating the costs of the three trial paths. "Gradient judgment": The judgment step for the cost distribution trend, including three branches: "C_L < C_M (left is better) → discard the right side R = M": It means that when the cost of the left boundary point is more optimal, update the right boundary of the search interval to the midpoint M. "C_R < C_M (right is better) → discard the left side L = M": It means that when the cost of the right boundary point is more optimal, update the left boundary of the search interval to the midpoint M. "C_M is the smallest (middle is better) → shrink both sides inward": It means that when the cost of the midpoint is the most optimal, shrink the search interval towards the midpoint. The branch of "Yes (converged)": It means that the search reaches the termination condition and enters the final path generation step. S4: Final Path Generation (PathGeneration) "Lock the final coordinate P_final": The operation of determining the optimal target point after the search converges. "Generate the final planned path": The operation of generating a Reeds-Shepp path from the starting point to the optimal target point P_final. "Send it to the control layer": The operation of transmitting the planned path as an instruction to the vehicle control layer. The "End" node: The process termination exit of the method.

[0039] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A nonholonomic constrained vehicle path planning method based on a computationally limited environment, characterized in that: Applied to the scenario where the target is a qualified area with a tolerance range, including the following steps: S1. Define the fuzzy target domain and the search interval: Determine the starting point of path planning and the corresponding fuzzy target domain of the qualified area; The fuzzy target domain is defined by the ideal center point and the tolerance range, and a linear search interval [D_min, D_max] is defined based on the fuzzy target domain; S2. Define the cost evaluation function: Define a cost function used to evaluate the quality of the Reeds-Shepp path; S3. Perform interval binary search to lock the optimal target point: Within the linear search interval, use the binary search strategy to iteratively search for the target point coordinates that minimize the value of the cost function, specifically including: S31. Initialize the search boundary: Let the left boundary L = D_min and the right boundary R = D_max; S32. Iterative search: Loop through the following sub-steps until the preset termination condition is met: S321. Calculate the midpoint: Take the midpoint M = (L + R) / 2; S322. Path generation and cost calculation: Based on the Reeds-Shepp curve algorithm, generate paths from the starting point to the L point, M point, and R point respectively, and calculate the corresponding path costs C_L, C_M, C_R; S323. Gradient judgment and interval update: Compare the magnitudes of C_L, C_M, C_R. If C_L < C_M, update the right boundary R of the search interval to M; If C_R < C_M, update the left boundary L of the search interval to M; If C_M is less than both C_L and C_R, shrink the search interval towards the midpoint M; S33. Output the result: Take the target point coordinates determined when the termination condition is met as the final optimal target point; S4. Generate and send the final path: Based on the starting point and the optimal target point, generate the final Reeds-Shepp path and send it to the vehicle control layer; Among them, the method optimizes the determination method of the target point through the interval binary search to adapt to the environment with limited computing resources.

2. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to claim 1, characterized in that: In step S1, the fuzzy target domain is the qualified area corresponding to reverse parking or parallel parking in the driving training teaching scenario.

3. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to claim 2, characterized in that: In step S1, the initial value range of the linear search interval [D_min, D_max] is [0.5m, 1.0m].

4. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to claim 1, characterized in that: In step S2, the evaluation factors of the cost function include at least one of path length, number of gear shifts, steering wheel rotation amplitude, and vehicle centering degree within the target area.

5. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to claim 1, characterized in that: In step S32, the preset termination condition is: the width |R - L| of the current search interval is less than the preset precision threshold ε, or the number of iterations reaches the preset maximum iteration limit.

6. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to claim 5, characterized in that: It further includes step S5. Dynamically adjust the search precision: Dynamically adjust the precision threshold ε and / or the maximum iteration limit according to the real-time load rate of the processor running this method.

7. The nonholonomic constrained vehicle path planning method based on a computationally limited environment according to any one of claims 1 to 6, characterized in that: The environment with limited computing resources is an in-vehicle low-end chip or an embedded teaching device.

Images