Occupation probability update method, apparatus, equipment and mobility tool based on spatiotemporal graph

Abstract

This application relates to a method, apparatus, device, and mobile tool for updating occupancy probability based on a spatiotemporal graph. When updating the occupancy probability, if historical points have observation points at the same location, the historical probability and observation probability are linearly calculated in logarithmic probability space to obtain a fused result, which is then converted into a target probability. Otherwise, the historical probability is decayed based on the current number of unobserved historical points to obtain the target probability. Finally, based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, a historical trajectory point set for the next update cycle is generated. This application transforms the traditional direct probability calculation method into a linear calculation in logarithmic probability space, consistent with Bayesian theory, thereby ensuring the stability and accuracy of the fusion process.

Description

Technical Field

[0001] This application relates to the field of probability update technology, and in particular to a method, apparatus, device and mobile tool for updating occupancy probability based on spatiotemporal graphs. Background Technology

[0002] Spatio-Temporal Graphs (ST graphs) play a central role in longitudinal decision-making and planning for autonomous driving. By gridding spatial (longitudinal distance, s-axis) and temporal (t-axis) dimensions, ST graphs enable the system to assess the positional and temporal relationships of the planned trajectory relative to predicted obstacles. In dynamic environments, the value of ST graphs is particularly prominent, as they help the system infer the states of other traffic participants (such as vehicles and pedestrians) and static obstacles over a future period. This assessment of future spatio-temporal occupancy is fundamental to path planning and collision detection, directly impacting the safety and comfort of autonomous vehicles. Occupancy probability is the basis of the ST graph's state representation, quantifying the likelihood that a spatio-temporal region at a specific time t and a specific longitudinal distance s is occupied by an obstacle. This probability distribution is a key input to the cost function in the downstream trajectory planning module, decisively influencing the optimization of the velocity curve and ultimately determining the safety and driving experience of the autonomous vehicle.

[0003] In existing technologies, updating the occupancy probability of an ST chart can employ a linear weighted and decaying method. This means that when a new prediction is made, a weighted average or other linear accumulation method is used to update the probability; when no new prediction is made, the probability decays linearly over time. However, this linear weighted and decaying method cannot provide the reliable probability estimates required for long-term stable operation in complex dynamic environments.

[0004] Therefore, how to stably and accurately update the occupancy probability of the spatiotemporal graph is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] This application provides a method, apparatus, device, and mobile tool for updating the occupancy probability based on a spatiotemporal graph, so as to stably and accurately update the occupancy probability of the spatiotemporal graph.

[0006] Firstly, this application provides a method for updating the occupancy probability based on a spatiotemporal graph, including:

[0007] Determine the set of historical trajectory points and the set of new observation points in the current update cycle; wherein, the set of historical trajectory points includes each historical point with historical probability, and the set of new observation points includes each observation point with observation probability;

[0008] Determine whether each historical point has an observation point in the same location;

[0009] If yes, then the historical probability and the observed probability are linearly operated on in the logarithmic probability space to obtain the fused result, and the result is converted into the target probability; if no, then the historical probability of the historical point is attenuated to obtain the target probability.

[0010] Based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, the historical trajectory point set for the next update cycle is generated.

[0011] Optionally, determining the set of historical trajectory points in the current update cycle includes:

[0012] Retrieve the historical trajectory point set generated in the previous update cycle;

[0013] By simulating the motion state of the current target, the original coordinate values ​​of each historical point in the spatiotemporal graph of the historical trajectory point set are adjusted to obtain the adjusted coordinate values ​​of each historical point.

[0014] From the set of historical trajectory points, delete historical points whose adjusted coordinate values ​​are less than a predetermined threshold to obtain the set of historical trajectory points for the current update cycle.

[0015] Optionally, based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, a historical trajectory point set for the next update cycle is generated, including:

[0016] The target probability of each historical point is compared with the deletion threshold, and the historical points whose target probability is less than the deletion threshold and their corresponding target probabilities are deleted.

[0017] Based on the retained historical points and the newly added observation points in the new observation point set, a historical trajectory point set for the next update cycle is generated.

[0018] Optionally, the historical probability and the observed probability are linearly operated on in the logarithmic probability space to obtain the fused result, and the result is converted into the target probability, including:

[0019] The historical probability of a historical point is converted into the first logarithmic probability, and the observation probability of an observation point is converted into the second logarithmic probability.

[0020] Linear operations are performed using the first log-log odds, the second log-log odds, and the weighting coefficients to generate the fused log-log odds.

[0021] The fused logarithmic odds are converted into the target probability.

[0022] Optionally, the target probability is obtained by attenuating the historical probability of a historical point, including:

[0023] Determine the original number of unobserved data points at historical points;

[0024] The original unobserved count is updated to obtain the current unobserved count;

[0025] Determine whether the current number of unobserved items is greater than the unobserved threshold;

[0026] If yes, then set the target probability of the historical point to zero; otherwise, use a decay factor to decay the historical probability of the historical point to obtain the target probability of the historical point.

[0027] If a historical point has observation points in the same location, the original number of unobserved observations for that historical point will be cleared to zero.

[0028] Optionally, after determining the new set of observation points for the current update cycle, the following may also be included:

[0029] Based on the coordinates of each observation point in the spatiotemporal graph, the corresponding original continuous observation count is determined, and the original continuous observation count is updated to generate the current continuous observation count.

[0030] The observation probability is adjusted based on the current number of consecutive observations at each observation point.

[0031] Optionally, the observation probability can be adjusted based on the current number of consecutive observations at each observation point, including:

[0032] Determine whether the current number of consecutive observations at the observation point is less than the stability threshold;

[0033] If so, the scaling factor is determined based on the current number of consecutive observations;

[0034] The observation probability of the observation point is scaled using the scaling factor to obtain the adjusted observation probability; wherein, if the coordinate value in the spatiotemporal graph does not have a corresponding observation point in the new observation point set, the original continuous observation number of the coordinate value is cleared to zero.

[0035] Secondly, this application provides an occupancy probability update device based on a spatiotemporal graph, comprising:

[0036] A determination module is used to determine the set of historical trajectory points and the set of new observation points in the current update cycle; wherein, the set of historical trajectory points includes each historical point with historical probability, and the set of new observation points includes each observation point with observation probability;

[0037] The judgment module is used to determine whether each historical point has an observation point in the same location; if so, the fusion module is triggered; if not, the attenuation module is triggered.

[0038] The fusion module is used to perform linear operations on historical probabilities and observed probabilities in logarithmic probability space to obtain the fused operation result, and convert the operation result into the target probability;

[0039] The attenuation module is used to attenuate the historical probability of historical points to obtain the target probability;

[0040] The generation module is used to generate a set of historical trajectory points for the next update cycle based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set.

[0041] Thirdly, this application provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0042] Memory, used to store computer programs;

[0043] When a processor executes a program stored in memory, it implements the steps of the above-described occupancy probability update method.

[0044] Fourthly, this application also provides a mobile tool, including the aforementioned electronic device.

[0045] Compared with the prior art, the technical solutions provided in this application have the following advantages: This application provides a method, apparatus, device, and mobile tool for updating occupancy probability based on a spatiotemporal graph. When updating the occupancy probability, if historical points have observation points in the same position, the historical probability and observation probability are linearly operated on in logarithmic probability space to obtain the fused operation result, and the operation result is converted into the target probability; otherwise, the historical probability is decayed according to the current number of unobserved historical points to obtain the target probability; finally, based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, the historical trajectory point set for the next update cycle is generated. This application, through this method, can transform the traditional direct probability calculation method into a linear operation in logarithmic probability space, consistent with Bayesian theory, thereby ensuring the stability and accuracy of the fusion process. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0049] Figure 1 This is a schematic diagram of a vehicle-entry interaction scenario provided in an embodiment of this application;

[0050] Figure 2 A schematic flowchart of an occupancy probability update method based on a spatiotemporal graph provided in an embodiment of this application;

[0051] Figure 3 A schematic diagram of a confidence accumulation mechanism provided in an embodiment of this application;

[0052] Figure 4a The logarithmic probability to probability conversion diagram provided in the embodiments of this application;

[0053] Figure 4b The logarithmic odds and inverse probability conversion diagram provided in the embodiments of this application;

[0054] Figure 5 A flowchart of probability fusion and attenuation provided for embodiments of this application;

[0055] Figure 6 A schematic diagram illustrating the update process of a probability grid provided in an embodiment of this application;

[0056] Figure 7 A flowchart of an ST graph probabilistic raster update method provided in this application embodiment;

[0057] Figure 8 A schematic diagram of an occupancy probability update device based on a spatiotemporal graph provided in this application embodiment;

[0058] Figure 9 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0059] The spatiotemporal graph is represented by a rasterized map. The occupancy probability of cells (s,t) within the ST raster is the basis for the state representation of the ST graph. It quantifies the probability that a spatiotemporal region at a specific time t and a specific longitudinal distance s will be occupied by an obstacle. This probability distribution is a key input to the cost function in the downstream trajectory planning module, which has a decisive impact on the optimization of the velocity curve and ultimately determines the safety and driving experience of autonomous vehicles.

[0060] Given the dynamic nature of the driving environment and the vehicle's motion, this probabilistic ST diagram must undergo continuous temporal evolution and updates. The core technical challenge lies in constructing a mathematically rigorous and functionally robust fusion framework to combine prior beliefs with new observational evidence. The complexity of this challenge is further reflected in the framework's ability to manage the lifecycle of dynamic targets, specifically including: target emergence: how to handle newly entering obstacles; target disappearance: how to handle obstacles that have left the perception range or are occluded; and target state transitions: how to respond to changes in the predicted trajectory of obstacles.

[0061] Therefore, the quality of this update mechanism is crucial, as it directly determines the motion planning characteristics of autonomous vehicles. An overly conservative update logic may lead to unnecessary deceleration or hesitation; conversely, an overly aggressive update strategy may compress safety redundancy, resulting in potentially dangerous maneuvers. This highlights the industry's urgent need for a probabilistic update method that can finely handle uncertainty and simulate the deliberate judgment of a human driver.

[0062] See Figure 1 This figure illustrates a vehicle-entry interaction scenario provided by an embodiment of this application. As shown, in a typical vehicle-entry interaction scenario, the autonomous driving system must process information from the perception and prediction module that has inherent uncertainties. For example, the future trajectory of the target vehicle obj2 has multiple possibilities and is assigned different probabilities.

[0063] In traditional schemes, the update of ST graph occupancy probability mainly focuses on two typical methods:

[0064] Linear weighted sum decay method: When there is a new prediction, the probability is updated using a weighted average or other linear accumulation method; when there is no new prediction, the probability decays linearly over time.

[0065] Odds fusion method: This function combines direct probability adjustment, probability-based calculations, and some heuristic rules. When there is a new prediction, odds are fused based on the old probabilities; when there is no new prediction, odds decay is calculated based on set parameters.

[0066] However, both of these methods have significant limitations in practical applications, especially in terms of mathematical rigor, predictability, and alignment with human driving logic.

[0067] The linear weighting and decay method is theoretically unrigorous and not Bayesian consistent, which may lead to probability updates that do not conform to statistical laws. This theoretical unrigor means that it may not be able to provide the reliable probability estimates required for long-term stable operation in complex dynamic environments. Furthermore, the method lacks smoothness and may be sensitive to noise, especially when there are large fluctuations between old and new probabilities. For safety-critical systems such as autonomous driving that require high precision and robustness, sensitivity to noise may lead to cumulative errors or unpredictable behavior, thereby causing safety risks.

[0068] For the odds fusion method, when the new probability is lower than the old probability, the fusion result may have an increased probability that exceeds the expected range. Furthermore, the decay logic of the odds formula is complex and difficult to control precisely. Moreover, the theoretical basis of this method is not rigorous. To prevent division by zero, the 1.01 term in the denominator of the probability calculation causes the result to deviate from the pure probability definition and may systematically underestimate the probability of extremely high probability events. In addition, this method does not fully conform to human driving logic and may lead to probability updates that are not intuitive (such as overestimation or abrupt changes). It is not suitable for high-precision scenarios. Human drivers tend to make incremental reactions by combining historical experience and current observations, while this method may lead to unnatural decisions.

[0069] Therefore, to overcome the shortcomings of traditional weighted average methods, such as lack of statistical consistency and sensitivity to noise, as well as the problems of logical complexity, unexpected results, insufficient theoretical basis, and inability to effectively handle "ghost" obstacles in odds fusion methods, this invention provides a spatiotemporal graph-based occupancy probability update method, apparatus, device, and mobile tool. This solution provides a stable, reliable, and time-consistent spatiotemporal environment representation, solving the problems of mathematical imprecision, numerical instability, poor predictability, lack of human driving intuition, and difficulty in optimization and maintenance in existing methods when handling dynamic obstacle occupancy probabilities. This provides a high-quality data foundation for downstream planning and decision-making modules.

[0070] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0071] The following disclosure provides numerous different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of the invention. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0072] See Figure 2 The above is a schematic flowchart of an occupancy probability update method based on a spatiotemporal graph, provided in an embodiment of this application, including:

[0073] S101. Determine the historical trajectory point set and the new observation point set in the current update cycle; wherein, the historical trajectory point set includes each historical point with historical probability, and the new observation point set includes each observation point with observation probability;

[0074] In this application, the update period refers to the period for updating the occupancy probability, which can be customized according to actual conditions. In each update period, the occupancy probability can be updated using the occupancy probability update method described in this application. When updating the occupancy probability in the current update period, this application first needs to determine the dataset used in the current update period, which includes the historical trajectory point set (old_points) and the new observation point set (new_points).

[0075] In this application, the historical trajectory point set refers to the dataset generated in the previous update cycle, and the dataset generated in the previous update cycle is used as the historical trajectory point set for the current update cycle; the new observation point set is the dataset of newly observed data. For distinction, this application refers to the data points in the historical trajectory point set as historical points, each with a corresponding historical probability, and the data points in the new observation point set as observation points, each with a corresponding historical probability.

[0076] Furthermore, both historical points and observation points are data points in the spatiotemporal diagram. If the spatiotemporal diagram is a two-dimensional spatiotemporal diagram, the coordinates of the historical points and observation points are in the form of (s, t), where the coordinate value s is the vertical distance and the coordinate value t is the time. If the spatiotemporal diagram is a three-dimensional spatiotemporal diagram, the coordinates of the historical points and observation points are in the form of (s, l, t), where the coordinate value s is the vertical distance, the coordinate value l is the horizontal distance, and the coordinate value t is the time. This provides a unified and temporally consistent environmental representation for joint horizontal and vertical planning. In this embodiment, only a two-dimensional spatiotemporal diagram is used as an example for explanation.

[0077] S102. Determine whether each historical point has an observation point in the same location;

[0078] If yes, then execute S103; otherwise, execute S104.

[0079] In this application, the spatiotemporal map can be represented as a raster map, a vectorized map, or a hybrid representation, etc. Specifically, a raster map divides the space of the spatiotemporal map into regular grid cells, maps data points to each grid cell, and then updates the occupancy probability of each grid cell. A vectorized map uses parametric geometric primitives (such as dynamic polygons and spline curves) to describe the area occupied by obstacles in the space of the spatiotemporal map, and the probability update process is transformed into updating the parameters of these primitives. A hybrid representation combines the advantages of raster and vectorized representations; for example, a fine raster is used for dynamically changing or highly uncertain areas, while a vectorized representation is used for obstacles with regular shapes and definite movements.

[0080] If the spatiotemporal graph is represented as a raster map, after determining the historical trajectory point set and the current new observation point set, this application needs to determine the minimum bounding rectangle boundary covering all the aforementioned data points to define the effective area of ​​the spatiotemporal graph to be processed in the current calculation cycle. Then, a temporary ST grid is initialized for this effective area, and each data point in the new observation point set is mapped and placed in the corresponding cell within the temporary ST grid according to its coordinate values. Therefore, when determining whether each historical point has an observation point in the same position, this application can check whether the grid cell where the historical point is located has a corresponding observation point. That is: if a historical point and an observation point are in the same grid cell, then the historical point is determined to have an observation point in the same position; if a historical point and all observation points are not in the same grid cell, then the historical point is determined not to have an observation point in the same position; if an observation point and all historical points are not in the same grid cell, then the observation point is determined to be a newly added observation point.

[0081] If the spatiotemporal graph is represented as a vectorized map, then when checking whether a historical point has an observation point at the same location, it is directly determined whether the coordinate values ​​of the historical point and the new observation point are the same. If the coordinate values ​​of the historical point and the new observation point are the same, then it is determined that the historical point has an observation point at the same location; otherwise, it is determined that the historical point does not have an observation point at the same location.

[0082] S103. Perform linear operations on the historical probability and observed probability in the logarithmic probability space to obtain the fused operation result, and convert the operation result into the target probability.

[0083] In this application, if historical points have observation points in the same location, the historical probability of the historical point and the observation probability of the observation point need to be fused. The fused probability is referred to as the target probability. To ensure the accuracy of probability fusion, this application employs a log-odds-based Bayesian filtering method. The probability values ​​in the interval [0,1] are mapped to the log-odds space (-∞,+∞) for linear operation. That is, the historical and observation probabilities are converted to log-odds, and then linear operations are performed in the log-odds space to obtain the fused result. This method is the classic binary Bayesian filtering implementation in log-odds form, which has the advantages of numerical stability and intuitive accumulation, avoiding the product saturation or rounding errors that may occur with simple probability superposition. After obtaining the fused result, it needs to be inversely converted back to probabilistic form to obtain the target probability.

[0084] S104. Attenuate the historical probability of the historical point to obtain the target probability;

[0085] In this application, if a historical point does not have an observation point at the same location, it means that the historical point has not been observed. In this case, the historical probability of the historical point needs to be attenuated in order to obtain the target probability of the historical point after attenuation.

[0086] S105. Based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, generate the historical trajectory point set for the next update cycle.

[0087] In this application, after all fusion and decay operations in the current cycle are completed, a final point set is generated. The data points in this final point set are historical points with target probabilities, as well as newly added observation points from the observation point set. Then, the historical trajectory point set of the previous update cycle is cleared, and the final point set generated at the end constitutes the new historical trajectory point set required for the next update cycle.

[0088] In summary, when updating the occupancy probability, if there are observation points in the same position as historical points, this application performs linear operations on the historical probability and observation probability in the logarithmic probability space, and converts the fused operation result into the target probability. In this way, this application can transform the traditional direct probability operation method into a linear operation in the logarithmic probability space, which is consistent with Bayesian theory, thereby ensuring the stability and accuracy of the fusion process.

[0089] In another embodiment of this application, the process of determining the set of historical trajectory points in the current update cycle includes:

[0090] Obtain the historical trajectory point set generated in the previous update cycle; by simulating the motion state of the current target, adjust the original coordinate values ​​of each historical point in the spatiotemporal map in the historical trajectory point set to obtain the adjusted coordinate values ​​of each historical point; delete historical points whose adjusted coordinate values ​​are less than a predetermined threshold from the historical trajectory point set to obtain the historical trajectory point set of the current update cycle.

[0091] In this application, after obtaining the historical trajectory point set generated in the previous update cycle, it is also necessary to perform coordinate adjustment and pruning operations on each historical point in the historical trajectory point set.

[0092] Specifically, this application first traverses the historical trajectory point set `old_points` and adjusts the coordinates of each historical point in the set. For example, the original coordinate values ​​of a historical point include s-coordinates and t-coordinates. Then, it simulates the motion state of the current target and adjusts the original coordinate values ​​of each historical point in the spatiotemporal graph, such as adjusting them based on `update_s` (spatial update) and `update_t` (temporal update). The current target can be a motor vehicle or a vulnerable traffic participant (such as a pedestrian or non-motorized vehicle). When applying this solution to different targets, only key parameters need to be adjusted according to the motion characteristics of different targets to effectively manage the spatiotemporal occupancy probability of different targets.

[0093] If the current target is a vehicle, simulating the motion state of the current target and adjusting the original coordinate values ​​of each historical point in the spatiotemporal map means: simulating the movement of the vehicle and the passage of time, adjusting the original coordinate values ​​of each historical point. In this application, the adjusted coordinate values ​​of each historical point are called the adjusted coordinate values. After adjustment, the position of each historical point in the spatiotemporal map can be moved to the right as a whole.

[0094] Furthermore, after adjusting the original coordinate values ​​of each historical point, this application also needs to delete historical points whose adjusted coordinate values ​​are less than a predetermined threshold from the historical trajectory point set to obtain the historical trajectory point set for the current update cycle. This predetermined threshold can be set to a small value; in this application, it is set to zero. If the adjusted coordinate value of a historical point is less than zero, that historical point is deleted. For example, if the original coordinate value is s or t, then when s or t is detected to be less than zero, the historical point is removed from old_points.

[0095] In summary, by performing coordinate adjustment and pruning operations on historical points as a preprocessing step, this application ensures that when updating the occupancy probability, only historical data points in the historical trajectory point set that are related to the current target's state and time are considered, thereby improving the efficiency and relevance of subsequent calculations.

[0096] In another embodiment of this application, after determining the new set of observation points in the current update cycle, the method further includes: determining the corresponding original continuous observation number based on the coordinate values ​​of each observation point in the spatiotemporal graph, updating the original continuous observation number to generate the current continuous observation number; and adjusting the observation probability based on the current continuous observation number of each observation point.

[0097] In this application, before the observation points in the new observation point set are formally used for fusion, a confidence adjustment mechanism is used to filter out perceptual noise and unstable "flash" targets. The confidence adjustment mechanism refers to the following: the system maintains a confidence accumulator, which records the number of frames continuously observed at each location in the spatiotemporal map. This is referred to as the continuous observation count. When a new observation point appears at a certain location in the spatiotemporal map, the system updates the cumulative value of the continuous observation count at the corresponding location. When no new observation point appears at a certain location in the spatiotemporal map, the system updates the cumulative value of the continuous observation count at the corresponding location to zero.

[0098] For example, for each cell corresponding to position (s,t) in the spatiotemporal graph, the original continuous observation count of the cell is first obtained. If a new observation point appears in the cell during the current detection period, the original continuous observation count is incremented by one to obtain the current continuous observation count, thereby updating the cumulative value.

[0099] By determining the current number of consecutive observations at each observation point, the observation probability can be adjusted. For example, if the current number of consecutive observations at an observation point is relatively small, it indicates that the observation point is a newly emerging observation point, which may be a "flash" target caused by sensor noise or environmental interference. Therefore, the observation probability of the observation point can be reduced to reduce the impact of the observation point.

[0100] In another embodiment provided in this application, the observation probability is adjusted based on the current number of consecutive observations at each observation point, including:

[0101] Determine if the current number of consecutive observations at the observation point is less than the stability threshold; if so, determine the scaling factor based on the current number of consecutive observations; use the scaling factor to scale the observation probability of the observation point to obtain the adjusted observation probability; if the coordinate value in the spatiotemporal graph does not have a corresponding observation point in the new observation point set, then clear the original number of consecutive observations of the coordinate value to zero.

[0102] See Figure 3The figure illustrates a confidence accumulation mechanism provided in this application embodiment. When processing a new observation point, the confidence accumulator is first updated based on the observation point's location. This involves updating the accumulated value of the corresponding consecutive observations in the confidence accumulator, for example, by adding one to the original consecutive observation count to obtain the current consecutive observation count. Then, it is determined whether the current consecutive observation count is less than a stability threshold. If it is, it indicates a possible interfering target, and the probability needs to be scaled based on the accumulated value of the current consecutive observation count to reduce the impact of the observation point. If the current consecutive observation count is not less than the stability threshold, it indicates that the observation point is a stably occurring target, and the observation probability of the observation point does not need to be adjusted. In this case, the scaling factor can be set to 1.0.

[0103] Furthermore, if the current number of consecutive observations is less than the stable threshold, a scaling factor can be determined based on the current number of consecutive observations so that the observation probability can be dynamically scaled according to the scaling factor. The current number of consecutive observations and the scaling factor are positively correlated, that is, the smaller the current number of consecutive observations, the smaller the scaling factor is set. For example: Set a stability threshold of 4 to determine if the current consecutive observation count of the observation point is less than 4. If not, no adjustment to the observation probability is needed. If the current consecutive observation count is 1, it means that the unit is a first hit. Set the scaling factor corresponding to 1 to a lower factor, such as 0.75, and multiply the observation probability of the observation point by 0.75 to obtain the adjusted observation probability. If the current consecutive observation count is 2, it means that the unit is a second hit. Set the scaling factor corresponding to 2 to a higher factor, such as 0.85, and multiply the observation probability of the observation point by 0.85 to obtain the adjusted observation probability. If the current consecutive observation count is 3, it means that the unit is a third hit. Set the scaling factor corresponding to 3 to a higher factor, such as 0.95, and multiply the observation probability of the observation point by 0.95 to obtain the adjusted observation probability.

[0104] It should be noted that if a coordinate value in the spatiotemporal graph does not have a corresponding observation point in the new observation point set, the original consecutive observation count for that coordinate value will be reset to zero. For example, if a cell in the confidence accumulator is not observed in this frame, the consecutive hits for that cell are interrupted, and the accumulated value of the consecutive observation count will be reset to zero. After probability scaling adjustment, the new observation points need to have their updated probability values ​​updated in the temporary ST raster for subsequent fusion.

[0105] In summary, this application introduces a confidence accumulator, which is a key data structure that can accumulate the observation confidence of each grid cell / location. The accumulated value inside corresponds to the number of consecutive observations of that cell, so as to dynamically adjust the occupancy probability of each new observation point according to the number of consecutive observations. This process is an important guarantee for suppressing sensing noise and improving the robustness of the system.

[0106] Furthermore, this application, through this method, can simulate the deliberate cognitive process of a human driver when faced with uncertain information. This mechanism, by tracking the number of consecutive observations of each spatiotemporal grid / location using a confidence accumulator, can effectively suppress "flash" targets caused by sensor noise or environmental interference, preventing the vehicle from making unnecessary and abrupt deceleration or braking actions due to erroneous "ghost" obstacles. This improves the smoothness and ride comfort of autonomous driving, reduces the risk of rear-end collisions that may result from such behavior, and enhances driving comfort and safety.

[0107] In another embodiment provided in this application, the process of performing linear operations on historical probabilities and observed probabilities in logarithmic probability space to obtain a fused result, and converting the result into a target probability, includes:

[0108] The historical probability of a historical point is converted into the first logarithmic probability, and the observation probability of an observation point is converted into the second logarithmic probability.

[0109] Linear operations are performed using the first log-log odds, the second log-log odds, and the weighting coefficients to generate the fused log-log odds; the fused log-log odds are then converted into the target probability.

[0110] In this application, when a new observation point occupies a historical point, the log-odds fusion process first needs to convert the historical probability of the historical point into the first log-odds and the observation probability of the observation point into the second log-odds. The specific conversion formula is as follows:

[0111]

[0112] In this formula, In is a logarithmic function with the mathematical constant e as the base, p is the probability, and L is the logarithmic odds. If p in the formula is the historical probability of a historical point, then the calculated L is the first logarithmic odds; if p in the formula is the observation probability of an observation point, then the calculated L is the second logarithmic odds.

[0113] After calculating the first and second log-odds, a linear operation can be performed using the first and second log-odds and weighting coefficients to generate the fused log-odds. The specific calculation formula is as follows:

[0114]

[0115] in, The first logarithmic odds generated through historical probability transformation. The second log-probability generated through the new observation probability transformation. These are weighting coefficients set based on the system state. This represents the log-odds ratio after fusion. The weighting coefficient is the weight assigned to the historical log-odds ratio during the log-odds fusion process. Its value determines the relative proportion of historical beliefs and new observational evidence in the updated probability. In this embodiment, the weighting coefficient can be set to 0.5, or other values ​​can be set according to actual needs.

[0116] This application employs a log-odds-based Bayesian filtering method to perform probability fusion. It maps the historical and observed probabilities within the interval [0,1] to the log-odds space of (-∞,+∞), and then fuses the two mapped log-odds. The fusion is represented by a weighted average. This method is the implementation of the classic binary Bayesian filtering in log-odds form. It has the advantages of numerical stability and intuitive accumulation, and avoids the product saturation or rounding error problems that may occur when simply superimposing probabilities.

[0117] After generating the fused log-odds, it is necessary to convert the fused log-odds into the target probability. In this application, the fused log-odds can be inversely converted back to probabilistic form using the Sigmoid function, as shown in the formula:

[0118]

[0119] in, The logarithmic odds after fusion The converted target probability. See also Figure 4a The diagram shows the positive transformation between logarithmic odds and probability provided in the embodiments of this application. Figure 4b The logarithmic probability and inverse probability conversion diagram provided in the embodiments of this application can be used according to... Figure 4a The historical probabilities of historical points are converted into log odds, and the observation probabilities of observation points are converted into log odds; after calculating the fused log odds, according to... Figure 4b Convert to target probability.

[0120] As can be seen, this application's weighted fusion based on Bayesian theory within the logarithmic probability space ensures the mathematical rigor of multi-frame information fusion and fundamentally avoids the numerical saturation and floating-point precision issues of probabilistic operations. Furthermore, since the core parameters are designed as tuning variables with clear physical or probabilistic interpretations and relatively orthogonal functions, the causal relationship between the algorithm's behavior and parameter adjustments is clear and controllable, providing significant reliability and maintainability convenience for engineering deployment. This rigorous fusion framework and interpretable parameter design provided by this application enhance system reliability and maintainability.

[0121] In another embodiment provided in this application, the process of attenuating the historical probability of a historical point to obtain the target probability includes:

[0122] Determine the original number of unobserved historical points; update the original number of unobserved historical points to obtain the current number of unobserved historical points; determine whether the current number of unobserved historical points is greater than the unobserved historical point threshold; if so, set the target probability of the historical point to zero; if not, use a decay factor to decay the historical probability of the historical point to obtain the target probability of the historical point; if the historical point has observation points at the same location, then clear the original number of unobserved historical points to zero.

[0123] In this application, a corresponding miss observer `age` is set for each historical point. The value of the miss observer is the number of unobserved observations. If there is no observation point at the same location as the historical point, it indicates a miss. In this case, the miss count of the observer `age` needs to be incremented, and the original unobserved count in the miss observation counter associated with that historical point is updated incrementally to obtain the current unobserved count for that historical point, so as to accurately record the number of consecutive unobserved frames for the historical point. For example, adding one to the original unobserved data will give the current unobserved count.

[0124] After determining the current number of unobserved instances, it is necessary to compare the current number of unobserved instances with the unobserved threshold (N_decay). This unobserved threshold is used to define the maximum time period during which a historically occupied unit / location can remain unobserved before being cleared by the system. When the number of unobserved instances for any historical point exceeds the N_decay threshold, the occupancy probability of that historical point will be forcibly set to zero. Through this explicit aging mechanism, the system can be provided with fine-grained control over the timeliness of "memory," thereby achieving a key balance between the two opposing requirements of clearing "ghost" obstacles and maintaining target persistence during brief occlusion periods. This effectively improves the robustness of dynamic environment perception. In this embodiment, N_decay can be set to 3, or it can be set to other values, which are not specifically limited here.

[0125] If the number of unobserved instances does not exceed the N_decay threshold, the historical probability of the historical point is exponentially decayed. That is, if the number of unobserved instances of a historical point does not exceed N_decay, the historical probability of the historical point will be exponentially decayed according to a preset decay factor λ, as shown in the following formula:

[0126]

[0127] in, The historical probability of a historical point. The target probability is the result of historical probability decay. The decay factor, which can be preset according to the system state, is used to determine the decay rate of historical probabilities. In this application, the decay factor ranges from [0,1], and its specific value directly determines the "forgetting" rate of historical information. When the value of λ approaches 1, the probability decay rate is slower, and the persistence of historical information is enhanced; conversely, when the value of λ approaches 0, historical information will be cleared more quickly. In this embodiment, λ can be set to 0.5.

[0128] Understandably, if a new observation point occupies a historical point, a state reset operation can be performed. For example, the miss counter (age) corresponding to the cell / location of the historical point can be reset to 0. In this way, the possible time decay process of the cell / location can be interrupted after it has been effectively observed in the current calculation cycle.

[0129] See Figure 5 The following is a flowchart of probability fusion and decay provided in this application embodiment. As shown in the figure, this application needs to traverse each historical point in the historical trajectory point set. For each historical point, based on the position of the historical point / the corresponding cell in the temporary grid, it is determined whether there is a new observation point. Two update logics are executed respectively, as shown in the figure: When processing a single historical point, it is first determined whether there is a new point occupying the position of the historical point; if so, the fusion path is used to merge the historical probability of the historical point and the observation probability of the observation point logarithmically and reset the miss counter age to 0; if not, the decay path is used to increment the miss counter age and determine whether the age setting is greater than N_decay; if so, the historical probability of the historical point is set to 0.0 (cleared); if not, the historical probability of the historical point is exponentially decayed. After each historical point has completed the above process, the processing ends.

[0130] See Figure 6 This is a schematic diagram illustrating the update process of the probability grid provided in an embodiment of this application. Figure 6 It can be seen that the fused state at time t-1 is obtained by fusing the historical state at time t with the current frame observation at time t. The specific process is as follows:

[0131] Fusion: The historical probability of "historical point A" at time t-1 is 0.85. At time t, it is covered by a new observation point "new observation A'", and the probability of "new observation A'" is 0.90. Through the log-probability fusion method of the present invention, its updated probability is 0.88. At the same time, "historical point B" also completes the probability update based on the same principle.

[0132] Decay: The historical probability of "historical point C" at time t-1 is 0.40. Since no new observations are obtained at time t, its probability decays to 0.24 according to the decay mechanism.

[0133] Added: The observation probability of "new observation D'" at time t is 0.65, and it is added to the fused state as a newly emerging target.

[0134] This process fully demonstrates the comprehensive capabilities of this application in processing dynamic information, integrating fusion, attenuation, addition, and removal: it can strengthen confidence by fusing new evidence, and gracefully reduce or forget historical information when evidence is lacking, thereby ensuring the real-time performance, accuracy, and security of the probabilistic raster.

[0135] In summary, in this application, if the miss counter value of a historical point exceeds the threshold N_decay, it indicates that the historical point has not been observed or confirmed for a long time. The system will directly set the historical probability of its corresponding cell / location in the temporary grid to zero, effectively clearing outdated or "ghost" obstacles and preventing them from having a lasting impact on subsequent decisions. This mechanism can both prevent the indefinite existence of "ghost obstacles" and cope with temporary occlusion, thereby directly improving the robustness and safety of autonomous driving environmental perception.

[0136] Furthermore, this application adopts a two-way lifecycle management mechanism that combines the number of unobserved objects and the decay factor, which can handle the entire lifecycle of the target more reasonably and smoothly, ensuring the temporal continuity of environmental representation, avoiding the frequent "appearance" and "disappearance" of the target in the system due to temporary frame loss, and providing more stable data input for downstream planning modules. Through this complete two-way lifecycle management, the temporal continuity of environmental perception is enhanced.

[0137] In another embodiment of this application, the process of generating the historical trajectory point set for the next update cycle based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set specifically includes the following steps:

[0138] The target probability of each historical point is compared with the deletion threshold, and historical points with target probabilities less than the deletion threshold and their corresponding target probabilities are deleted; based on the retained historical points and the newly added observation points in the new observation point set, the historical trajectory point set for the next update cycle is generated.

[0139] In this application, a deletion threshold (FLAGS_p_thd_delete) also needs to be set, which defines a preset probability lower limit. After all fusion and attenuation calculations for the current cycle are completed, each cell can be traversed and filtered according to the deletion threshold. When the historical probability of a historical point is lower than the target probability after attenuation and fusion, the cell will be considered invalid, and its occupancy probability in the temporary grid will be set to zero. This deletion threshold ensures the timely removal of obstacles with low confidence or those that no longer exist, avoiding interference with subsequent decisions. In this embodiment, FLAGS_p_thd_delete can be set to 0.21, or other values, which are not specifically limited here.

[0140] This application uses this method to retain only high-confidence units with an occupancy probability value not lower than the deletion threshold. The system then clears the historical point set from the previous moment, and the newly selected high-confidence occupancy point set forms the new historical trajectory point set required for the next update cycle.

[0141] Furthermore, after attenuating the historical probability of historical points to obtain the target probability, this application can determine whether the target probability is higher than the preset deletion threshold. If it is higher than the deletion threshold, the probability of the corresponding cell in the temporary grid is updated by the attenuated target probability. If it is not higher than the deletion threshold, there is no need to update the probability of the cell, and the probability value of the cell can be directly cleared to zero. There is no need to perform probability modification operation according to the deletion threshold after fusion and attenuation.

[0142] See Figure 7 This document presents a flowchart of a probabilistic raster update method for an ST graph, as provided in an embodiment of this application. In this embodiment, a two-dimensional ST graph, represented as a rasterized map, is used as an example to illustrate the probabilistic update method. As shown in the figure, the process includes a data input stage, a probability update stage, and a result output stage.

[0143] Data input phase: Requires inputting old_points (historical trajectory point set), age (miss count), new_points (new observation point set), confidence accumulator, update_s, and update_t.

[0144] Probability update phase:

[0145] Step 1: Coordinate Adjustment and Pruning: Update the coordinate values ​​of each historical point in the historical point set and remove historical points with coordinate values ​​less than zero to improve the efficiency and relevance of subsequent calculations.

[0146] Step 2: Determine the grid boundary and initialize the temporary grid: Based on all data points in the historical trajectory point set (old_points) and the current new observation point set (new_points), determine the minimum bounding rectangle boundary as the grid boundary and initialize the temporary grid.

[0147] Step 3: Observation confidence adjustment: Update the number of consecutive observations for new observation points. If the number of consecutive observations is less than the stability threshold, scale the observation probability according to the scaling factor.

[0148] Step 4: Historical point fusion and decay: When a historical point is occupied by an observation point, the historical probability and the observation probability are fused using logarithmic probability; when a historical point is not occupied by an observation point, the historical probability of the historical point is adjusted according to the decay factor.

[0149] Step 5: Generate the final point set

[0150] Based on the results of the first four steps, an updated final point set is generated.

[0151] Result Output Stage: At the end of the process, the final updated point set and state of this update cycle are output and used as the historical trajectory set for the next update cycle.

[0152] It should be noted that this application can be applied to the field of autonomous driving, as well as to other robotic systems, intelligent transport systems (ITS), or industrial automation scenarios that require real-time perception and state updates, such as path planning for intelligent warehouse AGVs (Automated Guided Vehicles) and obstacle avoidance for drones. In any scenario involving uncertain occupancy state updates in a dynamic environment, the inventive concept of this application can be used to update the occupancy probability.

[0153] As can be seen from the above embodiments, this application has at least the following technical points:

[0154] 1. Confidence dynamic adjustment method based on continuous observation: This application introduces a confidence dynamic adjustment mechanism, which tracks the number of consecutive hits of each spatiotemporal grid through a confidence accumulator and dynamically scales the initial probability based on this before probability fusion, thereby effectively suppressing perception noise and "flash" targets and improving the robustness of the system.

[0155] 2. Lifecycle management mechanism combining counting and decay: This application proposes a two-way lifecycle management mechanism. On the one hand, it introduces a missed observation counter and threshold to forcibly clear historical information that has not been observed for a long time. On the other hand, it combines a decay factor to perform frame-by-frame probability decay of targets that are temporarily occluded or disappear, simulating gradual "forgetting".

[0156] 3. Rigorous Probabilistic Fusion Framework Based on Log-Odds Space: This application abandons the traditional heuristic direct probability calculation and instead performs weighted fusion within the log-odds space. This method is consistent with Bayesian theory in principle and fundamentally avoids the numerical saturation and floating-point precision problems of probability calculation, ensuring the rigor and robustness of the fusion process.

[0157] 4. Orthogonal and Interpretable Parametric Design: The core parameters of this application, including the fusion weight coefficient (α), decay factor (λ), and unobserved threshold (N_decay), are all designed as tuned variables with clear physical or probabilistic interpretations and relatively orthogonal functions. This design decouples the control of different physical processes, greatly reduces the complexity of algorithm debugging and calibration, improves the understandability and maintainability of the system, and provides significant convenience for the engineering deployment of this safety-critical module. Furthermore, the fusion weight coefficient and decay factor are not limited to pre-set fixed values; adaptive dynamic adjustment or data-driven methods can also be adopted, making the fusion process more intelligent and improving the adaptability of the method.

[0158] See Figure 8 , Figure 8 A schematic diagram of an occupancy probability update device based on a spatiotemporal graph is provided for embodiments of this application. The device specifically includes:

[0159] The determination module 11 is used to determine the historical trajectory point set and the new observation point set in the current update cycle; wherein, the historical trajectory point set includes each historical point with historical probability, and the new observation point set includes each observation point with observation probability;

[0160] The judgment module 12 is used to determine whether each historical point has an observation point in the same position; if so, the fusion module 13 is triggered; if not, the attenuation module 14 is triggered.

[0161] The fusion module 13 is used to perform linear operations on the historical probability and the observed probability in the logarithmic probability space to obtain the fused operation result, and convert the operation result into the target probability;

[0162] The attenuation module 14 is used to attenuate the historical probability of historical points to obtain the target probability;

[0163] The generation module 15 is used to generate a set of historical trajectory points for the next update cycle based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set.

[0164] As an optional embodiment, the determining module includes:

[0165] The acquisition unit is used to acquire the historical trajectory point set generated in the previous update cycle;

[0166] The adjustment unit is used to adjust the original coordinate values ​​of each historical point in the spatiotemporal map of the historical trajectory point set by simulating the motion state of the current target, so as to obtain the adjusted coordinate values ​​of each historical point.

[0167] The deletion unit is used to delete historical points whose adjusted coordinate values ​​are less than a predetermined threshold from the historical trajectory point set, so as to obtain the historical trajectory point set of the current update cycle.

[0168] As an optional embodiment, the generation module is specifically used for:

[0169] The target probability of each historical point is compared with the deletion threshold, and historical points with target probabilities less than the deletion threshold and their corresponding target probabilities are deleted; based on the retained historical points and the newly added observation points in the new observation point set, the historical trajectory point set for the next update cycle is generated.

[0170] As an optional embodiment, the fusion module is specifically used for:

[0171] The historical probability of a historical point is converted into a first logarithmic probability, and the observation probability of an observation point is converted into a second logarithmic probability. A linear operation is performed using the first logarithmic probability, the second logarithmic probability, and the weighting coefficient to generate a fused logarithmic probability. The fused logarithmic probability is then converted into a target probability.

[0172] As an optional embodiment, the attenuation module is specifically used for:

[0173] Determine the original unobserved number of a historical point; update the original unobserved number to obtain the current unobserved number; determine whether the current unobserved number is greater than the unobserved threshold; if yes, set the target probability of the historical point to zero; if no, use a decay factor to decay the historical probability of the historical point to obtain the target probability of the historical point; wherein, if the historical point has observation points at the same position, the original unobserved number of the historical point is cleared to zero.

[0174] As an optional embodiment, the device further includes:

[0175] The update module is used to determine the corresponding original continuous observation number based on the coordinate values ​​of each observation point in the spatiotemporal diagram, and update the original continuous observation number to generate the current continuous observation number;

[0176] The adjustment module is used to adjust the observation probability based on the current number of consecutive observations at each observation point.

[0177] As an optional embodiment, the adjustment module is specifically used to: determine whether the current number of consecutive observations of the observation point is less than a stable threshold; if so, determine a scaling factor based on the current number of consecutive observations; scale the observation probability of the observation point using the scaling factor to obtain the adjusted observation probability; wherein, if the coordinate value in the spatiotemporal graph does not have a corresponding observation point in the new observation point set, the original number of consecutive observations of the coordinate value is cleared to zero.

[0178] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0179] See Figure 9 , Figure 9 A schematic diagram of an electronic device structure provided in this application embodiment includes a processor 21, a communication interface 22, a memory 23 and a communication bus 24, wherein the processor 21, the communication interface 22 and the memory 23 communicate with each other through the communication bus 24;

[0180] Memory 23 is used to store computer programs;

[0181] When the processor 21 executes the program stored in the memory 23, it implements the steps of the occupancy probability update method described in any of the above method embodiments, which will not be repeated here.

[0182] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0183] The communication interface is used for communication between the aforementioned terminal and other devices.

[0184] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0185] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0186] In another exemplary embodiment, a mobile tool is also provided, including the electronic device described in the above embodiments.

[0187] The mobile vehicle can be any device with mobility capabilities, including vehicles with autonomous or intelligent driving capabilities (including passenger vehicles (such as cars, buses, coaches, minibuses, etc.), cargo vehicles (such as ordinary trucks, box trucks, trailer trucks, enclosed trucks, tank trucks, flatbed trucks, container trucks, dump trucks, special structure trucks), special vehicles (such as logistics delivery vehicles, automated guided vehicles (AGVs), patrol vehicles, cranes, excavators, bulldozers, loaders, road rollers, off-road engineering vehicles, armored engineering vehicles, sewage treatment vehicles, sanitation vehicles, vacuum trucks, floor scrubbers, water sprinkler trucks, sweeping robots, food delivery robots, shopping guide robots, lawnmowers, golf carts, etc.), recreational vehicles (such as amusement vehicles, amusement park autonomous driving devices, balance bikes, etc.), rescue vehicles (such as fire trucks, ambulances, power repair vehicles, engineering emergency rescue vehicles, etc.)) and robots (such as sweeping robots, food delivery robots, etc.).

[0188] In another exemplary embodiment, a computer storage medium is also provided, wherein the program instructions, when executed by a processor, implement the steps of the occupancy probability update method described in any of the above method embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0189] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0190] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0191] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for updating occupancy probability based on a space-time graph, characterized in that, include: Obtain the historical trajectory point set generated in the previous update cycle; by simulating the motion state of the current target, adjust the original coordinate values ​​of each historical point in the spatiotemporal graph in the historical trajectory point set to obtain the adjusted coordinate values ​​of each historical point; From the set of historical trajectory points, delete historical points whose adjusted coordinate values ​​are less than a predetermined threshold to obtain the set of historical trajectory points in the current update cycle; wherein, the set of historical trajectory points includes each historical point with historical probability; Determine the set of historical trajectory points and the set of new observation points in the current update cycle; wherein, the set of new observation points includes each observation point with an observation probability; Based on the coordinates of each observation point in the spatiotemporal graph, the corresponding original continuous observation count is determined and updated to generate the current continuous observation count. It is then determined whether the current continuous observation count for each observation point is less than a stable threshold. If so, a scaling factor less than 1 is determined based on the current continuous observation count, and the observation probability of each observation point is scaled using this scaling factor to reduce the observation probability, resulting in an adjusted observation probability. If the coordinates in the spatiotemporal graph do not correspond to any observation point in the new observation point set, the original continuous observation count for the coordinates is cleared to zero. Determine whether each historical point has an observation point in the same location; If so, the historical probability and the observed probability are linearly operated on in the logarithmic probability space to obtain the fused operation result, and the operation result is converted into the target probability. At the same time, the number of unobserved points corresponding to the historical points is cleared to zero. If not, the historical probability of the historical point is attenuated to obtain the target probability, including: determining the original number of unobserved instances of the historical point; updating the original number of unobserved instances to obtain the current number of unobserved instances; determining whether the current number of unobserved instances is greater than the unobserved threshold; if yes, the target probability of the historical point is set to zero; if no, the historical probability of the historical point is attenuated using an attenuation factor to obtain the target probability of the historical point. Based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, the historical trajectory point set for the next update cycle is generated. The weighting coefficients in the linear operation and the attenuation factor used in the attenuation can be adaptively and dynamically adjusted according to the system state, including the planning interaction state, or adjusted based on a data-driven strategy.

2. The method of claim 1, wherein, The adaptive dynamic adjustment based on system states, including the planning interaction state, includes: obtaining the vehicle's planning input acceleration; comparing the planning input acceleration with the acceleration range of the obstacle prediction trajectory; if the planning input acceleration exceeds the acceleration range, determining that the current interaction state is planning-driven; in the planning-driven interaction state, adjusting the weight coefficient or the decay factor to reduce the weight of historical probability in the linear operation, or accelerating the decay of historical probability at historical points.

3. The method of claim 1, wherein, The method of attenuating the historical probability of a historical point to obtain the target probability further includes: obtaining the time index value of the historical point in the spatiotemporal graph; determining the time-related attenuation compensation amount based on the time index value; wherein, the larger the time index value, the larger the time-related attenuation compensation amount; and using the time-related attenuation compensation amount to further correct the target probability of the historical point, so that the probability attenuation rate of the far-term time region in the spatiotemporal graph is greater than the probability attenuation rate of the recent time region.

4. The method of claim 1, wherein, Based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set, a historical trajectory point set for the next update cycle is generated, including: The target probability of each historical point is compared with the deletion threshold, and the historical points whose target probability is less than the deletion threshold and their corresponding target probabilities are deleted. Based on the retained historical points and the newly added observation points in the new observation point set, a historical trajectory point set for the next update cycle is generated.

5. The method of claim 1, wherein, The historical probability and the observed probability are linearly operated on in logarithmic probability space to obtain the fused result, and the result is then converted into the target probability, including: The historical probability of a historical point is converted into the first logarithmic probability, and the observation probability of an observation point is converted into the second logarithmic probability. Linear operations are performed using the first log-log odds, the second log-log odds, and the weighting coefficients to generate the fused log-log odds. The fused logarithmic odds are converted into the target probability.

6. An occupancy probability update device based on a spatiotemporal graph, characterized in that, include: The determination module is used to determine the historical trajectory point set and the new observation point set in the current update cycle; wherein, the historical trajectory point set includes each historical point with historical probability, and the new observation point set includes each observation point with observation probability; the judgment module is used to determine whether each historical point has an observation point in the same position; if yes, the fusion module is triggered; if no, the decay module is triggered; the determination module includes: an acquisition unit, used to acquire the historical trajectory point set generated in the previous update cycle; an adjustment unit, used to adjust the original coordinate values ​​of each historical point in the historical trajectory point set in the spatiotemporal map by simulating the motion state of the current target, to obtain the adjusted coordinate values ​​of each historical point; and a deletion unit, used to delete historical points from the historical trajectory point set whose adjusted coordinate values ​​are less than a predetermined threshold, to obtain the historical trajectory point set of the current update cycle; The fusion module is used to perform linear operations on historical probabilities and observed probabilities in logarithmic probability space to obtain the fused operation result, and convert the operation result into the target probability. At the same time, the number of unobserved points corresponding to the historical points is cleared to zero. The update module is used to determine the corresponding original continuous observation number based on the coordinate values ​​of each observation point in the spatiotemporal diagram, and update the original continuous observation number to generate the current continuous observation number; The adjustment module is used to adjust the observation probability based on the current number of consecutive observations at each observation point; determine whether the current number of consecutive observations at the observation point is less than a stable threshold; if so, determine a scaling factor less than 1 based on the current number of consecutive observations; use the scaling factor to scale the observation probability of the observation point to reduce the observation probability, and obtain the adjusted observation probability; wherein, if the coordinate value in the spatiotemporal graph does not have a corresponding observation point in the new observation point set, the original number of consecutive observations of the coordinate value is cleared to zero. The attenuation module is used to attenuate the historical probability of a historical point to obtain the target probability, including: determining the original number of unobserved instances of the historical point; updating the original number of unobserved instances to obtain the current number of unobserved instances; determining whether the current number of unobserved instances is greater than the unobserved threshold; if so, setting the target probability of the historical point to zero; if not, attenuating the historical probability of the historical point using an attenuation factor to obtain the target probability of the historical point. The generation module is used to generate a set of historical trajectory points for the next update cycle based on each historical point, the target probability of each historical point, and the newly added observation points in the new observation point set; wherein, the weight coefficients in the linear operation and the decay factor used in the decay can be adaptively and dynamically adjusted according to the system state, including the planning interaction state, or adjusted based on a data-driven strategy.

7. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the occupancy probability update method according to any one of claims 1 to 5.

8. A mobile tool, characterized in that, Includes the electronic device as described in claim 7.

Images