Method for modeling tactical environment of motor vehicle
By constructing a tactical environment modeling method around motor vehicles, using perceptual data and geometric modeling modules to evaluate the state of the passing lane unit, introducing neutral and safe states, the problem of insufficient description of the state of occupancy grid unit in the prior art is solved, and the problem of insufficient description of the state of the occupancy grid unit in the existing technology is realized. Environmental modeling suitable for tactical decision-making and the effect of reducing accident risk.
Patent Information
- Application Number
- CN202280090447.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2021-12-22
- Filing Date
- 2022-12-22
- Publication Date
- 2024-11-01
AI Technical Summary
The existing space modeling method around motor vehicles is difficult to apply to tactical decisions, and the state description of the occupied grid cells is insufficient, so it cannot effectively support mid-term decisions.
A tactical environment modeling method is adopted to construct a geometric model of the passing lane through the perceptual device group and geometric modeling module, decompose it into units of the same length, and evaluate the occupancy state of the unit based on the perceptual data, introduce neutral and safe states, enrich the Unit description.
It provides a rich description of the state of occupancy grid unit suitable for tactical decisions, improves the applicability and accuracy of environmental modeling, reduces the risk of accidents, and ensures the integrity of information received by the decision module.
Smart Images

Figure CN118891489A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for modeling a motor vehicle environment. The invention also relates to a device for modeling a motor vehicle environment. Background Art
[0002] Smart or autonomous vehicles need to constantly make decisions based on the situation they find themselves in. In the literature, three levels of decision making by autonomous cars are distinguished:
[0003] - first level decisions, called operational decisions, which involve decisions made within a very limited first time window, for example in the order of milliseconds to 1 second,
[0004] - a second level of decision making, known as tactical decision making, which involves medium-term decisions, such as those made within a time interval greater than one second and limited to approximately 10 seconds,
[0005] - The third level of decision making, known as strategic decision making, involves long-term decisions, such as decisions made over a period of time greater than 10 seconds relative to the present moment.
[0006] Operational decisions are made at high frequency based on a fine and accurate representation of the near environment in which the vehicle is located. Strategic decisions, for their part, are made at low frequency and require only a high granularity representation of the vehicle's environment over a range of perhaps more than a kilometer. Both types of environmental representations, fine representations for operational decisions or high granularity for strategic decisions, are already widely used. On the other hand, environmental representations dedicated to tactical decision making are less prevalent. The present invention proposes to remedy this.
[0007] From the prior art, in particular from US 2018 / 0322777 A1, solutions are known that implement occupancy grids of the space around a motor vehicle. However, these grids have disadvantages:
[0008] - they model areas of fixed size and do not adapt the modeling to the environment in which motor vehicles travel, or
[0009] - The cells that make up the grid can only take three states: vacant, occupied, or undefined. Summary of the invention
[0010] The object of the present invention is to provide a device and a method for modeling the tactical environment of a motor vehicle, which overcomes the above-mentioned disadvantages and improves the device and method for modeling the tactical environment of a motor vehicle. In particular, the present invention allows the realization of such a device and a method, which are simple and reliable and make the environment modeling suitable for the traffic environment of the vehicle and provide a rich description of the state of the cells of the occupied grid, which is particularly suitable for tactical decision making.
[0011] To this end, the present invention relates to a method for modeling a tactical environment of a first automated motor vehicle, the tactical environment being indicative of environmental data of the automated motor vehicle capable of influencing tactical decision making involving mid-term decisions, the first automated motor vehicle comprising:
[0012] - sensor set,
[0013] -Decision-making module, and
[0014] - a module for geometric modeling of the set of traffic lanes to be considered by the decision module,
[0015] The traffic lane group includes a first sequence lane that directly interacts with the lane in which the first automated motor vehicle travels according to an interaction pattern defined in a predefined interaction pattern group, and a second sequence lane that interacts with the first sequence lane according to an interaction pattern defined in a predefined interaction pattern group,
[0016] The predefined interaction mode group includes at least one interaction mode of the automatic motor vehicle remaining in its traffic lane, the automatic motor vehicle changing lanes to a lane adjacent to its traffic lane, the automatic motor vehicle's lane merging with another traffic lane, and the automatic motor vehicle's lane being crossed by at least one other traffic lane.
[0017] Additionally, modeling methods include:
[0018] - In the first step, the geometric modeling module constructs the geometric models of the first sequence and the second sequence of traffic lane groups, and decomposes these traffic lanes into groups of units of the same length,
[0019] - a second step of evaluating, based on data from the set of perception means, a first occupancy state of each unit of the set of units as a vacant state, an occupied state, or an uncertain state, the uncertain state being assigned to units that are outside the field of view of the set of perception means and / or to units that are at least partially obscured by obstacles that are located between the first automated motor vehicle and the unit,
[0020] - A third step, based on the data from the perception device group, by analyzing the traffic located on the second sequence lane, defining the second occupancy state of each designated unit whose first occupancy state is uncertain as neutral or safe, which comprises the sub-steps of: assigning a neutral state to the designated unit if a second vehicle different from the first automatic motor vehicle and traveling on the second sequence lane occupies the first unit located at the intersection of the first sequence lane and the second sequence lane and downstream of the designated unit in the direction of travel of the first sequence lane,
[0021] - A fourth step is sending to the decision module the environment model including the association of the occupancy state with each cell of the geometric model, the occupancy state of the cell being the second state if defined, otherwise the occupancy state of the cell being the first state.
[0022] A third step, after the sub-step of assignment of the neutral state, may comprise a sub-step of assigning a safe state to a designated unit, comprising the determination of an emergency stop zone associated with each vehicle surrounding the first automated motor vehicle.
[0023] The identification of emergency stop zones may include:
[0024] - Determine the possible trajectory set of surrounding vehicles, then
[0025] -For each possible trajectory,
[0026] * Calculate the stopping distance of surrounding vehicles and
[0027] * Identify a cluster of neighboring cells that are on the possible trajectory, in front of the surrounding vehicles and at a distance less than the stopping distance of the surrounding vehicles,
[0028] The emergency stop zone for the surrounding vehicles is determined as the intersection of the neighboring cell cluster associated with each possible trajectory, and then
[0029] - assigning the safe state to each cell of the emergency stop zone whose second occupation state differs from the neutral state.
[0030] The second step of evaluating the first state of each cell of the group of cells may comprise superimposing the group of cells and an allocation map allocating the space covered by the field of view as vacant areas, occupied areas or uncertain areas, the allocation map originating from the group of sensing devices.
[0031] The status of a unit can be:
[0032] - is occupied, if a first number of points of cells located in one or more occupied areas of the allocation map is greater than a first specified threshold, otherwise
[0033] is empty, if a second number of points of cells located in one or more occupied areas of the allocation map is greater than a second specified threshold, otherwise
[0034] -uncertain.
[0035] The first and second specified thresholds may be an absolute value of a minimum number of points, or a minimum percentage of the total number of points contained in a cell.
[0036] The allocation map may include a representation of the space surrounding the first automated motor vehicle represented by polygons, an occupancy grid, or a difference map.
[0037] The sensing device may provide sensing data and a confidence index associated with the sensing data, and a first step of constructing a geometric model may include:
[0038] - receiving sensory data from the sensory device, and a confidence index associated with the sensory data,
[0039] - calculate the sub-step of the optimized discretization step size according to the confidence index,
[0040] -Transfer the optimized discretization step size to the substep of the geometric modeling module, and then
[0041] - A sub-step of receiving a geometric model from the geometric modeling module, the geometric model comprising a group of first and second sequence lane groups decomposed into cells of optimized length.
[0042] The sub-step of calculating the optimized discretization step size may include:
[0043] - minimize the number of negative false cells that are evaluated as being in a vacant, neutral or safe state, while it is actually occupied,
[0044] And optionally,
[0045] - Minimize the number of positive false cells that are evaluated as being in an occupied state when it is actually in a vacant state, a neutral state, or a safe state.
[0046] The invention furthermore relates to a device for modeling the environment of a first automated motor vehicle equipped with autonomous movement control means and a decision module. The device comprises hardware and / or software elements for implementing the method defined above.
[0047] The invention also relates to a computer program product comprising program code instructions recorded on a computer-readable medium for implementing the steps of the method defined above when said program is run on a computer. The invention also relates to a computer program product downloadable from a communication network and / or recorded on a computer-readable and / or computer-executable data medium, comprising instructions which, when the program is executed by a computer, direct the computer to implement the method defined above.
[0048] The invention also relates to a computer-readable data recording medium having recorded thereon a computer program comprising program code instructions for implementing the method defined above. The invention also relates to a computer-readable recording medium comprising instructions, which, when executed by a computer, direct the computer to implement the method defined above.
[0049] The invention also relates to a data carrier signal carrying a computer program product as defined above. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The figures show, by way of example, one embodiment of a device for modeling an environment of a motor vehicle and one execution of a method for modeling an environment of a motor vehicle according to the invention.
[0051] Figure 1 Represents a motor vehicle equipped with modeling equipment for the environment.
[0052] Figure 2 It is an alternative representation of a motor vehicle equipped with modeling equipment for the environment.
[0053] Figure 3 An example of modeling traffic lanes to be considered for tactical decision making for a motor vehicle is shown.
[0054] Figure 4 A first embodiment of a representation of the space occupancy around a motor vehicle is shown.
[0055] Figure 5 A second embodiment of a representation of the space occupancy around a motor vehicle is shown.
[0056] Figure 6 is a flow chart of an execution manner of the method for modeling an environment according to the present invention.
[0057] Figure 7 A first modeling mesh of a traffic lane generated at a first discretization step of decreasing length is shown.
[0058] Figure 8 A second modeling mesh of the traffic lane generated with a second discretization step size that is larger than the first discretization step size is shown.
[0059] Fig. 9 The merging process between the traffic lane modeling grid and the line of sight of the motor vehicle's perception device group is shown.
[0060] Fig.10 The merging process between the traffic lane modeling grid and the occupancy grid representation of the data from the perception device is shown.
[0061] Fig.11 The merging process between the traffic lane modeling mesh and a polygonal representation of data from a perception device is shown.
[0062] Fig.12 Classification of first and second sequence lanes of interest according to one of vacant, occupied, and uncertain states is shown.
[0063] Fig.13 A first portion of the traffic lane is shown, the state of which is neutral, and a second portion of the traffic lane is shown, the state of which is safe. DETAILED DESCRIPTION
[0064] Example reference to a motor vehicle 100 of an embodiment of a modeling apparatus for a tactical environment of a motor vehicle Figure 1 Described later.
[0065] The motor vehicle 100 may be any type of motor vehicle, in particular a coach, a van, a truck or a public transport vehicle, such as a bus or a shuttle. According to the described embodiment, the motor vehicle 100 is an autonomous vehicle and will be indicated as such in the following description.
[0066] The illustration is therefore given in a non-limiting manner. In particular, the motor vehicle may be a non-automatic vehicle equipped with a driver assistance system, in particular a driver assistance system corresponding to automation greater than or equal to level 2, that is to say corresponding to partial automation of the vehicle.
[0067] In the remainder of this document, the term "tactical environment of the automated vehicle 100" refers to environmental data of the automated vehicle 100 that may affect tactical decision making. This may involve environmental data that may affect decisions made within a time interval greater than 1 second and limited to a dozen seconds.
[0068] The automatic vehicle 100 moves in a specified environment. The automatic vehicle 100 includes a driving assistance system 70 and an automatic control device 20 of the automatic vehicle 100 .
[0069] The driving assistance system 70 sends a movement control C to the automatic control device 20. The movement control C may include a longitudinal movement control and / or a lateral movement control. The longitudinal movement control may include a first torque command for an engine group of the vehicle and / or a second torque command for a vehicle brake actuator. The lateral movement control includes a rotation angle of a steering wheel of the automatic vehicle 100.
[0070] The driving assistance system 70 comprises in particular a modeling system 10 of the tactical environment of the automated vehicle 100 and a decision module 3. The role of the system 10 is to construct a modeling M_ENV_T of the tactical environment of the automated vehicle, which modeling will be sent to the decision module 3. This modeling M_ENV_T enables the decision module 3 to make one or more tactical decisions. These decisions will then be converted into movement control C of the automated vehicle 100.
[0071] The system 10 for modeling the tactical environment of the autonomous vehicle 100 includes, in particular, the following elements:
[0072] - a data storage device in which is recorded a high-resolution map including a description of the road network on which the automated vehicle 100 is travelling Figure 1 ,
[0073] - a geo-positioning system 2 which enables Figure 1 On the positioning automatic vehicle 100,
[0074] - a building block 4 of a geometric representation of the lane of interest of the automated vehicle 100,
[0075] - a device 5 for sensing the environment of the autonomous vehicle 100,
[0076] A computing unit 6 comprising a microprocessor 61 , a local electronic memory 62 and a communication interface 63 , enabling the microprocessor 61 to receive data from the geographic positioning system 2 , the geometrical modeling module 4 and the perception means 5 and to transmit data to the geometrical modeling module 4 .
[0077] land Figure 1 Provides geometric, topological, and semantic descriptions of road networks.
[0078] In particular, the Figure 1 It contains the information required to determine, from the position of the automatic vehicle 100 determined by the geolocation system 2, all possible paths whose starting point is determined by said position.
[0079] land Figure 1 The described road network consists of roads broken down into traffic lanes. The traffic lanes themselves are broken down into lane segments; Figure 1 Associate an identifier with each lane segment.
[0080] In the remainder of this document, the term "lane" or "traffic lane" refers to the Figure 1 Defines a unique identifier associated with a lane segment.
[0081] Geo-locating the automated vehicle 100 by the geo-locating system 2 includes determining an identifier of the lane on which the automated vehicle 100 is traveling, and the vehicle's position on that lane segment.
[0082] Figure 2 A high level description of the process implemented in the driver assistance system 70 regarding tactical level decision making is provided. The process is broken down into perception X1 of the environment of the automated vehicle 100, followed by structuring of the perception X2, followed by decision making based on the structured perception X3.
[0083] Therefore, based on the perception X1 of the tactical environment of the automatic vehicle 100, in particular when receiving the data D-ENV from the perception device group 5, the driving assistance system 70 creates a structured model M_ENV_T of the tactical environment of the automatic vehicle 100 and sends the structured model M_ENV_T to the decision module 3.
[0084] The construction of the structured model M_ENV_T includes the following two steps:
[0085] - A first step of creating, by module 4, a geometric description LGM of a set of navigation lanes that may play a role in the making of tactical decisions,
[0086] - A second step of constructing the environment data D_ENV by the microprocessor 61 from the LGM geometric description.
[0087] The geometric description or LGM grid created by module 4 is a non-central and discrete representation (in the form of a grid consisting of a series of interconnected cells) of the navigation lane in the length direction and along the navigation lane.
[0088] In alternative embodiments not described herein, the lane may also be decomposed into cells in the width direction. In this case, the LGM grid will be a non-central and discrete representation (a grid form consisting of a series of interconnected cells) of the navigation lane in the length and width directions.
[0089] The allocentric representation uses a globally fixed reference frame, in contrast to the egocentric representation which uses a reference frame that moves with the vehicle.
[0090] Each cell of the LGM grid extends over a length equal to the discretization step size PDIS and over the full width of the navigation lane. The length of the cell is taken into account in the direction of travel on the traffic lane in question. Therefore, all cells of the LGM grid have the same length, equal to the PDIS step size. Under an embodiment, the fixed step size PDIS value is determined according to the accuracy of the perception device group 5 and / or the accuracy of the geolocation device 2, in particular the fixed step size value is included between 10 centimeters and 500 centimeters. The fixed step size PDIS value may be included in other value ranges.
[0091] The discretization step size PDIS can also be modified to optimize the reliability of the environmental data associated with each cell, in particular the reliability of the occupancy state of each cell, as described later in this paper.
[0092] Figure 3 An example of an LGM grid associated with an automated vehicle 100 is provided. The right-pointing arrow indicates the path that the automated vehicle 100 will follow. In this case, the LGM grid includes modeling of the following lanes, corresponding to the lanes of interest in the current situation of the automated vehicle 100:
[0093] - the lane 151 on which the automated vehicle 100 is travelling and the lane 152 which it will follow after turning right,
[0094] a lane 153 adjacent to the lane of travel of the automated vehicle 100 , which is located immediately to the left and in the same direction of travel as the lane 151 ,
[0095] - Lane 154 coming from the left, which will cross the path of the automated vehicle 100 .
[0096] Other navigation lanes are not modeled in the LGM grid because they have no impact on the decision made by the automated vehicle 100 at that given moment.
[0097] According to a preferred embodiment of module 4, lanes of interest may be of the first or second sequence. In this embodiment, a set of possible roads for the automatic vehicle 100 is determined within a specified time range or distance range based on the specified position of the automatic vehicle 100. Each possible road is then considered to determine the first and second sequence lanes of interest:
[0098] - a first sequence of lanes is defined as directly interacting with the possible road according to one of a set of predefined interaction modes, and
[0099] - A second sequence lane is defined to directly interact with a first sequence lane according to an interaction mode defined in a set of predefined interaction modes.
[0100] In one embodiment, the set of predefined interaction modes may include:
[0101] - the automated vehicle 100 remains in its lane of travel, and / or,
[0102] - a lane change of the automated vehicle 100 to a lane adjacent to its lane of travel, and / or
[0103] - a merging of the lane of automated vehicle 100 with another lane of traffic, and / or
[0104] The lane of automated vehicle 100 is crossed by at least one other traffic lane.
[0105] In one embodiment, the microprocessor 61 controls the construction of the LGM when transmitting a construction request REQ(LOPT) to the module 4, which request may include the optimized discretization step size LOPT. This will be explained in detail later in this article.
[0106] The perception device 5 may include any perception device capable of characterizing the spatial occupancy of the environment, such as one or more radars and / or cameras and / or lidars mounted on the automatic vehicle 100. The perception device 5 has a field of view, which is represented by the field of view FOV (the acronym for Field Of View) in the rest of this document. The field of view FOV is defined by the range of the perception device 5. In the rest of this document, points or areas outside the field of view are referred to as unobservable. According to this definition, an area can be observable but obscured, for example, by an obstacle (vehicle, building, etc.) located between the perception device group 5 and the area.
[0107] Within the field of view FOV, data from a set of perception devices 5 provides an operational representation of the occupancy of the space surrounding the automated vehicle 100.
[0108] The representation of the surrounding space occupancy can take different forms. Some representations decompose the space around the automated vehicle 100 into occupied, vacant, or uncertain regions as described, including polygonal representations, representations in the form of probabilistic or explicit occupancy grids, or difference map representations.
[0109] When the occupancy of the surrounding space is represented by polygons, each polygon determines the outline of a region that can be defined by a state. Each segment of the polygonal outline can also provide a state information (especially when a segment delimits an object of a certain type).
[0110] Occupancy grid representation uses a two-dimensional grid to grid the surrounding space, the cell size is fixed and usually quite small. The space bounded by each cell is characterized by one or more information, such as occupancy state and / or occupancy probability and / or the height of the object occupying the cell.
[0111] Representations of disparity maps are constructed based on images from a pair of stereo cameras. These representations are presented in the form of a three-dimensional grid of disparity images (inversely proportional to depth).
[0112] Figure 4 An example of a space occupation represented by polygons 40 , 41 , 42 is shown, with polygon 41 being the complement of polygons 40 and 42 . Figure 5 An example of spatial occupancy represented by an occupancy grid 51 is shown.
[0113] The information from the sensing means 5 thus allows the space to be classified into said vacant areas, said occupied areas and said uncertain areas. An area is called vacant if it does not contain any obstacles; it is called occupied if it includes at least one obstacle; it is called uncertain if it is outside the field of view FOV or it is at least partially obscured by obstacles. For example, in Figure 4 In FIG. 4 , polygon 40 represents an occupied area, polygon 41 represents a blocked area and polygon 42 represents a vacant area.
[0114] The driver assistance system 10 can also include a communication network 7, such as a CAN network, of the automated vehicle 100. The communication network 7 can in particular provide the microprocessor 61 with movement data of the automated vehicle 100 and surrounding vehicles, such as speed, acceleration and jerk.
[0115] The microprocessor 61 can also access the shared database 8 via the cellular network, which includes contextual information such as map updates, work zone notifications, etc. . . .
[0116] In an embodiment of the invention, the computer 61 allows the execution of software comprising the following modules communicating with each other:
[0117] a module 611 for constructing a geometric model of a set of lanes, which communicates with the geometric modeling module 4 and the set of perception devices 5,
[0118] a module 612 for evaluating a first occupancy state of each cell of the geometric model, which communicates with the set 5 of sensing means,
[0119] A module for defining a second occupancy state of the unit 613 , which communicates with the group of communication means 5 and the communication network 7 .
[0120] - A transmission module 614 of the environment model, which communicates with the decision module 3 .
[0121] In a first step E1 , a geometric model or LGM mesh of a set of first and second sequences of lanes ENS_VOIES is constructed, the lanes being decomposed into a set of cells ENS_C of equal length.
[0122] The first step E1 of constructing the geometrical model comprises a sub-step E14 of receiving the LGM geometrical model from the geometrical modelling module 4 .
[0123] In the rest of this paper, this geometric model is also referred to as the LGM mesh.
[0124] In one embodiment, the geometric modeling module periodically sends updates to the geometric model.
[0125] In one embodiment, the receiving sub-step E14 may be preceded by the following sub-steps:
[0126] - Sub-step E11, receiving the sensing data and the confidence index IC associated with the sensing data from the set of sensing means 5,
[0127] - Sub-step E12, calculating the optimized discretization step size POPT according to the confidence index IC,
[0128] - Sub-step E13, transmitting the optimized discretization step size POPT to the geometric modeling module 4.
[0129] The length of the cell, that is, the sampling size of the traffic lane, is a key parameter in constructing the M_ENV_T tactical environment model.
[0130] In fact, as described later in this paper, the construction of the tactical environment model M_ENV_T is based on characterizing the state of each cell of the LGM grid as uncertain, vacant, occupied, neutral or safe (the latter two states are determined by analyzing the flow of traffic located on the second lane of the LGM grid). Knowledge of the state of the cells allows the automated vehicle to make decisions while minimizing the risk of accidents. Therefore, the error rate in the information provided by the LGM grid must be below a maximum error threshold.
[0131] However, as will be seen later in the description of steps E2 and E3, the state of the cell is defined on the basis of data from the perception means 5, these data advantageously including a confidence index IC which varies according to the perception conditions. It is therefore essential that the size of the cell, that is to say the sampling size of the traffic lane, is adapted to the confidence index of the data from the perception means 5.
[0132] However, the confidence index may be undefined or wrong. In this case, there is a risk of choosing a discretization step size that is too precise compared to the localization uncertainty, which increases the probability of assigning wrong states to cells. For example, too many false negatives may be obtained, that is, cells that are wrongly evaluated as empty, which increases the risk of collisions.
[0133] One solution is to increase the discretization step size of the LGM grid, since the rules for determining the state of the LGM grid cells (these rules are described in detail in the description of steps E2 and E3) will have the effect of evaluating the state of the cells in a more pessimistic way. In other words, due to the rules applied to the calculation of the cell state, increasing the discretization step size reduces false negatives, that is to say the detection of false vacant cells, and increases the surface area of the areas that are mistakenly considered to be occupied, due to the increase in the cell length.
[0134] Figure 7 and 8 The effect of adjusting the discretization step size is illustrated. The first figure shows a first mesh LGM 1 generated with a first discretization step size P1 of decreasing length. A vehicle 101 located to the left of the automated vehicle 100 occupies a first series of cells 71 extending over a first length L1.
[0135] The same scene in Fig. 9 It is shown that the second grid LGM2 is generated with a second discretization step size P2 (greater than the step size P1) for generating the second grid LGM2. The vehicle 101 then occupies a second series of cells 72 having a length L2 greater than the length L1.
[0136] If the positioning of the vehicle 101 is accurate, the first step length P1 is more appropriate than the second step length P2. On the other hand, if the positioning of the vehicle 101 is not very accurate, the second step length P2 is more appropriate than the first step length P1.
[0137] Therefore, in a second sub-step E12 , an optimized value POPT of the discretization step size is calculated, taking into account both the uncertainty index associated with the sensing means 5 and the maximum percentage of cells that are falsely considered empty.
[0138] In one embodiment, the optimization value POPT may be a configuration variable of the method, which has been previously determined by simulation.
[0139] An example of a simulation method capable of calculating the optimized value POPT is described below:
[0140] - First, record a large dataset from the vehicle’s sensors in order to obtain a sufficiently large number of samples to be as representative as possible of the situations encountered by the vehicle,
[0141] -Secondly, the data is replayed by simulation to evaluate the process of assigning states to each cell of the LGM grid; then, while changing the discretization step size, the function of the number of false negatives as a function of the discretization step size is obtained.
[0142] - then a target false negative rate that must not be exceeded is set; the false negative rate is strictly reduced as the discretization step size increases, and a binary method can be used to find the minimum discretization step size that can meet the maximum false negative rate.
[0143] Then, in a third sub-step E13 , the optimized value POPT of the discretization step size is sent to the module 4 .
[0144] Then, in a third sub-step E14 , the LGM grid is received from the module 4 with a discretization step equal to the optimization value POPT.
[0145] Then, a step E2 is performed to determine the first occupancy state of each unit of the unit group ENS-C as a vacant state, an occupied state or an uncertain state, based on the data from the set of sensing means 5 .
[0146] Fig. 9 A first merging process E21 between the LGM grid coming from the module 4 and the data coming from the set of sensing devices 5 is shown, which consists in determining a first state of certain cells of the LGM grid according to the field of view FOV of the set of sensing devices 5 .
[0147] according to Fig. 9 According to the logic shown, any cell 91, 92 of the LGM grid that is not completely within the field of view FOV is considered to be completely unobservable and its first state is uncertain because it is not known whether it is completely vacant or at least partially occupied.
[0148] In a second process E22, the space covered by the field of view FOV is of interest. The set of sensing devices provides a map 103 of the space allocated as vacant, occupied or uncertain areas, which may be obscured by obstacles (e.g. cars or buildings). Process E22 consists in merging the cells of the LGM grid with the map 103.
[0149] Fig.10 is a first illustration of a second merging process, wherein the allocation map 103 is represented by an occupancy grid. The second process E22 is applied to cells A1 to A6 of the LGM grid.
[0150] All points of cell A1 are superimposed on the vacant area of graph 103: cell A1 is therefore considered vacant. Some points of cell A2 are superimposed on the occupied area of graph 103 while other points of cell A2 are vacant: cell A2 is therefore considered fully occupied. Some points of cell A3 are superimposed on the uncertain area of graph 103 while other points of cell A3 are vacant: cell A3 is therefore considered fully uncertain.
[0151] In the second merge process E22, the logic described in Table 1 is thus applied to combine the states of the first and second regions of the allocation map 103, which are included in the same cell of the LGM grid.
[0152] [Table 1]
[0153]
[0154] Fig.11 is a second illustration of the second merging process E22, wherein the representation map 103 is represented by occupied polygons. The merging process is applied to cells B1 to B6 of the LGM grid.
[0155] All points of cell B1 are superimposed on the vacant area of Figure 103: cell B1 is therefore considered vacant. Some points of cell B2 are superimposed on the uncertainty area of Figure 103, while other points of cell B2 are vacant: cell B2 is therefore considered completely uncertain. The treatment of cell B3 is the same as that of cell B2. Cell B4 is superimposed on the occupied, vacant and uncertain areas of Figure 103: cell B4 is therefore considered completely occupied.
[0156] This description is not intended to be limiting, and in particular other embodiments of FIG. 103 are possible.
[0157] In one embodiment, the status of a unit of the LGM may be:
[0158] - occupied, if a first number of points of cells located in one or more occupied regions of the allocation map 103 is greater than a first specified threshold MIN_OCC, otherwise
[0159] - vacant, if a second number of points of cells located in one or more vacant regions of the allocation map 103 is greater than a second specified threshold MIN_LIB, otherwise
[0160] - Uncertain.
[0161] Furthermore, the first and second specified thresholds MIN_OCC, MIN_LIB may correspond either to an absolute value of the minimum number of dots, or to a percentage of the minimum number of dots to the total number of dots contained in the cell.
[0162] For example, if the threshold MIN_LIB is set to 90% and if the threshold MIN_OCC is set to 15%, the rules to be applied may be defined as follows:
[0163] - If a cell contains at least 15% of the points in the occupied state, its state is occupied, otherwise
[0164] - if it contains at least 90% of the points of the vacant state, then its state is vacant, otherwise,
[0165] - otherwise its status is undefined.
[0166] In one embodiment, a distinction may be made between cells of the LGM grid whose uncertain state corresponds to the "obstacle blocked" state and cells of the LGM grid whose uncertain state corresponds to the "outside the field of view FOV" state.
[0167] Fig.12 The first and second sequence lanes of interest are classified into one of three states: vacant, occupied, and uncertain. Note that the cells of the LGM grid are not Fig.12 only portions of the first and second sequence lanes are shown.
[0168] exist Fig.12 In the driving situation shown, the automated vehicle 100 approaches the entrance to the roundabout. It moves on a route including the entrance to the outer lane of the roundabout, in particular on the through lane V0.
[0169] The first sequence and second sequence of lanes associated with the route of the automated vehicle 100 include:
[0170] - a sequence lane V11, which corresponds to a portion of the roundabout outer lane located upstream of the automatic vehicle 100 entering the roundabout,
[0171] - a second sequence lane V21, which is a roundabout entrance lane, adjacent to the through lane V0 and merging toward the inner lane of the roundabout,
[0172] The second lane V22 of the second sequence corresponds to a portion of the lane within the roundabout, connects with the lane V21 at point 122 , and the lane V22 is located upstream of the point 122 .
[0173] The second vehicle 121 travels on the second sequence lane V21 parallel to the automated vehicle 100. The presence of the second vehicle determines a portion of the lane V220 in an occupied state, the occupied lane portion defining an area of the ground occupied surface including the second vehicle 121.
[0174] Moreover, in Fig.12 In the configuration shown, the presence of the second vehicle 121 does not allow, based on data from the perception device of the automated vehicle 100, to determine whether the first sequence lane portion V11 is vacant or occupied; similarly, it does not allow to determine whether the second sequence lane portion V22 is vacant or occupied. The lane portions V21 and V22 are obscured and therefore in an uncertain state.
[0175] exist Fig.12 In the case shown, other lane portions of the lane of interest are vacant.
[0176] Therefore, at the end of step E2, the cells of the first and second sequence lanes of interest are determined to be in a vacant, occupied or uncertain state.
[0177] A third step E3 is then performed, based on data from the perception device group 5, when a second vehicle 101, which is different from the automatic vehicle 100 and travels on the second sequence lane, occupies a first unit located at the intersection of the first sequence lane and the second sequence lane and downstream of the designated unit along the travel direction of the first sequence lane, a second occupancy state is defined for each designated unit whose first occupancy state is uncertain, which includes a sub-step E31 of assigning a neutral state to the designated unit.
[0178] In other words, in step E3, attention is paid to the elements whose state cannot be determined by processing the data coming from the set of perception means 5. For this purpose, use is made of the data relating to the first and second sequences of lanes of interest.
[0179] In particular, in a first sub-step E31 , the neutral state is determined for the cells of the first sequence of lanes of the LGM grid, which cells were assigned an uncertain state in step E2 because they were obscured by the second vehicle 121 inserted between them and the automated vehicle 100 .
[0180] In the remainder of this document, the term “neutral” characterizes the state of a cell of the LGM grid whose traffic cannot interfere with the trajectory of the automated vehicle 100 .
[0181] For example, in Fig.13In the traffic configuration shown, the position of the second vehicle 131 prevents any vehicle located in the portion V131 of the first sequence lane V11 from intruding into the traffic lane of the automatic vehicle 100. In other words, the unit of the lane portion V131 located upstream of the vehicle 131 is in a neutral state due to the presence of the vehicle 131 located on both the first sequence lane V11 and the second sequence lane V21.
[0182] Sub-step E31 thus comprises
[0183] - detecting a second vehicle 131 different from the automated vehicle 100 and traveling on a second sequence of designated lanes, occupying the first cell located at the intersection of the first sequence of lanes and the second sequence of designated lanes;
[0184] - determining a set of cells located upstream of the first cell and whose first occupancy state is indeterminate,
[0185] - determining that the second state of each cell of the group of cells is neutral.
[0186] Furthermore, after sub-step E31 , the third step E3 may comprise a sub-step E32 of assigning a safe state to a designated unit whose state is uncertain, which comprises determining an emergency stop zone associated with each dynamic object around the automated vehicle 100 .
[0187] In fact, certain cells of the LGM grid, whose state remains uncertain after the end of sub-step E31 , may be assigned a safe state if they are located in the emergency stop zone of the vehicles surrounding the automated vehicle 100 .
[0188] Sub-step E32 consists in determining a set of vehicles ENS_VE surrounding the automated vehicle 100. In one embodiment, the determination of the set of surrounding vehicles ENS_VE is performed on the field of view FOV. Alternatively, the determination of the set of ENS_VE can be performed on an area included in the field of view FOV.
[0189] Then for each specified vehicle V in the set ENS_VE i , determine the specified vehicle V i Possible trajectory T ij Group ENS_T i , especially in terms of road infrastructure.
[0190] Then, for each possible trajectory T ij , calculate vehicle V i In the trajectory T ij Stop distance D_STOP ij To this end, in one embodiment, the use of data from the sensor set 5 and / or the communication network 7 allows the determination of the vehicle V iThe instantaneous speed of the vehicle V i The current position of each trajectory T ij , so determine the adjacent unit group G ij , which is located on the trajectory T ij On the vehicle V i In front and with the vehicle V i The distance is less than the vehicle V i Stop distance D_STOP ij .
[0191] Advantageously, with the vehicle V i Associated emergency stop zone Z_STOP i It is then determined to be respectively associated with each trajectory T ij The associated adjacent cell group G ij The intersection of.
[0192] Finally, a safe state is assigned to any cell of the LGM grid.
[0193] - its status is undefined after the end of sub-step E31, and
[0194] a vehicle V which is completely included in the group of vehicles ENS_VE surrounding the automated vehicle 100 i Emergency stop area Z_STOP i middle.
[0195] In other words, only when the cell of the LGM grid is located at the vehicle V i The intersection of all possible trajectories, and any point of the unit with the vehicle V i The distance is less than the allowed vehicle V i The LGM grid cell can only be considered as the shortest stopping distance of the trajectory reaching the point due to the vehicle V i existence and safety.
[0196] exist Fig.13 In the driving situation shown, the group of surrounding vehicles ENS_VE includes two vehicles V1 and V2. Only one trajectory can be considered for vehicle V1. Associated with this trajectory is a stopping distance D_STOP11, which allows to define a group of cells G11 covering the stopping distance D_STOP11. The cells of group G11 in the first uncertain state will then adopt the second safe state.
[0197] Likewise, only one trajectory can be considered for the vehicle V2. Associated with this trajectory is a stopping distance D_STOP21, which allows defining a group of cells, group G21, covering the stopping distance D_STOP21. The cells of group G21 in the first uncertain state will then adopt the second safe state.
[0198] Then the fourth step E4 is performed to transmit the environment model M_ENV_T to the decision module 3, which includes the association of the occupancy state with each unit of the LGM geometric model. If it is defined in step E3, the occupancy state of the unit is the second state, otherwise the occupancy state of the unit is the first state defined in step E2.
[0199] Thus, in step E4 , the model M_ENV_T is constructed by associating an enrichment state with each cell of the LGM grid. In fact, as shown in Tables 2 and 3 below, the processing performed in steps E1 to E3 allows enriching the information coming from the set of sensing means 5 .
[0200] Table 2 shows the first embodiment described above, where the same uncertain state is assigned to cells whose state is unknown, either because they are outside the FOV or because they are obscured by obstacles.
[0201] [Table 2]
[0202]
[0203] Table 3 shows a second implementation, where the uncertain state is decomposed into two different sub-states:
[0204] - the "unobservable" state assigned to cells located outside the field of view FOV,
[0205] - A shielded state, in which the unit is at least partially shielded by an obstacle located between the automated vehicle 100 and the unit.
[0206] [Table 3]
[0207]
[0208] After determining the enrichment state of each cell according to one of the described embodiments, the LGM grid is transmitted to the decision module 3 .
[0209] The method according to the invention differs from the prior art, in particular document US2018 / 0322777A1, in which the occupancy grid is defined only on a simple space of interest corresponding to the space in front of the automated vehicle, and the road topology is not taken into account when constructing this grid. In fact, first of all, the modeling method according to the invention takes into account the geometry of the navigation lanes. Moreover, the LGM grid includes not only the navigation lanes on which the automated vehicle travels, but also lanes that may interact with the lanes of the automated vehicle. Therefore, unlike the prior art, the method according to the invention is not limited to the modeling of lanes facing automated vehicles.
[0210] The modeling method according to the invention thus has several advantages. Firstly it provides a compact representation of the environment with which the vehicle interacts, since only the lane of interest is modeled. This involves a dependent representation of the environment, since it focuses on the areas that constrain tactical decision making.
[0211] Furthermore, the modeling according to the invention enriches the perception data, in particular regarding the units that are not observable by the perception device, by interpreting the traffic around the automated vehicle 100. This interpretation process allows enriching the state of the unit with two new states, a neutral state and a safe state, in order to allow the decision module to exploit these states.
[0212] Moreover, the modeling takes into account the uncertainties in localization and perception while ensuring the integrity of the information provided to the decision-making module, in particular by adjusting the discretization step size of the LGM grid according to the confidence index of the perception data.
[0213] Furthermore, the modeling approach according to the invention is able to take into account different types of representations of perceptual data.
[0214] Finally, due to its geometric nature, the modelling according to the invention simplifies the subsequent processing performed in the decision module, in particular by facilitating the calculation of risk measures such as, for example, the delay before collision.
Claims
1. A method of modeling a tactical environment of a first automated motor vehicle (100), the tactical environment indicating environmental data of the automated motor vehicle (100) that can influence tactical decision making involving mid-term decisions, the first automated motor vehicle (100) comprising: - a sensing device group (5), - a decision module (3), and - a module for geometric modeling (4) of the set of traffic lanes (ENS_VOIES) to be considered by the decision module (3), The traffic lane group (ENS_VOIES) includes: a first sequence lane (V11), which directly interacts with the lane through which the first automatic motor vehicle (100) passes according to an interaction mode defined in a predefined interaction mode group; and a second sequence lane (V21, V22), which interacts with the first sequence lane (V11) according to an interaction mode defined in the predefined interaction mode group. The predefined interaction mode group includes at least one interaction mode of the automated motor vehicle (100) staying in its lane, the automated motor vehicle (100) changing lanes to a lane adjacent to its lane, the lane of the automated motor vehicle (100) merging with another lane, and the lane of the automated motor vehicle (100) being crossed by at least one other lane. The modeling method comprises: - In the first step (E1), the geometric modeling module (4) constructs the geometric model (LGM) of the traffic lane groups (ENS_VOIES) of the first sequence (V11) and the second sequence (V21, V22), and decomposes these traffic lanes into groups of units of the same length (ENS_C), - a second step of evaluating, based on data from the set of sensing means (5), a first occupancy state of each cell of the set of cells (ENS_C) as a vacant state, an occupied state, or an uncertain state, the uncertain state being assigned to cells located outside the field of view (FOV) of the set of sensing means (5) and / or to cells at least partially obscured by obstacles located between the first automated motor vehicle (100) and the cells, - A third step (E3), based on the data from the sensing device group (5), by analyzing the traffic located on the second sequence lane, defines the second occupancy state of each designated unit whose first occupancy state is uncertain as neutral or safe, which includes the following allocation sub-step (E31): if a second vehicle (131) different from the first automatic motor vehicle (100) and traveling on the second sequence lane (V21) occupies the first unit located at the intersection of the first sequence lane (V11) and the second sequence lane (V21) and downstream of the designated unit along the travel direction of the first sequence lane, the neutral state is allocated to the designated unit, - A fourth step (E4) of sending to the decision module (3) an environment model (M_ENV_T) comprising an association of an occupancy state with each cell of the geometric model (LGM), the occupancy state of the cell being the second state if defined, otherwise the occupancy state of the cell being the first state.
2. The modeling method according to claim 1, characterized in that The third step (E3), after the sub-step (E31) of assigning the neutral state, comprises a sub-step (E32) of assigning the safe state to the designated unit, which comprises determining an emergency stop zone associated with each vehicle (Vi) surrounding the first automated motor vehicle (100). (Z_STOPi), determine the emergency stop zone (Z_STOPi) including: - Determine the set (ENS_Ti) of possible trajectories (Tij) of surrounding vehicles (Vi), then -For each possible trajectory (Tij), oCalculate the stopping distance (D_STOPij) of the surrounding vehicles (Vi) and o determining a neighboring cell group (Gij) that is located on a possible trajectory, in front of a surrounding vehicle (Vi) and at a distance from the surrounding vehicle that is less than the stopping distance (D_STOPij) of the surrounding vehicle, The emergency stop zone (Z_STOPi) of the surrounding vehicles (Vi) is determined as the intersection with the neighboring cell cluster associated with each possible trajectory (Tij), and then - The safe state is assigned to each cell of the emergency stop zone (Z_STOPi) whose second occupied state differs from the neutral state.
3. Modeling method according to one of the preceding claims, characterized in that A second step (E2) of evaluating a first state of each cell of the set (ENS_C) of cells comprises superimposing the set (ENS_C) of cells with a map (103) for assigning the space covered by the field of view (FOV) as vacant, occupied or uncertain areas, the map (103) originating from the set (5) of sensing means, characterized in that the state of the cell is - is occupied, if a first number of points of cells located in one or more occupied areas of said allocation map (103) is greater than a first specified threshold (MIN_OCC), otherwise - empty, if a second number of points of cells located in one or more empty regions of the allocation map (103) is greater than a second specified threshold (MIN_LIB), otherwise -uncertain, And it is characterized in that the first and second specified thresholds (MIN_OCC, MIN_LIB) can be the absolute value of the minimum number of dots, or the minimum percentage of the total number of dots contained in the unit.
4. The modeling method according to claim 3, characterized in that The allocation map (103) includes a representation of the space surrounding the first automated motor vehicle (100) represented by polygons, an occupancy grid, or a difference map.
5. Modeling method according to one of the preceding claims, characterized in that The sensing device (5) provides sensing data and a confidence index (IC) associated with the sensing data, and is characterized in that a first step (E1) of constructing a geometric model comprises: - a sub-step (E11) of receiving, from said set of sensing means, sensing data and a confidence index (IC) associated with said sensing data, - a sub-step (E12) of calculating the optimized discretization step size (POPT) according to the confidence index (IC), - passing the optimized discretization step size (POPT) to a sub-step (E13) of the geometric modeling module (4), followed by - A sub-step (E14) of receiving a geometric model from said geometric modeling module (4), said geometric model comprising a first and a second sequence of lane groups (ENS_VOIES) decomposed into a group (ENS_C) of cells of optimized length (POPT).
6. The modeling method according to claim 5, characterized in that The sub-step (E12) of calculating the optimized discretization step size (POPT) includes - minimize the number of negative false cells that are evaluated as being in a vacant, neutral or safe state, while it is actually occupied, and optionally, - Minimize the number of positive false cells that are evaluated as being in an occupied state when it is actually in a vacant state, a neutral state, or a safe state.
7. A modeling device (10) for the environment of a first automatic motor vehicle (100), the first automatic motor vehicle being equipped with a motion autonomous control device (20) and a decision module (3), the modeling device comprising hardware and / or software elements (1, 2, 4, 5, 6, 7, 8, 61, 62, 63, 611, 612, 613) for implementing a modeling method according to one of claims 1 to 6.
8. A computer program product, comprising program code instructions recorded on a computer-readable medium, for implementing the steps of the method according to any one of claims 1 to 6 when the program is run on a computer.
9. A computer-readable data recording medium having recorded thereon a computer program including program code instructions for implementing the method according to any one of claims 1 to 6.
10. Data carrier signal carrying a computer program product according to claim 8.
