Method for randomly selecting a number of trajectory candidates for a vehicle

Abstract

Method for the random selection of a number of trajectory candidates for a vehicle, in which in a first step a reference line is created from discrete support points, a reference longitudinal movement is created based on the created reference line and then a speed curve is obtained by means of a quasi-stationary simulation and the selection is made based on the created reference longitudinal movement and the obtained speed curve.

Description

[0001] The invention relates to a method for randomly selecting a number of trajectory candidates for a vehicle, in particular for a motor vehicle, and an arrangement for carrying out the method. State of the art

[0002] A fundamental distinction is made between assisted, automated, and autonomous driving. Vehicles with assisted or automated systems can perform steering, acceleration, and braking maneuvers along and across the lane without human intervention. Autonomous driving is also referred to as highly automated or fully automated driving. In this context, the term "automated driving" is generally used, which is intended to encompass all different levels of automation.

[0003] In automated driving, a planning module creates a trajectory to describe the movement to be made. A trajectory, also known as a path curve, describes a spatial curve along which a body, in this case a vehicle, moves.

[0004] The created trajectory is passed to controllers that are to follow it. A popular method for determining an optimal trajectory is to randomly create several trajectory candidates and then select the best one. A sample can consider a final state defined by position, orientation, velocity, and acceleration.

[0005] In a known method for creating a trajectory, an end time is first specified, at which the final state is to be reached. The remaining states are set depending on the mode. During free travel without a preceding vehicle, various final speeds are specified based on the target speed. During following a preceding vehicle, the so-called target object, the randomly selected final positions and final speeds are selected depending on the predicted position of the target object at the end time.

[0006] One feature of trajectory creation is the planning horizon, which is limited to a few seconds. However, additional information is usually available, for example, through the use of a map. This information can describe the route, speed limits, or even the movement of the target object. Furthermore, a higher-level strategic or tactical planning algorithm can specify specific points, such as a stopping point where a standstill is achieved. Disclosure of the invention

[0007] Against this background, a method having the features of claim 1 and an arrangement according to claim 9 are presented. Embodiments emerge from the dependent claims and from the description.

[0008] The presented method is used for the random selection of a number of trajectory candidates for a vehicle. In a first step, a reference line is created from discrete sampling points. Based on the created reference line, a reference longitudinal movement is created. A velocity profile is then obtained using a quasi-stationary simulation. The selection is then made based on the created reference longitudinal movement and the obtained velocity profile.

[0009] In quasi-stationary simulation, a speed profile is imposed on a reference line, assuming a static world. This takes longitudinal and lateral dynamics, curve progression, speed limits, stopping points, and target objects into account. Forward and reverse simulation is used to optimally connect the start and target states. A target object is included via a simulation of an ACC controller.

[0010] A trajectory candidate is a possible trajectory that is proposed for selection to become the trajectory that is implemented.

[0011] This results in a focused creation of a trajectory by using longitudinal motion models and controllers to obtain a predictive reference motion.

[0012] The presented method is based on the investigation of the following problems, which are examined in more detail below: (1) Focused sampling A practical limitation of sample-based trajectory generation is the available computing power. The more trajectory candidates that can be generated and evaluated, the better the resulting trajectory. The presented method enables targeted or focused sampling, so that only potentially useful trajectory candidates need to be generated and considered. This can either reduce computing power, allowing the use of cheaper hardware, or allow a better trajectory to be found. (2) Prospective sampling The described method is based on the creation of a reference movement, which, due to its low computational effort, can extend beyond the planning horizon for trajectory creation. This allows the future road layout, future speed limits, and movements of the target object to be taken into account in the samples. This is particularly advantageous or even necessary at high speeds, since a comfortable speed adjustment, for example, to reach a standstill at a specified stopping point, is not feasible within the planning horizon of the trajectory. (3) No mode change Previous methods use different modes for sample selection. To ensure a high level of comfort for passengers, the transition between modes should be as unnoticeable as possible, i.e., smooth. For other road users, fundamentally similar behavior between different modes is required for good interaction with the automated vehicle. The presented method enables fundamentally identical sample selection for journeys with and without a target object. The selection of a mode is thus eliminated.

[0013] In the presented method, a reference longitudinal movement is created based on a reference line using models and controllers for specifying a longitudinal movement. This is used in the sample selection to achieve the three advantages mentioned above. The dynamic driving limits of the vehicle, the curvature of the reference line, the speed limits, stopping points, and, if applicable, a target object are taken into account.

[0014] The described arrangement is configured to carry out the presented method and has an evaluation unit for this purpose. The arrangement can be implemented in hardware and / or software. Furthermore, the arrangement can be integrated into a control unit of a vehicle, in particular a motor vehicle, or can be designed as such a control unit.

[0015] Further advantages and embodiments of the invention will become apparent from the description and the accompanying drawings.

[0016] It is understood that the features mentioned above and those to be explained below can be used not only in the combination specified in each case, but also in other combinations or on their own, without departing from the scope of the present invention. Short description of the drawings Fig. 1 shows two graphs of speed over a distance along a path. Fig. Figure 2 shows a GV diagram in two representations. Fig. 3 shows a flow chart of a possible sequence of the presented procedure. Fig. Figure 4 shows two graphs showing speed curves over time. Fig. Figure 5 shows exemplary speed curves for different final speeds in a graph. Fig. 6 shows a vehicle with an object driving ahead in a highly simplified, purely schematic representation. Embodiments of the invention

[0017] The invention is illustrated schematically in the drawings using embodiments and is described in detail below with reference to the drawings.

[0018] The proposed method requires a reference line consisting of discrete nodes. The following information should be provided for each node: position coordinates, orientation, curvature, and distance along the reference line from the starting position. This is optionally provided for each speed limit. If a stopping point is present, the speed limit at the corresponding node is set to zero.

[0019] First, the quasi-stationary simulation is performed. For this, only the curvature and distance are required. An exemplary curvature curve is shown above in Fig. 1 shown.

[0020] Fig. 1 shows in a first graph 10, on whose abscissa 12 the distance along path s in [m] and on whose ordinate 14 the curvature ĸ in [1 / m] is plotted, a curve 16 to illustrate the curvature over the distance for the quasi-stationary simulation.

[0021] A second graph 30, whose abscissa 32 represents the distance along path s in [m] and whose ordinate 34 represents the velocity v in [m / s], shows a first curve 36 and a second curve 38. The first curve 36 illustrates the steady-state velocity limited to 20 m / s. The second curve 38 shows the velocity from the quasi-steady-state simulation.

[0022] Quasi-steady simulation is used in racing to estimate lap times. It determines how fast a vehicle can travel along a track while taking into account limited longitudinal and lateral acceleration. The acceleration limits are usually speed-dependent and can be represented as a speed-vs-vs diagram. Such a diagram shows Fig. 2.

[0023] Fig. 2 shows a GCF diagram. On the left side, in a first Fig. a coordinate system shown with the axes left turn a y 52, braking 54, right turn 56 and driving a x 58. The speed v is plotted on another axis 60. A point 62 illustrates the maximum speed v max .

[0024] On the right side of the Fig. 2 shows a graph on whose abscissa 72 the lateral acceleration a yand at its ordinate 74 the longitudinal acceleration a x is applied.

[0025] The creation of a velocity curve in the quasi-stationary simulation can in principle, as shown in the flow chart of the Fig. 3 is shown, look like: According to Fig. 3, the vertices of the curvature are determined in a first step 100. These are the local maxima of the absolute curvature.

[0026] In a second step 102, the simulation begins with an acceleration phase. The starting speed for the first interpolation point is given. The lateral acceleration is calculated from the speed and the curvature. If the lateral acceleration is greater than the maximum permitted lateral acceleration at the current speed, the program continues to step 104. Otherwise, the maximum possible longitudinal acceleration at the current speed and for the existing lateral acceleration is calculated. The determined longitudinal acceleration is used to calculate the speed at the next interpolation point using a position-discrete acceleration step. The longitudinal acceleration is limited if necessary so as not to exceed a given speed limit. If the longitudinal acceleration adjusted in this way is negative and therefore a deceleration, the program continues to step 104.Otherwise, the next sampling point is considered and the process starts again. This process runs forward along the reference line. In a third step 104, the system switches to the deceleration phase. For this, the system jumps to the next vertex. If no vertex is present, the last sampling point is selected. The maximum possible speed is calculated for the current sampling point. The acceleration limits are observed.

[0027] In a fourth step 106, the braking process runs backward along the reference line. The speed is given for the current sampling point. The lateral acceleration is calculated. The maximum possible deceleration (negative longitudinal acceleration) is determined for the lateral acceleration and the current speed. This deceleration is then applied to perform a backward integration step and calculate the speed of the previous sampling point. The longitudinal acceleration / deceleration is adjusted if necessary so as not to exceed a given speed limit. If the sampling point was already visited during the acceleration phase and the speed from the braking phase is not higher than the speed from the acceleration phase, the backward movement is aborted. The program jumps to the sampling point considered in step 104 and switches to step 102.

[0028] The quasi-stationary simulation results in a velocity curve as shown in the lower graph 30 with reference number 38 in Fig. 1. A time curve can be calculated from the speed and distance curve.

[0029] The speed curve from the quasi-stationary simulation takes into account the curvatures in the reference line as well as speed restrictions and thus serves as a forecast for the route.

[0030] To also consider a target object, adaptive cruise control (ACC) is simulated. This essentially works as follows: (1) The start is at the starting position. The speed and starting time are given. (2) The predicted state of the target object is determined for the current simulation time. (3) From the state of the simulated ego vehicle, namely pose, speed, and that of the target object, namely pose, speed, acceleration, etc., the target longitudinal acceleration of the ego vehicle is determined by means of a controller for an ACC. (4) The target longitudinal acceleration is limited by the GV diagram. Furthermore, it is adjusted so that the velocity of the quasi-steady simulation is not exceeded at the next sampling point when integrating to the next sampling point. (5) If no next sampling point is available or other termination conditions are met, the process is aborted here. Otherwise, an integration step is performed to the next sampling point of the reference line. As with quasi-stationary simulation, the integration is performed discretely in position rather than discretely in time. (6) The integration time is determined. This determines the simulation time for the next sampling point. The next sampling point is switched to. The state of the ego vehicle, e.g., pose, is determined from the reference line. The process continues with (2).

[0031] A stopping point can either be controlled by a separate controller or be considered as a stationary phantom object in the ACC controller.

[0032] From the simulation of the ACC controller, curves for time, speed, and acceleration are now available. One such curve is shown in Fig. 4 shown.

[0033] Fig. 4 shows in a first graph 200, on whose abscissa 202 the time t ACCin [s] and on whose ordinate 204 the speed v in [m / s] is plotted, a first curve 206 and a second curve 208. The first curve 206 shows the speed curve of a target object. The second curve 208 shows the speed curve of the following vehicle. The illustration illustrates following a target object that initially travels at a constant speed, then decelerates, and then continues traveling at a constant speed.

[0034] The ACC controller attempts to follow the target object within the specified time gap. When the target object brakes, the speed is reduced and adjusted to its speed.

[0035] In a second graph 230, on whose abscissa 232 the time t ACC in [s] and at its ordinate 234 the longitudinal acceleration in [m / s 2], a first curve 236 and a second curve 238 are shown. The first curve 236 shows the longitudinal acceleration profile of the target object, and the second curve 238 shows the longitudinal acceleration profile of the following vehicle.

[0036] The curves from the quasi-stationary simulation and the simulation of the ACC controller can now be used as a reference for a focused sample selection. This particularly applies to samples for longitudinal motion, i.e., speed and acceleration. This is exemplified in Fig. 5 shown.

[0037] Fig. 5 shows, in a graph 300, on whose abscissa 302 the distance in [m] and on whose ordinate 304 the speed in [m / s] is plotted, exemplary speed curves for different final speeds with longitudinal accelerations adapted to the reference curve.

[0038] A first curve 310 shows a reference curve, a second curve 312 shows a curve for v e = 10.0 m / s, a third curve 314 shows a course for v e = 9.0 m / s, a fourth curve 316 shows a course for v e = 8.0 m / s, a fifth curve 318 a course for v e = 7.0 m / s, a sixth curve 320 a course for v e = 6.0 m / s, a seventh curve 322 a course for v e = 5.0 m / s, an eighth curve 324 a course for v e = 4.0 m / s, a ninth curve 326 a course for v e = 3.0 m / s, a tenth curve 328 a course for v e = 2.0 m / s, an eleventh curve 330 a course for v e = 1.0 m / s and a twelfth curve 332 a course for v e = 0.0 m / s

[0039] Various randomly generated speed curves are shown, with the final speed and final acceleration selected based on the reference. Especially during braking, i.e., braking before a curve or when approaching a slow target object, selecting a longitudinal acceleration other than zero is useful and is enabled by the longitudinal motion reference.

[0040] Furthermore, the longitudinal movement reference enables an estimate of the earliest time at which a certain position along the reference line can be reached.

[0041] Fig. 6 shows a purely schematic representation of a vehicle 400 and a preceding object 402, which is also referred to as a target object and is typically also a vehicle.

[0042] To implement the method, a reference characteristic curve 410 is first created, based on which a reference longitudinal movement 412 is created. A velocity profile 414 is then created using a quasi-stationary simulation. Based on the created reference longitudinal movement 412 and the velocity profile 414, the selection is made, and a number of trajectory candidates 416 are obtained. From this number of trajectory candidates 416, a trajectory 418 can be selected, which is then passed on to a controller 420 for implementation. A suitable evaluation unit 422 is provided to implement the method.

Claims

[1] Method for the random selection of a number of trajectory candidates (416) for a vehicle (400), in which in a first step a reference line (410) is created from discrete support points, a reference longitudinal movement (412) is created based on the created reference line (410) and then a speed profile (414) is obtained by means of a quasi-stationary simulation and the selection is made based on the created reference longitudinal movement (412) and the obtained speed profile (414). [2] Method according to claim 1, wherein, if a preceding object (402) is detected, object tracking is carried out in a second step. [3] Method according to claim 1 or 2, in which a future road course, future speed limits and movements of the preceding object (402) are taken into account. [4] Method according to claim 3, in which a distance control cruise control is simulated. [5] Method according to claim 4, in which curves for time, speed and acceleration are obtained by simulating the adaptive cruise control. [6] Method according to one of claims 1 to 5, in which a trajectory (418) is selected from the number of trajectory candidates (416) and is passed on to controllers (420) which are to follow it. [7] Method according to one of claims 1 to 6, in which the following information is given for each support point: position coordinates, orientation, curvature and distance along the reference line (410) starting from a starting position. [8] Method according to one of claims 1 to 7, in which driving dynamic limits of the vehicle (400), a curvature of the reference line (410), speed limits and stopping points are taken into account. [9] Arrangement for randomly selecting a number of trajectory candidates (416) for a vehicle (400), with an evaluation unit (422) which is designed to carry out a method according to one of claims 1 to 8.

Images