Method for controlling an at least partially automated vehicle

Abstract

Method for controlling an at least partially automated vehicle to avoid and / or align the vehicle in a collision situation, wherein the vehicle is controlled based on a scalar constellation parameter

Description

The present invention relates to a method for controlling an at least partially automatically operated vehicle for avoiding or aligning the vehicle in collision situations.Vehicles operated at least partially in an automated manner are part of the descriptions of future mobility scenarios. An at least partially automatedly operated vehicle is a vehicle in which all or at least parts of the driving task are taken over by vehicle systems. If the entire driving task is taken over, this is referred to as a fully automated or highly automated vehicle. The vehicle drives in an automated manner, for example by independently detecting the course of the road, other road users or obstacles and calculating the corresponding control commands at least partially in the vehicle and forwarding them to the actuators in the vehicle, as a result of which the course of the vehicle is influenced accordingly. A human driver is no longer involved in the driving task in a fully automated or highly automated vehicle.From the standpoint of vehicle safety, the introduction phase of automated and highly automated vehicles in mixed traffic is of particular interest. Many engineers worldwide are working to prevent accidents in the future from occurring due to at least partially automated driving functions and highly automated vehicles. However, these at least partially automated vehicles of the future will move long in mixed traffic with a majority of conventional vehicles. Passive safety will therefore also remain important for highly automated vehicles, since accidents must still be expected in mixed traffic, for example if, as before, a conventional vehicle with a human driver causes an accident. In the urban sector, accidents in the crossing region with crossing and turning conventional vehicles are of particular interest here in automated vehicles with priority.Prior ArtVarious documents are concerned with the mitigation of collisions. The transition from the driving maneuver for mitigation is usually discrete in the prior art, i.e. the documents are concerned either with planning driving maneuvers or with maneuvers for mitigation of collisions. The documents mentioned below often find a block diagram in which the crash avoidance is firstly operated, a determination is made according to whether the collision is unavoidable and, if so, various collision severity values for possible trajectories are subsequently determined and a decision is then made.According to DE 2012 110 21004, the method starts directly upon the detection of an unavoidable collision. Details of the decision making are not disclosed because they represent a complex process.It is interesting in DE 10 2014 218 549 that the aspect of crash severity determination is detailed using a method of hybrid artificial intelligence. The finding of the inducibility of the collision as entry criterion, as is used in the prior art, means, due to situational blurring and deficiencies of sensing (narrow sensing) and recognition (narrow perception) in reality, that the system only carries out the evaluation of the collision position too late in many collisions, since the situation was not yet clearly recognizable as being unavoidable. An obvious solution would be work with probabilities and different scenarios. This leads to increased computing effort and complex handling of the case distinction within the algorithm, as described, for example, in WO 09 09 09 09 2374 A1.Advantages of the InventionAgainst this background, the present invention provides a method which represents a uniform and nevertheless practicable planning and triggering of the response patterns of emergency braking and avoidance maneuvers and an alignment (pre-crash alignment; CAL) preceding a collision for influencing the imminent collision position in intersection situations.The invention is based on the realization that the determination of whether collision avoidance is still possible in the present collision situation and, if not, how a suitable orientation of the vehicle has to look at the imminent collision can be mapped within an algorithm by mapping to a scalar constellation parameter.In the present case, a collision situation can be understood to mean a traffic situation in which two road users would collide with one another with unchanged continuation of their movements.A collision situation typically exists between the vehicle and a collision partner. The collision partner may likewise be a vehicle in the present case.The present invention has particular features over the prior art, these being:• Simultaneity: The intervention planning of the reaction patterns collision avoidance and alignment takes place in a common algorithm step.• Efficiency: Scheduling in a common algorithm step requires a particularly efficient method to have enough computational capacity in the critical situations for scheduling the response. In this case, a reduced feature space Q specific to the invention is used, which permits planning and triggering of the maneuver in only one dimension and, in addition, also represents a simple value for characterizing the geometry of the collision position. This allows mapping to a scalar constellation parameter q.• Safety: Implementation of the CAL strategy can be carried out in a safe manner. If a maneuver is not safe taking into account typical errors in the prediction or with regard to sensing, etc., it must not be called up. By "nonsecure" herein it can be understood that it leads to significant degradation over the state without CAL intervention.Since the method of the present invention is based on a simple feature space Q, it can be described in contrast to the prior art without the usual "black box" which contains the decision criteria and their evaluation.This may be advantageous in the verification and validation of algorithms. This is because a low complexity algorithm is still available to classical review methods (e.g., FMEA, FTA).According to one specific embodiment of the method, the scalar constellation parameter is ascertained as a function of a first length or a first width of the vehicle or the second length or second width of a collision partner.According to one embodiment of the method, the first length and the second length are estimated vehicle lengths.According to one embodiment of the method, the first width and the second width are estimated vehicle widths.The lengths and widths can be effected in this case as a function of environmental data of the vehicle, from which the geometric dimensions of the vehicle or at least of the vehicle type can be derived. If the vehicle type is derived, the lengths and widths can be taken from a database stored locally in the vehicle or from a database available online via an Internet connection.According to one embodiment of the method of the present invention, a novel reference trajectory is determined as a function of the scalar constellation parameter q, which is also called a "safety boost" trajectory. The determination of the trajectory can be determined on the basis of one of the three following fundamental reaction patterns:• Initiating a previously defined, additional change in the travel of the vehicle by acceleration in the event of a imminent lateral collision;• Initiating a situationally calculated additional change in the travel of the vehicle by acceleration in the event of a imminent lateral collision;• Initiation of full braking by a latency-associated brake system during the acceleration phaseA reference trajectory (reference trajectory) originates from the trajectory planning as part of the motion planning of a system for controlling an at least partially automatically operated vehicle. In the present case, trajectory planning can be understood to mean that a time law (time law) is added to a geometric path.A further aspect of the present invention is a computer program which is set up to carry out all the steps of the method according to the present invention.A further aspect of the present invention is a machine-readable storage medium on which the computer program according to the present invention is stored.A further aspect of the present invention is an electronic control unit which is configured to carry out all the steps of the method according to the present invention.Embodiments of the present invention are explained in more detail below with reference to drawings.They show FIG. 1 shows a block diagram of a system for controlling an at least partially automatedly operated vehicle; FIG. 2 shows a schematic illustration of an intersection scenario; FIG. 3 shows different two-dimensional collision positions; FIG. 4 shows a further block diagram of a system for controlling an at least partially automatically operated vehicle; FIG. 5 is a block diagram of an algorithm for implementing the method of the present invention; FIG. 6 is a heat map for a decision process of the present invention; FIG. 7 shows a decision matrix; FIG. 8 shows a schematic illustration of a reaction maneuver; FIG. 9 shows the activation of the vehicle in the case of the execution of a safety boost; FIG. 10 shows the activation of a safety boat in combination with automatic emergency braking.FIG. 1 shows a block diagram of a system for controlling an at least partially automatedly operated vehicle, in which a method according to the present invention is implemented.Block 10 represents the detection module (I.B.R.). This module provides input parameters to all other modules.Block 12 represents the task planning module (Engl. Mission Planning, Route Planning). In this module, the sequence of waypoints from a given start to a given destination is determined based on available static and dynamic data, such as. This is done by ascertaining maps or traffic data.Block 14 represents the behavior planning ("behavior planning"). Behavior planning is a decisive component of an at least partially automatically operated vehicle. In this case, a sequence of actions is created which the vehicle must perform in order to reach its destination and in the process to comply with the traffic rules and to avoid collisions with other road users. The behavior module transfers motion specification (motion specification) to the following module.Block 16 represents the motion planning (motion). The movement planning takes place therein. The method of the present invention is particularly suitable for implementation in or on this module. In the drawing, it is shown as a simultaneous emergency + CAL planning (emergency and collision alignment planning module 17.This modular complex passes over a reference trajectory (engl. Reference Trajectory) to the control module 18 (engl. Control) of the vehicle ego.The reference trajectory originates from the trajectory planning as part of the motion planning. In the present case, trajectory planning can be understood to mean that a time law (time law) is added to a geometric path.Another embodiment that can be implemented based on the present invention relates to simultaneous maneuver planning in the dimension-reduced feature space Q.q, from the feature space Q, is a scalar constellation parameter q, which can describe an existing or expected collision position in an intersection scenario (i.e. a scenario playing in the two-dimensional space (plane)) essentially by means of a mapping to a single dimension.FIG. 2 shows a schematic illustration of such an intersection scenario and, by way of example, the effects on the constellation parameter q.In the illustrated intersection scenario, a vehicle ego enters an intersection coming from the left and arrives there at a collision sign opp coming from below. The intersection scenario is mapped onto the constellation parameter q at a time t 0 in the future. The time t 0 describes the time of the start of a collision between the vehicle ego and the collision partner opp. The constellation parameter q indicates the distance of the respective next corner of the respective vehicle. In the figure, diamonds are indicated in each case. For the vehicle ego, this is the right front corner. For the collision partner, the left front corner is opp.In particular, the point of impact with respect to the vehicle structures of the participants is described. For example, in one embodiment, constellation parameter q may be defined as the distance that the approaching facing corners of the colliding vehicles take at time t 0 l and w are typical or measured vehicle lengths or widths, depending on the algorithm. Instead of typical values for l and w, position-estimated values for the length and width can also be used. It is also conceivable that different values for the length and width are set for the vehicle and the collision partner.If the constellation parameter q assumes the value 0, this means that there is no longer any distance between the respective reference points of the vehicle ego and the collision encounter opp, and consequently contact of the two corners and thus a collision of the vehicle ego with the collision encounter opp takes place or will take place. If the value of q + / - the typical length+the typical width of a vehicle, this means that a lateral collision will occur. In the illustrated examples, at a negative value of q, the vehicle ego encounters the side of the collision sign opp. With a positive value of q, the collision sign opp encounters the side of the vehicle ego.FIG. 3 shows the various two-dimensional collision positions and the range of the scalar constellation parameter q at vehicle lengths l1, and l2and vehicle widths w1, w2.The decisive factor for the control of the vehicle operated at least partially in an automated manner is the estimation of the parameters q and q̂ (q at t0). Assuming that the speeds of the vehicle ego v e ve and of the collision sign opp v 0 are constant (constitutive velocity (cv) assumption) and the vehicles at the time of the estimation have the distances d ego and doppto the collision location, the following approximation q and q̂for the time until the collision t c can also describe a journey past and can therefore be used for both reaction patterns avoidance and alignment of the collision. In this case, q would not describe the distance of the corners at the time of collision, but rather indicates the minimum distance of these corners over time. In this parameter space, both reaction patterns collision avoidance and alignment can therefore be planned simultaneously.FIG. 4 shows a block diagram of a system for controlling an at least partially automatedly operated vehicle for implementing the method of the present invention. A model of a possible implementation of the present invention is shown.The following table serves as a legend for the presentation.T propulsionTime for Driving the Drive with a maxT brakeTime for Activating the Brake with a mint propulsionDuration of the Drive Activation with a maxt brakeDuration of the Actuation of the Brake with a mina maxrequested acceleration during boosta minRequested Delay during BoostIf a value for q is present and a decision has already been made to react to the collision scenario, a system according to FIG. 4 can be used.The control device 40 receives input signals from a vehicle dynamic sensor system 41. Optionally, the sensor system can provide values for the inclination or the distance s traveled.The control device 40 receives input signals from a surroundings sensor system 40. this sensor system may be designed in such a way that it delivers, for example, kinetic parameters such as distance s, speed v and acceleration a of the collision sensor opp or of a vehicle driving in front.A drive model 401 of the vehicle is present in the control device 40. This includes, among other things, the performance characteristic of the vehicle. There is a brake model 402 of the vehicle.The input signals of environment sensor system 40 enter into a calculation specification 403 for a setpoint value of an acceleration or brake booster. This setpoint value is also included in the calculation of the parameters for controlling the vehicle.In the present case, a boost can be understood to mean the application of particularly high energy for a short period of time for acceleration or deceleration.For the control of the vehicle, a calculation rule 404 includes all the external input signals and the internal input values in order to calculate control parameters 43 for the drive system of the vehicle and the control parameter for the brake system 44.Figure 5 shows a block diagram of an algorithm for implementing the method of the present invention.In block 51, an image of the environment of the vehicle ego can be ascertained via an environment sensor system 42. Features (engl. Features) are extracted and fed to block 52. In block 52, the detection (perception), connection (fusion) and prediction (prediction) of the environmental parameters take place. The features (engl. Feature) is supplied to the motion planning module (motion planning module) 50.In block 50, a block 501 is found in which the core of the present invention is advantageously carried out, namely the mapping of a collision situation or possible collision situation to a constellation parameter q.In block 502, a simultaneous check of the present situation and consideration of the determined environmental features and the constellation parameter q then takes place in order to establish the best possibility of reacting for the vehicle operated at least partially in an automated manner. Basically, two response patterns are collision avoidance and collision alignment. The first is preferable because this can avoid an imminent collision with high probability. If this option is omitted, the collision situation of a then unavoidable collision can be improved in that an optimal alignment of the collision partners with respect to one another can take place.The result of the block 502 is transferred into a block 503 for trajectory planning. The reference trajectory determined there is then transferred to the vehicle control module 53.FIG. 6 shows a so-called heat map for the decision process about the reaction pattern to be applied.For the heat map, the constellation parameter q is related to the speed v ego, i.e. the speed of the vehicle ego. The "temperature" indicates a severity of collision depending on the amount of degradation in velocity, Δv. The more the speed is reduced in the event of a collision, the more harmful the collision acts on the occupants of the vehicle ego and of the vehicle of the collision encounter opp. On the basis of the heat map, it is possible to decide which minimally damaging collision can still be achieved by aligning the vehicle ego with the vehicle of the collision front.Particular point of impingement J3 cabin (cabin), J6 corner (corner) and J8 cabin (my cabin) of its own are identified in the heat map by arrows.Below the heat map, the collision positions of FIG. 3 are schematically shown as a reference.The heat map may be created using machine learning and artificial intelligence techniques.FIG. 7 shows a decision matrix which can be used for the decision about the response pattern applied.The constellation parameter is plotted on the x-axis of the matrix. The constellation parameters are plotted on the y-axis. The matrix indicates, among other things, which maximum or minimum value of the matrix.Constellation parameters depending on saved or obtained distances are then achieved. This makes it possible to make the decision as to which reaction pattern could fail the most advantageous for the occupants of the vehicle ego.The decision for a response pattern may make a sub-function based on the decision margin. The accident constellations achievable for the vehicle ego have a constellation parameter q between q min. and q max. From these values, the decision function can choose a constellation with the parameter q t. For each combination of q min. and q max a decision is defined which corresponds to a desired q t. The figure shows 5 reaction strategies which have proved to be particularly effective. This is: "suppressed" (i.e. no substantial control intervention takes place; "rear end" (i.e. the vehicle ego will strike the vehicle of the collision sign opp in the rear region. "front axle" (i.e. the collision takes place in the region of the front axle. A vehicle axle represents a particularly fixed structure that can absorb a large amount of collision energy; "front part" ("front structure"), i.e. the collision will take place in the front structure of the vehicle; "boost", this reaction strategy is based on the vehicle ego being accelerated strongly in order to achieve avoidance of the collision. However, this set is not complete.The white arrows indicate in which direction the clearance [q min., q max.] shrinks as the situation progresses in time. At the time of the collision, both values are the same and thus lie on the diagonal. Movements downwards or to the right are not possible, since this would correspond to an expansion instead of a reduction in the clearance.As soon as a reaction strategy is found, the combined braking and steering intervention is optimized under the boundary conditions of the reaction strategy in order to achieve a crash corresponding to the strategy and at the same time low in energy.Example case of hit in the opposing vehicle cabin.This initial situation (without CAL) leads to a high impulse exchange and strong forces on the occupants. Both vehicles are deformed to a great extent and, in particular, the opposing driver's cab is pushed in.The CAL function here shifts the point of impingement to the extreme end of the opposing vehicle side. The forces between the vehicles are very greatly reduced, and the vehicles are hardly deformed. At the same time, the collision energy is minimized by a reduced speed. Overall, the forces on the occupant are reduced to such an extent that airbag deployment can be dispensed with in this case.A further exemplary embodiment of the present invention relates to the so-called safety boost.A possible accident constellation in the intersection region is a rear collision in which a vehicle A that has hit has not fully made it across the intersection and a further vehicle B that crosses the lane hits the rear of the vehicle A, for example in the region of the C-pillar. Vehicle A cannot align with an AEB in less than 1 second, which would only move the impact point further toward the passenger compartment. Acceleration is also difficult because increasing the speed in a collision always involves a risk because the kinetic energy is increased and the relative speed to vehicles traveling ahead is increased.The object of this exemplary embodiment is safety boost, a new safety maneuver, which in the case described allows vehicle A to move out of the risk zone without collision, without increasing the speed beyond the time range of a "start-up second". FIG. 8 schematically shows this maneuver. Vehicle A is advancing a defined distance forward about 500 ms before an expected rear meeting (in the example, about 1 m in addition). Immediately after the hazardous situation, the vehicle has the original speed and has thus not generated any additional risk of a rear-end collision, since the relative speed to vehicles to be driven is maintained.This maneuver requires fine tuning of actuators for acceleration and deceleration. The braking maneuver can advantageously still be initiated during the acceleration process. This embodiment is characterized in that the specific execution of the safety boost takes place at the identical initial and final speeds. Advantageously, this function is anchored in the Vehicle Motion Control (VMC) and in the sense of a reflex agent which triggers this reaction pattern directly in the VMC.For example, the vehicle with right to drive ego can be struck laterally by a vehicle with waiting opp while continuing the current longitudinal dynamics at an intersection.If, when the vehicle ego approaches or crosses an intersection, it is detected, for example by means of a surroundings sensor system, that another vehicle is on a lateral collision course, it is possible to react with a dedicated acceleration profile in the event that the prediction of the collision area is in the rear area of the vehicle ego, which may be a) predefined or b) calculated in situ.FIG. 9 shows the activation of the vehicle in the case of the execution of a safety boost.FIG. 9 shows in dashed lines (or red) and traverses (or blue) the adjustment of acceleration and deceleration, which gives the desired profile in the superposition.This allows exact positioning of the vehicle ego relative to the opponent by adjusting the timing of the actuators.Special Case: Electrified VehiclesThe following measures are proposed for this purpose:Electric vehicle / hybrid: at the time of the function triggering, the electric motors operate in the highest possible overload range (i.e. electric motors can boost without problems for several seconds to minutes; limiting factor is the generation of heat)For a realization according to b), a very reliable environment sensor system with a short cycle time is required. Embodiment a) is easier to implement, since it has proven successful.FIG. 10 shows the adaptation of the above-described maneuver safety boost in the transition to automatic emergency braking (engl. AEB).FIG. 10 shows in dashed lines (or red) and traverses (or blue) the adjustment of acceleration and deceleration, which gives the desired profile in the superposition.If it is determined during the safety boost that either a vehicle driving ahead is greatly decelerated or the vehicle A would be hit despite the maneuver, the braking profile can be completed simply for full braking. This can be done without further latency, since the brake has already been physically activated and no further delay TTL (time-to-lock) arises.Another positive effect of the present invention is that by better estimating the collision position and decreasing the situational blur, the imminent collision and its accelerations can be more accurately predicted. This opens up possibilities of improved pre-crash actuators and also improved adversivity in the in-crash phase.In one exemplary embodiment, this information, which consists of the collision position, mapped onto the constellation parameter q and the relative speed v rel at the time of the collision, can be used to actuate restraining means to the situation adapted by these two parameters. In this case, in one exemplary embodiment, the activation is carried out as a comparison of a function value f RHS, which is dependent on at least these parameters, with a predefined, restraint system-specific threshold value THD RHS with:In this case, f RHS can also additionally be dependent, for example, on other parameters, such as the occupant mass m occ or the seat position s occ i.e., f RHS= f RHS( q, v rel, m occ, s occ). This function can be formulated specifically for each occupant in his respective seat position or can be adapted by further variables. The function can also be designed such that, for a specific retaining means, in addition to the information as to whether it is to be activated or not, it also provides an additional parameter which changes the property of the retaining means as a function of at least the parameter q.In another simple embodiment, for example, in a table form for a specific restraining means, for a respectively specific interval of the values of q=[q n... q n+1], q n< q n+1 and a respectively specific interval of v rel= [ v reln,... v reln+1], v reln< v reln+1 a specific behavior of a specific restraining means RHS can be predefined, e.g. activated / not activated, or if it is activated, optionally additionally a specific parameter is added to the activation, which parameter is stored in this table, e.g. the level of a force effect, providing this restraining means (e.g. at a belt force limiter).References included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedDE 2012 110 21004

[0005] DE 10 2014 218 549

[0006] WO 09 09 09 2374 A1

[0006]

Claims

Method for controlling an at least partially automatically operated vehicle for avoiding and / or aligning in a collision situation of the vehicle, wherein the control of the vehicle takes place on the basis of a scalar constellation parameter.Method according to Claim 1, wherein the collision situation is related to a collision partner and the collision partner is a second vehicle, wherein the scalar constellation parameter is determined as a function of a first length and / or a first width of the vehicle and / or the second length and / or second width of a collision partner.The method of claim 2, wherein the first length and / or the second length are estimated vehicle lengths.The method of claim 2 or 3, wherein the first width and / or the second width are estimated vehicle widths.Method according to one of the preceding claims, wherein a reference trajectory for controlling the vehicle is determined as a function of the scalar constellation parameter.Computer program which is set up to carry out all the steps of a method according to one of the preceding claims.A machine readable storage medium having stored thereon the computer program according to claim 6.Electronic control unit which is configured to carry out all the steps of a method according to one of Claims 1 to 5.

Images