Method and control unit for influencing the driving dynamics of a vehicle

Abstract

The present invention relates to a method for influencing the driving dynamics of a vehicle, wherein data-driven interventions (116) for actuators (104) of the vehicle are determined using a data-driven algorithm (112) and, in parallel, conventional interventions (118) for the actuators (104) are determined using a conventionally verifiable algorithm (114), wherein a comparison is carried out between the data-driven interventions (116) and the conventional interventions (118) and, using a result of the comparison, interventions (106) to be carried out for the actuators (104) are determined.

Description

Field of the invention

[0001] The invention relates to a method for influencing the driving dynamics of a vehicle, a corresponding control unit, and a corresponding computer program product. State of the art

[0002] A vehicle's driving dynamics controller influences the vehicle's driving dynamics through interventions. These interventions are calculated by a rule-based algorithm and executed by the vehicle's actuators. For example, the interventions can include braking, steering, and / or drive interventions. The rule-based algorithm is specifically applied to a model or model variant of the vehicle during test drives. The rule-based algorithm is traceable and therefore its functional safety can be ensured.

[0003] The application is time-consuming because a large number of algorithm parameters are configured by an application engineer. Only a small number of parameters can be changed at a time to allow for monitoring the effects of the change. Disclosure of the invention

[0004] Against this background, the approach presented here introduces a method for influencing the driving dynamics of a vehicle, a corresponding control unit, and a corresponding computer program product according to the independent claims. Advantageous further developments and improvements of the approach presented here result from the description and are described in the dependent claims. Advantages of the invention

[0005] A data-driven algorithm is not configured by an application engineer. It is generated through machine learning or artificial intelligence using training data or test drives, without human intervention. A data-driven algorithm may not be transparent or verifiable, and its functional safety cannot be ensured using conventional methods. However, during vehicle operation, the data-driven algorithm, like a rule-based algorithm, issues interventions for the vehicle's actuators. These interventions can influence the vehicle's driving dynamics more effectively than those of a rule-based algorithm. Nevertheless, a residual risk remains.

[0006] In the approach presented here, two algorithms are executed in parallel within a vehicle. One of the algorithms was automatically created for the vehicle using machine learning and is therefore not directly verifiable. This algorithm is referred to here as the data-driven algorithm. The other algorithm is a conventionally created algorithm and is therefore verifiable. This algorithm is referred to as the conventionally verifiable algorithm. Both algorithms are fed with input variables from the vehicle and calculate interventions for the vehicle's actuators in parallel. As long as the interventions of the data-driven algorithm influence the vehicle's driving dynamics better than the interventions of the conventionally verifiable algorithm, the interventions of the data-driven algorithm are executed.If the interventions of the data-driven algorithm have a worse impact on the vehicle's driving dynamics than the interventions of the conventionally safe algorithm, the interventions of the conventionally safe algorithm are executed.

[0007] The approach presented here allows a data-driven algorithm that is inherently unsafe to be indirectly secured to achieve functional safety by executing a conventionally safe algorithm in parallel and switching to the interventions of the conventionally safe algorithm if the interventions of the data-driven algorithm are worse than those of the conventionally safe algorithm.

[0008] A method for influencing the driving dynamics of a vehicle is proposed, wherein data-driven interventions for vehicle actuators are determined using input variables of the vehicle and a data-driven algorithm, and in parallel, conventional interventions for the actuators are determined using the input variables and a conventionally verifiable algorithm, a comparison is carried out between the data-driven interventions and the conventional interventions, and interventions to be executed for the actuators are determined and sent to the actuators using a result of the comparison.

[0009] Ideas for embodiments of the present invention can be considered to be based, among other things, on the thoughts and findings described below.

[0010] A vehicle can be a two-track vehicle, such as a car or truck. It can also be a single-track vehicle, such as a motorcycle. Vehicle actuators can be integrated into the braking system, drive system, steering system, and / or chassis. Specifically, the actuators can be the vehicle's brakes. In this case, a corresponding vehicle dynamics control system can be called ABS / ESP.

[0011] Input variables can be measured values ​​taken directly at the vehicle. Input variables can also be input from the vehicle's driver. Input can also originate from a vehicle assistance system.

[0012] A data-driven algorithm can be called an AI (artificial intelligence) algorithm. A conventionally safe algorithm can be called a deterministic algorithm. For simplicity, the conventionally safe algorithm can be called a conventional algorithm.

[0013] An intervention can be triggered by a control signal. These control signals are specifically coded for the respective actuator being controlled. Both algorithms output identically coded control signals for the same actuator. For example, for a vehicle's braking system, both the data-driven algorithm and the conventionally secured algorithm determine uniformly coded control signals for the braking system. Therefore, comparable control signals can be compared to determine the interventions. The selected control signals can then be sent to the actuators to trigger the interventions.

[0014] The comparison can, for example, be part of a conventionally secured algorithm. Alternatively, the comparison can be performed using a separate comparison algorithm.

[0015] As a rule, data-driven interventions can be determined as the interventions to be executed. Conventional interventions can be determined as the interventions to be executed if they are superior to the data-driven interventions. An intervention can be considered "better" if it ensures greater deceleration while simultaneously preventing wheel lock-up. An intervention can also be considered "better" if it is deemed safer in terms of functional safety. The data-driven algorithm can identify superior interventions than the conventional algorithm in many areas. However, the conventional algorithm can reliably determine interventions even at the limits of vehicle dynamics. This prevents, for example, the vehicle from being under-braked (i.e., braked too weakly) by the data-driven interventions.This also prevents the vehicle from being over-braked by the data-driven interventions, which would lead to the wheels locking up.

[0016] Conventional interventions can be determined as the interventions to be performed if they deviate from data-driven interventions by more than a predefined tolerance. Small deviations can be tolerated. If the deviations exceed the tolerance, a switch to conventional interventions may be necessary. The tolerance could, for example, be 10 percent of the conventional interventions. A distinction can be made here as to whether this would result in stronger or weaker deceleration of the vehicle. Such a tolerance can be measured, for example, as braking torque / Newton-meter [Nm], braking force / Newton [N], brake pressure [Bar], or actuator signals such as valve opening times.

[0017] Conventional interventions can be selected as the interventions to be performed if they are more effective than data-driven interventions for a predefined duration. Within this timeframe, the deviation can exceed the tolerance. This temporal tolerance can, for example, buffer overshoots from conventional interventions. The duration can range from, for example, 100 milliseconds to 500 milliseconds.

[0018] The data-driven algorithm can be specifically trained for a particular vehicle configuration. This allows the data-driven algorithm to be very well adapted to the vehicle.

[0019] A conventionally secureable algorithm can be generic for different vehicles, thus eliminating the application overhead for each individual vehicle model. However, a conventionally secureable algorithm can also be essentially the same for different vehicles. This means that the conventionally secureable algorithm may be less well-suited to the specific vehicle, vehicle model, or vehicle configuration. In other words, a conventionally secureable algorithm can be secure, but it may not be specifically optimized for that vehicle.

[0020] At least one parameter of the conventionally verifiable, generic algorithm can be parameterized for the vehicle. For example, basic parameters such as the vehicle's unladen weight, wheelbase, or center of gravity can be parameterized to at least roughly characterize the vehicle.

[0021] A deviation between data-driven and conventional interventions can be integrated to form a deviation sum. Conventional interventions can be selected as the interventions to be executed if the deviation sum exceeds a tolerance level. Integrating the deviation reveals when conventional interventions deviate from data-driven interventions in one direction within the tolerance range for an extended period. If the deviation fluctuates between positive and negative, the integral does not exceed the tolerance level. Integrating the deviation can also detect small systematic errors in the data-driven algorithm. In such cases, a switch to conventional interventions can be made for safety. The tolerance level could, for example, be 10 percent of the conventional interventions.A distinction can be made here as to whether this would lead to a stronger or weaker deceleration of the vehicle. This tolerance sum can be expressed, for example, with a unit [Nms] (newton-meter-seconds) or [Ns] (newton-seconds).

[0022] The interventions to be implemented can be a combination of data-driven and conventional interventions. The ratio of these interventions can depend on the result of the comparison. For example, the ratio could depend on the integrated total deviation. In this case, as the deviation increases, the conventional interventions can be increasingly represented in the interventions to be implemented. However, the optimized component of the data-driven interventions is not lost in this process.

[0023] If conventional interventions are determined to be the required interventions, a message can be stored in the vehicle's diagnostic memory. This diagnostic memory, or fault memory, can be read during a workshop visit. Alternatively, the fault message can be transmitted directly to the vehicle manufacturer or supplier via radio. The data-driven algorithm can then be updated, for example. Normally, data-driven interventions should be used, as the data-driven algorithm is optimized for the vehicle. Using conventional interventions should be the exception. The messages in the diagnostic memory allow for verification of the data-driven algorithm's effectiveness. If no messages are stored in the diagnostic memory for many vehicles over an extended period, the functional safety of the data-driven algorithm can also be demonstrated.

[0024] The method is preferably computer-implemented and can be implemented, for example, in software or hardware or in a hybrid form of software and hardware, for example in a driver assistance system.

[0025] In particular, individual or preferably all of the process steps to be carried out within the framework of the procedure can be automated, i.e., performed by one or more machines. For example, data-driven interventions and / or conventional interventions can be automatically determined to automatically influence vehicle dynamics. Furthermore, the data-driven interventions and the conventional interventions can be automatically compared with each other, and the interventions to be carried out can be automatically determined.

[0026] The approach presented here also creates a control unit, the control unit being designed to carry out, control or implement the steps of a variant of the procedure presented here in appropriate facilities.

[0027] The control unit can be an electrical device with at least one processing unit for processing signals or data, at least one storage unit for storing signals or data, and at least one interface and / or a communication interface for reading or outputting data embedded in a communication protocol. The processing unit can be, for example, a signal processor, a so-called system ASIC, or a microcontroller for processing sensor signals and outputting data signals depending on the sensor signals. The storage unit can be, for example, flash memory, an EPROM, or a magnetic storage device. The interface can be configured as a sensor interface for reading sensor signals from a sensor and / or as an actuator interface for outputting data signals and / or control signals to an actuator.The communication interface can be configured to read or output data wirelessly and / or via a wired connection. The interfaces can also be software modules, such as those found on a microcontroller alongside other software modules.

[0028] Also advantageous is a computer program product or computer program with program code that can be stored on a machine-readable carrier or storage medium such as a semiconductor memory, a hard disk memory or an optical memory and is used to carry out, implement and / or control the steps of the method according to one of the embodiments described above, in particular if the program product or program is executed on a computer, in a control unit or a device.

[0029] It should be noted that some of the possible features and advantages of the invention are described herein with reference to different embodiments. A person skilled in the art will recognize that the features of the control unit and the method can be suitably combined, adapted, or exchanged to arrive at further embodiments of the invention. Brief description of the drawing

[0030] The following describes embodiments of the invention with reference to the accompanying drawing, whereby neither the drawing nor the description is to be interpreted as limiting the invention. Fig. Figure 1 shows a block diagram of a vehicle dynamics control system according to an exemplary embodiment.

[0031] The figure is merely schematic and not to scale. Identical reference symbols denote identical or equivalent features. Embodiments of the invention

[0032] Fig. Figure 1 shows a block diagram of a vehicle dynamics control system 100, in particular a motorcycle. The vehicle dynamics control system 100 uses input variables 102 and, based on these, outputs interventions 106 to be performed on actuators 104 of the vehicle. The input variables 102 can be measured variables 108 of the vehicle and / or requirements 110 for interventions via vehicle controls and / or vehicle assistance functions.

[0033] For the vehicle dynamics control system 100 presented here, two algorithms 112 and 114 are executed in parallel. The first algorithm 112 is a data-driven algorithm 112. The second algorithm 114 is a conventionally safe algorithm 114. The data-driven algorithm 112 determines data-driven interventions 116. The conventionally safe algorithm 114 determines conventional interventions 118. The conventionally safe algorithm 114 is also referred to as the conventional algorithm 114.

[0034] Both algorithms 112 and 114 process the input variables 102, which are at least substantially the same. The input variables 102 of the conventional algorithm 114 can be filtered in a filter 120 before processing to avoid unrealistic input variables 102. Algorithms 112 and 114 can be executed on separate processing units of the vehicle or on a shared processing unit of the vehicle.

[0035] The data-driven algorithm 112 is based on an artificial intelligence specifically trained for the vehicle. Due to the training and machine learning, the data-driven interventions 116 often intervene more effectively in the vehicle dynamics than the conventional interventions 118. However, with the data-driven algorithm 112, a direct correlation between the input variables 102 and the data-driven interventions 116 is not readily apparent. Therefore, demonstrating the functional safety of the data-driven algorithm 112 is difficult.

[0036] The conventionally safe algorithm 114 is rule-based and deterministic. The relationship between the input variables 102 and the conventional interventions 118 is directly traceable. Therefore, the functional safety of the conventionally safe algorithm 114 has been demonstrated.

[0037] In the approach presented here, the data-driven interventions 116 and the conventional interventions 118 are compared. Depending on the result of the comparison, either the data-driven interventions 116 or the conventional interventions 118 are sent to the vehicle's actuators 104 and executed as the interventions 106 to be performed.

[0038] In one embodiment, the data-driven interventions 116 are typically used as the interventions 106 to be executed. Only if the conventional interventions 118 are superior to the data-driven interventions 116 are the conventional interventions 118 used as the interventions 106 to be executed. Thus, the vehicle can typically benefit from the training of the data-driven algorithm 112. In critical situations, the conventional algorithm 114 serves as a fallback for the data-driven algorithm 112.

[0039] In one embodiment, the conventional interventions 118 are only used when they are a tolerance better than the data-driven interventions 116. Since both algorithms 112 and 114 process essentially the same input variables 102 and have the same goal, the two interventions 116 and 118 can be very similar. A minor deviation is acceptable.

[0040] In one embodiment, the system switches to conventional interventions 118 only after a predefined time period. When algorithms 112 and 114 react to changes in the input variables 102, the changes in interventions 116 and 118 may differ. For example, the conventional interventions 118 may exhibit an overshoot if the control parameters of the conventional algorithm 114 are less well adapted to the vehicle than those of the data-driven algorithm 112, which is specifically trained for the vehicle.

[0041] In one embodiment, a deviation between the data-driven interventions 116 and the conventional interventions 118 is integrated. If the integral becomes larger than a certain percentage of the deviations, the conventional interventions 118 are selected, at least partially, for the interventions 106 to be executed. This allows a small but persistent deviation in one direction to be detected and corrected.

[0042] In one embodiment, the conventional algorithm 114 performs the comparison of the interventions 116 and 118. The conventional algorithm 114 thus controls the output of the interventions 106 to be executed. For the comparison, the conventional algorithm 114 reads in the data-driven interventions 116 and compares them with the self-calculated conventional interventions 118.

[0043] In an alternative embodiment, an independent comparator 122 performs the comparison and controls the output of the interventions 106 to be executed. The comparator 122 reads the data-driven interventions 116 and the conventional interventions 118, compares them, and outputs the interventions 106 to be executed.

[0044] In one embodiment, the interventions 106 to be performed are mixed in a mixer 124 using the data-driven interventions 116 and the conventional interventions 118. The interventions 106 are mixed proportionally from the data-driven interventions 116 and proportionally from the conventional interventions 118. A mixing ratio is set based on the result of the comparison and is, for example, variable between zero and one. The mixing ratio can also depend, for example, on the integral of the deviation.

[0045] Possible embodiments of the invention are summarized below or presented using slightly different wording.

[0046] A safety mechanism, particularly for an anti-lock braking system based on artificial intelligence, is presented.

[0047] Anti-lock braking systems (ABS) for motorcycles are traditionally applied specifically for each vehicle type. The ABS is parameterized to suit the specific characteristics of the respective motorcycle. This process is complex and expensive, as it can only be performed by experienced application engineers.

[0048] By using AI-based approaches, this application effort can be significantly reduced. The algorithm learns the motorcycle's specific characteristics directly and does not need to be manually adapted to the respective model. Furthermore, the resulting deceleration can exceed that of conventional ABS.

[0049] One problem with the AI ​​approach is proving that the "learned" algorithm is "safe." This is a fundamental problem with AI algorithms. The approach presented here allows such an AI approach to be implemented in vehicles while simultaneously ensuring their safety.

[0050] In motorcycles, one of the worst possible errors occurs when the learned AI approach "under-brakes" the motorcycle. In this case, the algorithm prevents or reduces deceleration because it expects the wheel to lock up or the rear wheel to lift off the ground, even though stronger deceleration would be perfectly possible and a conventionally applied controller could achieve it.

[0051] The approach presented here implements two algorithms. Normally, the AI ​​algorithm is used to control the ABS. However, it is monitored by a conventional, generic algorithm. If the motorcycle's deceleration falls below the level that the conventional algorithm would guarantee, the control regime switches, and the conventional controller takes over ABS control. In this case, it is no longer necessary to optimize the conventional, generic algorithm for performance, but only for safety. This significantly reduces the application effort.

[0052] A motorcycle's ABS system uses specific sensor signals as input to regulate the slip between the tire and the road. The goal is to set the slip that results in maximum deceleration, while simultaneously preventing the wheel from locking up.

[0053] In the approach presented here, two controllers regulate slip, with only one controller in command at any given time. The first controller is an AI algorithm that has learned the slip level required to achieve maximum deceleration without wheel lock-up. This algorithm is normally in command. The second controller is a conventional algorithm that regulates slip. These algorithms are generally very safe (from a functional safety perspective) because they are well-established through years of experience.

[0054] The second controller constantly monitors the first controller. As long as the first controller sets a better slip than the second controller would itself, the second controller remains in observation mode. However, if the first controller sets a lower slip or causes the wheels to lock, the second controller actively switches from observation to command mode and actively regulates the wheel slip itself.

[0055] By applying this combined control strategy, all 'safety loads' are transferred to the conventional controller and not to the AI ​​controller, making the entire control strategy easier to approve for road traffic.

[0056] Several implementations are conceivable for the second controller. The second controller can be implemented identically for every motorcycle. Alternatively, the second controller can be a generic algorithm that can be adapted to different motorcycles through parameterization. Parameters could include, for example, weight, center of gravity, and / or wheelbase. The second controller can also be applied in a motorcycle-specific manner.

[0057] The command switch between the two controllers can, for example, only occur after a confirmation time interval has elapsed. The command switch can also only occur after a certain error magnitude has been exceeded. Furthermore, continuous crossfading can be performed based on the difference between the approaches.

[0058] The two controllers can be hosted on the same control unit. Likewise, the two controllers can be hosted on different control units.

[0059] The switch can be operated either by the second control or an independent monitoring control.

[0060] The driver can be informed of the regime change via HMI (Human-Machine Interface). Alternatively or additionally, the event can be written to the diagnostic memory or transmitted to the manufacturer wirelessly.

[0061] The approach presented here uses two control algorithms for ABS control on the motorcycle: a conventional one and an AI-based one.

[0062] The approach presented here can be used particularly for two-wheeled motorized vehicles with ABS. Likewise, the presented approach can be used in similar applications where AI approaches offer advantages in control, but proof of functional safety is not feasible, for example, ESP or ABS in two-track vehicles.

[0063] Finally, it should be noted that terms such as "comprising," "encompassing," etc., do not exclude other elements or steps, and terms such as "a" or "an" do not exclude a plurality. Reference numerals in the claims are not to be considered as limitations.

Claims

[1] Method for influencing driving dynamics of a vehicle, wherein data-driven interventions (116) for actuators (104) of the vehicle are determined using a data-driven algorithm (112) and, in parallel, conventional interventions (118) for the actuators (104) are determined using a conventionally secure algorithm (114), wherein a comparison is carried out between the data-driven interventions (116) and the conventional interventions (118) and interventions (106) to be carried out for the actuators (104) are determined using a result of the comparison. [2] The method of claim 1, wherein the conventional interventions (118) are determined as the interventions (106) to be performed if the conventional interventions (118) are better than the data-driven interventions (116). [3] Method according to claim 2, wherein the conventional interventions (118) are determined as the interventions (106) to be performed if the conventional interventions (118) deviate from the data-driven interventions (116) by more than a predefined tolerance. [4] Method according to one of claims 2 to 3, wherein the conventional interventions (118) are determined as the interventions (106) to be carried out if the conventional interventions (118) are better than the data-driven interventions (116) for longer than a predefined period of time. [5] Method according to one of the preceding claims, in which a deviation between the data-driven interventions (116) and the conventional interventions (118) is integrated to form a deviation sum, wherein the conventional interventions (118) are determined as the interventions (106) to be carried out if the deviation sum is greater than a tolerance sum. [6] Method according to one of the preceding claims, in which the interventions (106) to be carried out are mixed from the data-driven interventions (116) and the conventional interventions (118), wherein a mixing ratio is dependent on the result of the comparison. [7] Method according to one of the preceding claims, wherein the conventionally securable algorithm (114) is generic for different vehicles. [8] Method according to claim 7, wherein at least one parameter of the generic algorithm (114) is parameterized for the vehicle. [9] Method according to one of the preceding claims, wherein the data-driven algorithm (112) is trained for a configuration of the vehicle. [10] Method according to one of the preceding claims, in which a message is stored in a diagnostic memory of the vehicle when the conventional interventions (118) are determined as the interventions (106) to be carried out. [11] Control device, wherein the control device is designed to carry out, implement and / or control the method according to one of the preceding claims in corresponding devices. [12] Computer program product which is designed to instruct a processor, when executing the computer program product, to execute, implement and / or control the method according to one of claims 1 to 10. [13] A machine-readable storage medium on which the computer program product according to claim 12 is stored.

Images