Platform for path planning system development for autonomous driving systems

Abstract

A platform for path planning system development for autonomous driving systems is provided. The invention relates to a method (100) and apparatus (10) for developing new path planning features for autonomous driving systems (ADS) by using federated learning, utilizing production vehicles. To achieve this, the output of the "in-test" path planning module is evaluated in a closed loop in order to generate a cost function which is subsequently used to update or train the path planning model of the path planning development module (210).

Description

[0001] Cross-reference to related applications

[0002] This patent application claims priority to European Patent Application No. 21171222, filed on April 29, 2021, entitled “Method for Automated Development of a Path Planning Module for an Automated Driving System”, which has been assigned to the assignee and is expressly incorporated herein by reference. Technical Field

[0003] This disclosure relates to methods and systems for performance evaluation and development of path planning modules for vehicles equipped with automated driving systems (ADS). Specifically, the invention relates to closed-loop evaluation of the vehicle's path planning development module and its subsequent updates / training. Background Technology

[0004] Research and development activities related to autonomous vehicles have exploded in recent years, exploring many different approaches. An increasing number of modern vehicles are equipped with Advanced Driver Assistance Systems (ADAS) to improve vehicle safety and, more generally, road safety. ADAS, such as Adaptive Cruise Control (ACC), Collision Avoidance Systems, Forward Collision Warning, etc., are electronic systems that assist the driver while driving. Currently, research and development are underway in many technical areas associated with ADAS and Autonomous Driving (AD). In this document, ADAS and AD will be referred to as the general term Automated Driving Systems (ADS), corresponding to all the different levels of automation defined, for example, by SAE J3016 levels (0-5) (and especially levels 4 and 5) of driving automation.

[0005] In the near future, ADS solutions are expected to be applied to most new vehicles on the market. ADS can be understood as a complex combination of various components, defined as a system in which the perception, decision-making, and operation of a vehicle are performed by electronics and mechanics rather than a human driver, and can be defined as introducing automation into road traffic. This includes vehicle handling, destination selection, and awareness of the surrounding environment. While the automated system can control the vehicle, it allows the human operator to delegate all or at least some of the responsibilities to the system. ADS typically combines various sensors such as radar, lidar, sonar, cameras, navigation systems (e.g., GPS), odometers, and / or inertial measurement units (IMUs) to perceive the vehicle's surroundings, allowing the advanced control system to interpret the sensor information to identify appropriate navigation paths and obstacles, free space areas, and / or relevant landmarks.

[0006] Much of the current effort in developing ADS revolves around safely launching the first system to market. However, once this is achieved, improving the system in a safe and efficient manner will be extremely important to achieve cost reductions and performance improvements. Typically, there are significant costs associated with the development and validation of the security of ADS (or “ADS features”), particularly those related to field testing and understanding how the system performs in traffic. Furthermore, there are additional challenges in managing the vast amounts of data generated by ADS-equipped vehicles to develop and validate various ADS features, not only from the perspective of data storage, processing, and bandwidth, but also from a data security / privacy perspective.

[0007] Numerous research reports exist outlining various methods proposed for finding and executing such safe and reliable paths, and further, there exists an entire field of research dedicated to finding optimal paths given a diverse set of constraints. Analytical methods focus on finding paths given predictions and states of objects surrounding the vehicle. Another class of methods for obtaining paths for autonomous vehicles relies on machine learning or deep neural networks to perform the task of selecting suitable paths. Many currently known methods depend on detailed modeling of the surrounding environment to ultimately capture these models within the (complex) set of constraints of the following optimization problem for finding the optimal path. However, as the operating conditions and domains of ADS become increasingly complex, finding and executing such “safe” paths may prove extremely difficult, if not impossible.

[0008] Furthermore, when providing a path to be executed by ADS, a component or module is typically created that provides recommendations for the path and another component to check whether that path can be safely executed. This second checker component is often implemented using heuristics, as this is one of the few ways to determine the integrity / safety of a component. Another approach is to leave it to formal arguments or equations. However, heuristic-based approaches require the implementation team to be familiar with all possible scenarios and, further, to be error-free in determining the heuristics. As mentioned, with the increasing complexity of ADS's operating scenarios and domains, constructing such a complete set of heuristics to check path safety may prove daunting. Similarly, formal approaches fall into the same category of problems, attempting to explain all possible outcomes. For some solutions, it can only integrate one type of maneuver (vertical or lateral) to handle violations of its safe zone, as such a combination proves difficult to formally demonstrate.

[0009] Therefore, new solutions are needed in this field to facilitate the development, verification, and validation of ADS path planning capabilities, enabling the continuous delivery of safer and higher-performing systems. Typically, and preferably, improvements will be made without significantly impacting the size, power consumption, and cost of the airborne system or platform. Summary of the Invention

[0010] Therefore, the object of the present invention is to provide a method for the automatic development of a path planning development module for a vehicle performed by an onboard computing system, a computer-readable storage medium, an apparatus for the automatic development of a path planning development module for a vehicle, and a vehicle including such an apparatus, which mitigate, alleviate, or completely eliminate all or at least some of the disadvantages of currently known solutions.

[0011] This objective is achieved by a method for the automated development of a path planning development module for a vehicle, performed by an onboard computing system as defined in the appended claims, a computer-readable storage medium, an apparatus for the automated development of a path planning development module for a vehicle, and a vehicle including such an apparatus. The term "exemplary" will be understood in this context as used as an example, illustration, or description.

[0012] According to a first aspect of the invention, a method for automatically developing a path planning development module for a vehicle equipped with Adaptive Digital Assist (ADS), executed by an onboard computing system, is provided. The method includes obtaining candidate paths from the path planning development module. The path planning development module is configured to generate candidate paths for the vehicle to execute based on a path planning model and data indicating the vehicle's surrounding environment (e.g., sensor data, IMU data, HD map data, GNSS data, etc.). The method further includes obtaining a risk map of the vehicle's surrounding environment. The risk map is formed based on the vehicle's actuation capability and the location of a free space region in the surrounding environment, wherein the actuation capability includes an uncertainty estimate of the actuation capability, and the location of the free space region includes an uncertainty estimate of the estimated location of the free space region. The risk map includes risk parameters for each of a plurality of region segments included in the vehicle's surrounding environment. Furthermore, the risk map further has a time component that indicates the temporal evolution of the risk parameters of the region segments based on the predicted time evolution of the free space region, at least within a duration defined by the predicted duration of the candidate path.

[0013] The method further includes determining a risk value for a candidate path based on risk parameters of the region segment to which the candidate path intersects. Furthermore, if the determined risk value meets at least one risk criterion, the method further includes:

[0014] • Generate a first control signal at the output that indicates the instructions for executing candidate paths in ADS.

[0015] • Obtain at least one quality parameter of the candidate path to be executed, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort of the executed candidate path.

[0016] • Determine the cost function that indicates the performance of the path planning development module based on at least one obtained quality parameter.

[0017] • Update one or more parameters of the path planning model using an optimization algorithm configured to optimize a determined cost function.

[0018] According to a second aspect of the invention, a (non-transitory) computer-readable storage medium is provided for storing one or more programs configured to be executed by one or more processors of an in-vehicle processing system, the one or more programs including instructions for performing methods according to any of the embodiments disclosed herein. This aspect of the invention has similar advantages and preferred features to the first aspect of the invention previously discussed.

[0019] As used herein, the term "non-transitory" is intended to describe computer-readable storage media (or "memory") that exclude the propagation of electromagnetic signals, but is not intended to otherwise limit the types of physical computer-readable storage devices covered by the phrase computer-readable media or memory. For example, the terms "non-transitory computer-readable media" or "tangible memory" are intended to cover types of storage devices that do not necessarily store information permanently, including, for example, random access memory (RAM). Program instructions and data stored in a non-transitory form on a tangible computer-accessible storage medium can be further transmitted via a transmission medium or signals (such as electrical signals, electromagnetic signals, or digital signals) that can be transmitted via communication media (such as networks and / or wireless links). Therefore, the term "non-transitory" as used herein is a limitation on the medium itself (i.e., tangible, not signaling), rather than a limitation on the persistence of data storage (e.g., RAM versus ROM).

[0020] Furthermore, according to a third aspect of the invention, an apparatus is provided for the automatic development of a path planning development module for a vehicle equipped with ADS. The apparatus includes control circuitry configured to obtain candidate paths from the path planning development module. The path planning development module is configured to generate candidate paths for the vehicle to execute based on a path planning model and data indicating the vehicle's surrounding environment. The control circuitry is further configured to obtain a risk map of the vehicle's surrounding environment, wherein the risk map is formed based on the vehicle's actuation capability and the location of a free space region in the surrounding environment. The actuation capability includes an uncertainty estimate of the actuation capability, and the location of the free space region includes an uncertainty estimate of the estimated location of the free space region. The risk map includes risk parameters for each of a plurality of region segments included in the vehicle's surrounding environment. Furthermore, the risk map further has a time component that indicates the time evolution of the risk parameters of the region segments based on the predicted time evolution of the free space region, at least within a duration defined by the predicted duration of the candidate path. The control circuitry is further configured to determine a risk value of the candidate path based on the risk parameters of the region segments to intersect with the candidate path, and if the determined risk value satisfies at least one risk criterion, the control circuitry is further configured to:

[0021] • Generate a first control signal at the output that indicates the instructions for executing candidate paths in ADS.

[0022] • Obtain at least one quality parameter of the candidate path to be executed, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort of the executed candidate path.

[0023] • Determine the cost function that indicates the performance of the path planning development module based on at least one obtained quality parameter.

[0024] • Update one or more parameters of the path planning model using an optimization algorithm configured to optimize a determined cost function.

[0025] Furthermore, according to another aspect of the invention, a vehicle is provided comprising a set of onboard sensors configured to monitor the vehicle's surrounding environment. The vehicle further includes an automated driving system (ADS) having a perception system and a production platform path planner, the perception system being configured to generate perception data based on sensor data obtained from one or more of the set of onboard sensors. Further, the vehicle includes: a path planning development module configured to generate candidate paths for the vehicle; and means for automated development of the path planning development module according to any embodiment disclosed herein. Similar advantages and preferred features exist for this aspect of the invention as for the first aspect of the invention previously discussed.

[0026] The method, computer-readable storage medium, system, and vehicle proposed in this paper correspondingly provide a learning platform for production systems and sensors utilizing ADS to perform a joint learning process on new software and hardware versions of various path planning features in autonomous driving applications. More specifically, the learning approach proposed in this paper relies on a generated risk map that estimates and quantifies a view of the “risk” of the autonomous vehicle’s surrounding environment to determine the safety of generated candidate paths, while simultaneously enabling the subsequent training of a path planning model to optimize for quality and comfort. This readily provides advantages in the following aspects:

[0027] • By leveraging the available production resources in launched vehicles to further develop the path planning system, new path planning features for autonomous driving applications can be developed cost-effectively and in a timely manner.

[0028] • Robust and reliable safety assurance that allows for closed-loop evaluation of new path planning features.

[0029] Other embodiments of the invention are defined in the dependent claims. It should be emphasized that the term "comprising / including," as used in this specification, is used to indicate the presence of a stated feature, integral, step, or component. It does not exclude the presence or addition of one or more other features, integrals, steps, components, or combinations thereof.

[0030] These and other features and advantages of the present invention will be further illustrated below with reference to the embodiments described herein. Attached Figure Description

[0031] Other objects, features, and advantages of embodiments of the present invention will become apparent from the following detailed description with reference to the accompanying drawings, in which:

[0032] Figure 1 This is a schematic flowchart illustrating a method for automatically developing a path planning module for a vehicle, executed by an onboard computing system according to an embodiment of the present invention.

[0033] Figure 2 This is a schematic block diagram representation of a system for the automatic development of a path planning module for vehicles according to an embodiment of the present invention.

[0034] Figure 3 This is a schematic top view illustration of a risk diagram according to an embodiment of the present invention, wherein some of the different components contribute to the risk diagram.

[0035] Figure 4 It is relative to Figure 3 The diagram shows a schematic top view of the risk map at a subsequent time step.

[0036] Figure 5 This is a schematic top view illustration of a risk map according to an embodiment of the present invention, wherein the area segments are indicated in the form of a grid.

[0037] Figure 6 This is a schematic side view of a vehicle including an automatic development device for a path planning module for a vehicle, according to an embodiment of the present invention. Detailed Implementation

[0038] Those skilled in the art will understand that the steps, services, and functions described herein can be implemented using a single hardware circuit, software that runs in conjunction with a programmed microprocessor or general-purpose computer, one or more application-specific integrated circuits (ASICs), and / or one or more digital signal processors (DSPs). It will also be understood that, when the invention is described in relation to methods, it can also be implemented in one or more processors and one or more memories coupled to the one or more processors, wherein the one or more memories store one or more programs that, when executed by the one or more processors, perform the steps, services, and functions disclosed herein.

[0039] In the following description of exemplary embodiments, the same reference numerals denote the same or similar parts.

[0040] In this context, the risk map estimates and quantifies a view of the “risk” across the ADS (i.e., the environment surrounding the vehicle itself), and preferably includes uncertainties in sensor capabilities, detection and prediction, and the available capabilities of the vehicle platform.

[0041] As mentioned in the background section, current known processes for developing and launching new path planners for autonomous vehicles require significant manual work to generate paths that can be considered "safe." It should be noted that the term "safe" in this context will be understood as statistically safe above a probability threshold; for example, a path can be considered safe if the probability of an accident on it is determined to be less than a threshold defined by safety criteria. Accordingly, the term "safe" can be understood as having no risk, or more specifically, as a quantifiable measure of risk below a threshold.

[0042] Continuing, to mitigate this need for manual work, this paper proposes using a real-time risk map, derived from the ADS's own observation of its current situation, as a checker. The risk map includes all potential sources of risk, such as uncertainties in measurements from its own sensors, predictions of movement of available objects, the risk of new objects appearing, and the capabilities of the vehicle platform on which the ADS operates. If the risk accumulated when the ADS's recommended path is executed on the risk map is sufficiently low, it can be accepted. Therefore, using the risk map as a checker ensures that the path always carries a sufficiently low risk.

[0043] As a result, since the generated candidate paths are compared with the risk graph and are therefore considered safe for execution, it is possible to utilize these candidate paths to train the algorithm responsible for generating these paths during the closed-loop process to optimize for the quality / comfort parameters. It should be noted that this assumes here that the path planning model of the "path planning development module" has reached a sufficient level of maturity to generate "safe" paths, such that the risk graph evaluation is primarily used for failsafe purposes to prevent the execution of "unsafe" paths. To achieve this "sufficient level of maturity," a "risk graph" as disclosed herein can be used as a reference frame, and the path planning model can be trained against this reference frame during the open-loop process.

[0044] Furthermore, the risk graph can be extended with additional models using the same steps as extending it across new operational domains of the ADS, and in this way provides accurate risk estimates. In this manner, the risk graph is easier to maintain, and compared to heuristic methods, it further provides a more general concept for examining suggested paths for the ADS.

[0045] More specifically, this invention proposes the use of a risk graph that captures all the influences of the surrounding environment, detection, and prediction, as well as the uncertainties in the estimation of these surrounding environments and the uncertainties in the vehicle's own capabilities. Unlike optimization problems that are viewed as complex problems constrained by a large number of variables, most of these optimization problems are directly reflected in the risk graph. Accordingly, this can lead to more tractable and better-performing solutions, and also to solutions that are easier to extend to new domains. As our understanding of the world in which ADS operates increases, models of the supporting systems can be attached and fed into the risk graph to continue using the same methods for ensuring a "safe" path.

[0046] One of the main problems with traditional methods for formal path checking is that they do not account for uncertainties. Furthermore, since these models heavily rely on different parameters (such as braking capacity) to evaluate the effectiveness of the proposed path (or current state), the accuracy of these parameters is extremely important. To the best of the inventors' knowledge, there are no other suggestions on how to incorporate these aspects into currently known models. Risk maps, on the other hand, inherently provide all of these estimates by including the uncertainties of all aspects of the ADS. Therefore, using risk maps for path checking not only reduces the required implementation effort compared to formal or heuristic methods used for path checking, but also further incorporates the dynamic uncertainties inherent in the operating environment of the ADS. The result is a more reliable automated path checking and planning solution.

[0047] In other words, compared to traditional path planning solutions that rely on heuristics for security assessment, risk graphs provide an automated tool for risk-aware path planning, minimizing the manual labor required during development. Furthermore, since the model used for the surrounding world will inevitably need to be improved as the ADS increases its capabilities, so too will the risk graph, as it can be based on much of the same model upon which the ADS relies. Moreover, as it scales to more complex operational scenarios, the model needs to be in place to allow the ADS to operate autonomously there, making updated versions of the risk graph readily available. On the other hand, heuristic checkers would need to be updated and developed for every change in the operational domain, making them difficult to maintain.

[0048] The risk map approach proposed in this paper further provides a dynamic way to consider all potential uncertainties of the environment and the capabilities of the vehicle / ADS itself when assessing the safety of the proposed path. Compared to current methods, the risk map will contain information that allows for dynamic estimation of risk from all parts of the ADS, including sensors, perception systems, and the vehicle platform and actuators. This will not only allow estimates from the risk map to be more accurate compared to static models, but also allow the examined path to consider the more complex behavior of surrounding objects and some of the central problems of overcoming uncertainties surrounding static models. The risk map aggregates uncertainties from all sources to create a holistic view of the risk. It considers uncertainties in sensor readings, perception system predictions, object models (scene / behavioral models), and vehicle platform capabilities (braking, steering, etc.).

[0049] Understanding the safe execution paths of an Adaptive Distributed Module (ADS) is a key component of security-driven ADS. In any given situation, an ADS is exposed to an excessive number of risks and uncertainties. Quantifying each of these risks and uncertainties in the currently perceived state of the world provides the aforementioned "risk map" for the ADS. Figures 3 to 5 The diagram illustrates a schematic example of a risk map. This risk map (where each location around an ADS is associated with a risk) is used accordingly to assess the risk of traversing locations on the map. In other words, given a path, the risk map is used to assess the risk of that path. Therefore, the risk map is used to calculate the safety level (i.e., the risk level) of a path.

[0050] Figure 1 This is a schematic flowchart representation of a method 100 for the automated development of a path planning development module for a vehicle, performed by an onboard computing system according to some embodiments of the present invention. The vehicle is equipped with an Automated Driving System (ADS), which in this context includes, for example, all the different levels of automation defined by SAE J3016 levels (0-5) of driving automation (and especially levels 4 and 5). In other words, the term ADS encompasses both Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD).

[0051] More specifically, Method 100 involves a closed-loop evaluation of the vehicle's path planning module and its subsequent updates. The term "closed-loop" evaluation can also be referred to as "real-time testing" and can be understood accordingly as an evaluation of the output of the path planning development module, whose generated paths are intended to be executed by the vehicle in a real-world scenario.

[0052] Method 100 includes obtaining 101 candidate paths from a path planning development module. The term "path planning development module" can also be referred to as a "module in test" (MUT), meaning "new" (under development) software and / or hardware for path planning features in automotive applications. In other words, in this context, a path planning module can be understood as software and / or hardware configured to generate candidate paths based on perception data (e.g., raw or processed sensor data), where the path planning module is currently "under development" and not yet "in production" (e.g., unverified / validated). A vehicle, or more precisely, a vehicle's ADS, can naturally be equipped with a "production path planner," i.e., a path planning module, which is part of a production platform / system configured to generate paths to be executed by the vehicle's ADS. The term "obtain" will be interpreted broadly herein and encompasses receiving, retrieving, collecting, and acquiring.

[0053] Accordingly, the path planning development module is configured to generate candidate paths for vehicle execution based on a path planning model (e.g., a machine learning algorithm or artificial neural network) and data indicating the vehicle's surrounding environment. Thus, data indicating the surrounding environment (e.g., perception data) serves as input to the path planning model, and the output of the path planning model is one or more candidate paths. Data indicating the vehicle's surrounding environment can, for example, be perception output generated by the vehicle's perception system (e.g., fused sensor data). However, in some embodiments, data indicating the vehicle's surrounding environment can be sensor data obtained directly from one or more onboard sensors. Furthermore, the path planning model can also utilize the vehicle's actuation capabilities as input. As used herein, the term "actuation capability" can include one or more of the vehicle's braking capability, acceleration capability, and steering capability, as readily understood by those skilled in the art.

[0054] Further, method 100 includes obtaining a risk map of the vehicle's surrounding environment, wherein the risk map is formed based on the vehicle's actuation capability and the location of free space regions in the surrounding environment. More specifically, the actuation capability includes an uncertainty estimate of the actuation capability, and the location of the free space regions includes an uncertainty estimate of the estimated location of the free space regions. The risk map further includes risk parameters for each of a plurality of region segments included in the vehicle's surrounding environment. Furthermore, the risk map includes a time component that indicates the temporal evolution of the risk parameters of the region segments based on the predicted time evolution of the free space regions, at least within a duration defined by the predicted duration of the candidate path. (Refer to below...) Figures 3 to 5 Let's discuss risk maps in more detail.

[0055] For example, each risk parameter may indicate at least one of the following: the probability of an accident event occurring if the vehicle occupies the associated area segment, and the probability of violating a predefined safety / risk threshold. Furthermore, the location of the free space area may include the location of external objects in the vehicle's surrounding environment. Additionally, the estimated location of external objects may include the uncertainty of the external object's location, the trajectory of any dynamic object within the external object, and the uncertainty of the estimated trajectory of the dynamic object. In other words, the free space area in this context can be understood as an area in the vehicle's surrounding environment where no objects (e.g., other vehicles, pedestrians, obstacles, animals, bicycles, static objects, etc.) exist. Therefore, the obtained location of the free space area can be understood as an estimate of the area where no external objects (static and dynamic objects) exist, and an estimate of the uncertainty of this determination (i.e., the risk of the determined location of the free space area is not actually free). Further, in some embodiments, the location of the free space area further includes an estimate of the location of the "drivable area," wherein, in addition to the estimate of the area where no objects exist (and the associated uncertainty), the estimate also includes the location / presence (and associated uncertainty) of the road surface or "road-like" surface.

[0056] For example, the free space area can be derived from sensor data from one or more onboard sensors configured to monitor the vehicle's surroundings. However, sensor data can also originate from other sensors near the vehicle, such as those mounted on other vehicles or infrastructure components and obtained via V2V or V2X communication networks.

[0057] Therefore, in some embodiments, the location of the free space region is determined by a dedicated module of the vehicle, which is configured to acquire sensor data indicating the vehicle's surrounding environment and derive the location of the free space region relative to the vehicle based on the sensor data. Thus, there is no need for intermediate steps or layers of detection objects before obtaining the location of the free space region; that is, the "free space region" can be obtained directly. For example, a signal emitted by a lidar can propagate a certain distance through space before reflecting off a surface, and the area between the lidar and that surface can then be defined as the "free space region" without any operation or step to define the surface from which the signal reflects.

[0058] Method 100 further includes determining the risk value of the candidate path 103 based on risk parameters of the area segment to which the candidate path intersects. In other words, the check of the candidate path obtained 101 is performed based on the risk map obtained 102.

[0059] The steps to determine the 103 risk value may include: (a) checking that the candidate path is considered sufficiently safe to execute in the current time instance, and / or (b) checking that the candidate paths provided by the path planning development module have a sufficiently low risk over time. The first metric marked (a) above is referred to herein as the “Total Risk Value,” while the second metric marked (b) above is referred to herein as the “Overall Risk Value.” In this context, the term “total” is considered to encompass “cumulative.”

[0060] However, if the determined risk value meets at least one risk criterion, i.e., if the candidate path 101 is obtained through the “risk check” 104 according to the risk map, the method further includes generating at the output a first control signal 105 indicating instructions for the ADS to execute the candidate path. More specifically, the instructions for executing the candidate path 101 are generated by the decision and control block of the ADS, and appropriate control signals are sent to one or more actuators of the vehicle to control the vehicle’s steering, acceleration, and deceleration, etc.

[0061] Further, method 100 includes obtaining at least one quality parameter of the candidate path executed 106, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort during the execution of the candidate path. Some examples of applicable quality parameters include lateral acceleration, longitudinal acceleration, vertical acceleration, rate of change of acceleration, vibration, steering oscillation frequency, path length, roll acceleration, pitch acceleration, yaw acceleration, fuel consumption, and other key performance indicators (KPIs) for ride quality. Additionally, in some embodiments, at least one quality parameter includes at least one threat assessment metric, such as, for example, Brake Threat Number (BTN), Time to Collision (TTC), Time to Brake (TTB), Headway, and Time After Intrusion (PET).

[0062] Method 100 further includes: determining, 107, a cost function indicative of the performance of the path planning development module based on at least one obtained quality parameter; and updating, 108, one or more parameters of the path planning model by means of an optimization algorithm configured to optimize the determined cost function. Here, it can be noted that optimizing the cost function does not necessarily mean that the quality parameter should always be minimized or maximized. For example, if a threat assessment metric such as BTN or PET is used as one of the quality parameters employed in the cost function, simply minimizing that parameter would not be advantageous. This is primarily because, since the “safest” maneuver is typically to bring the vehicle to a complete stop, the algorithm is likely to be trained to bring the vehicle to a complete stop. Therefore, a predefined specification (i.e., min) can be used. x (x, x) threhold ) or max x (x, x)threhold )) Optimize the parameters toward a specific maximum or minimum value.

[0063] In this disclosure, the terms cost function, loss function, and error function are used interchangeably. The purpose of a cost function, as defined herein, is to provide means for updating a path planning model to maximize the desired output and minimize the undesired output from the path planning model. More specifically, the cost function is preferably set relative to one or more predefined objectives (i.e., predefined target values). For example, one or more (predefined) defined quality parameter thresholds may be used to form the “objective” of the cost function. In other words, the purpose of the cost function is to minimize or maximize one or more quality parameter values ​​associated with a candidate path, but preferably only up to a specific point. As mentioned above, this is to avoid the potential situation where the path planning is optimized to “achieve” the “safest” option at a static level (or at least very conservatively “driving,” which would be detrimental to the user experience). However, some fixed constraints may also be imposed on the optimization algorithm (e.g., the vehicle must move from A to B) to avoid this situation.

[0064] Therefore, in some embodiments, one or more quality parameters may form constraints on the optimization algorithm, wherein an example of such quality parameter (forming constraints) may be path completion as mentioned above. In other words, the optimization algorithm is further configured to employ one or more quality parameters as constraints in the optimization. Additionally or alternatively, the optimization algorithm may be further configured to employ at least one risk criterion (and / or some threat assessment metrics) as constraints in the optimization.

[0065] In other words, the cost function can be determined / defined using a combination of performance and safety-related functions. To incorporate safety, the determined risk values ​​(total and / or summative) of the path can be used. Additionally, quality / comfort and other performance-related functions can include fuel consumption, comfort level, rate of change of acceleration and / or acceleration used, etc. These different aspects can be included in the "learning" process as a weighted combination in the cost function, or as constraints on the optimization problem / algorithm.

[0066] In addition, an equivalent solution would be to determine the “reward function” corresponding to 107 in the evaluation, and then update one or more model parameters of the 108 path planning model by an optimization algorithm configured to maximize the determined “reward function”.

[0067] Returning to the determination of risk value 103, in some embodiments, the step of determining the risk value 103 includes determining the total risk value of the candidate path 109 based on risk parameters of a set of area segments to intersect with the candidate path. Therefore, at least one risk criterion includes a total risk threshold. Furthermore, the step of generating a first control signal 105 at the output indicating instructions for the ADS to execute the candidate path is performed only if the determined total risk value 109 is lower than the total risk threshold. Additionally, in some embodiments, the step of determining the total risk value 109 includes summing the risk parameters of a set of area segments to intersect with the candidate path 110.

[0068] Further, according to some embodiments, the step of determining the risk value 103 further includes determining the total risk value of the 111 path planning development module based on risk parameters of a set of area segments intersecting with at least one previously executed path. Here, the set of area segments intersecting with at least one previously executed path is associated with a previously obtained risk map for the same period as the at least one previously executed path. In other words, for each previously executed path, a set of area segments belonging to the risk map of that particular path is used to estimate the total risk value. It should be noted that the term "executed path" will be interpreted broadly and includes "at least a portion of the path." As will be readily understood by those skilled in the art, the path length can depend on various factors, and its execution can also depend on various factors. More specifically, while some may consider a path to be executed only when the "complete" path across the entire planning scope is executed, others may consider each portion of the executed "complete path" (at a specific sampling rate) to constitute a separate "executed path." However, the term "executed path" as used herein encompasses both interpretations. Additionally, it should be noted that if there is no “previously executed path”—that is, the candidate path being evaluated is the first path of the driving session—the total risk can be ignored as a measure of that “first” instance, or set to a predefined (starting) value below the associated threshold.

[0069] In addition, at least one risk criterion includes a total risk threshold, and the step of generating a first control signal at the output indicating instructions for ADS to execute candidate paths is performed only when the total risk value is lower than the total risk threshold.

[0070] Further, in some embodiments, the step of determining the total risk value 111 includes averaging the risk parameters of a set of area segments that intersect with at least one previously executed path 112 in order to determine the average risk exposure of the path planning development module 115, wherein the average risk exposure defines the total risk value.

[0071] As previously mentioned, the group of regional segments that intersect with at least one previously executed path are associated with a risk map previously obtained for the same period as that at least one previously executed path.

[0072] Furthermore, in some embodiments, the step of averaging the risk parameter includes, for each time sample:

[0073] • Determine the current risk value of the 113 path planning development module based on a set of area segments that intersect with the most recently executed candidate path since previous time samples.

[0074] The current composite risk value is determined by summing the current risk value with the previously calculated composite risk value of the previous time sample. Accordingly, in some embodiments, the previous composite risk value may be defined as the sum of the risk parameters of the area segments that intersect with the already executed paths.

[0075] • The 115 average risk exposure is determined by dividing the current composite risk value by the total number of samples used up to and including the current time sample. Here, the total number of samples can be "the total number of previously executed paths".

[0076] In other words, for each time sample (after a path for execution has been selected), the risk for that time instance (i.e., the aggregated risk value described earlier) is calculated by summing the risk parameters of the areas intersecting the planned path based on the risk map at that time instance. This "current risk value" is then summed with the previously calculated risk values ​​for each executed path to obtain the "current composite risk value." It should be noted that these previously calculated risk values ​​are determined based on the risk maps of these time samples. The total risk value is then calculated by dividing the "current composite risk value" by the total number of samples used up to and including the current time sample.

[0077] Continuing, in some embodiments, method 100 includes obtaining an alternative path from the vehicle's production platform. Accordingly, the production platform includes a validated / certified path planning development module configured to generate the alternative path. Then, if the determined risk value fails to meet at least one risk criterion, method 100 includes generating a second control signal 116 at the output to execute the obtained alternative path, and optionally, also includes disabling the path planning development module 117. Accordingly, if the path planning development module cannot provide a sufficiently safe path, the alternative path can be understood as a fallback option, and therefore the ADS is configured to provide an alternative path such that the ADS always has a "safe option" to rely on. The alternative path can be provided by the production platform's path planning module or any other dedicated module used to generate alternative paths for the vehicle.

[0078] Furthermore, in some embodiments, method 100 further includes: if the determined risk value fails to meet at least one risk criterion, generating a second control signal at the output to enable the path planner of the vehicle's production platform for generating candidate paths to be executed by the vehicle. In other words, if the total risk exceeds a threshold and / or if the total risk exceeds a threshold, the vehicle can be configured to return to the path planning module dependent on the production platform to generate a path to be executed by the vehicle.

[0079] Furthermore, in the joint learning scheme, the parameters of the update 108 for each of the multiple vehicles can be advantageously integrated in a central unit or cloud unit, thereby allowing a “global update” to be pushed to the entire vehicle fleet. Therefore, in some embodiments, method 100 further includes transmitting parameters of one or more updates 108 of the path planning model to a remote entity (e.g., a back-end or fleet management system). Additionally, method 100 may include receiving a set of globally updated parameters of the path planning model of the path planning module from the remote entity. This set of globally updated parameters is accordingly based on information obtained from multiple vehicles including the corresponding path planning development module. Method 100 may then include updating the path planning model of the path planning module 107 based on the received set of globally updated parameters.

[0080] The path planning model can be updated or trained "in real time" at a specified update frequency, either while the vehicle is moving or when it is stationary. In the latter case, and according to some embodiments, the output from the cost function can be stored in a suitable data storage medium and used later in a more "rigid" update process (e.g., at the end of a trip or while the vehicle is charging). Alternatively, the model parameters are updated only at a remote entity (i.e., in the background), so the output from the cost function is sent to the remote entity, where it is integrated across the entire vehicle fleet and subsequently pushed to each vehicle as globally updated model parameters. However, in an alternative embodiment, the quality parameters used to form the cost function are aggregated and stored intermediately, whereby the cost function is determined based on the aggregated data.

[0081] Optionally, the executable instructions for performing these functions are included in a non-transitory computer-readable storage medium or in another computer program product configured to be executed by one or more processors.

[0082] Figure 2 This is a schematic block diagram representation of an apparatus or system 10 for the automatic development of a path planning development module 210 for a vehicle equipped with ADS, according to an embodiment of the present invention. Generally speaking, Figure 2The information flow is described through exposure to events in the vehicle's surrounding environment, exposure to the path evaluation process, and further exposure to transmission and subsequent integration in the "background" 2. The apparatus or system 10 includes various control circuits configured to perform functions of the methods disclosed herein, wherein the functions may be included in a non-transitory computer-readable storage medium or other computer program product configured to be executed by the control circuits. Figure 2 This invention is used to better illustrate the invention by depicting various “modules,” each of which links to one or more specific functions described above.

[0083] It should be noted that the "candidate paths" generated by the path development module 210 and compared in the "risk assessment module" 230 may include multiple candidate paths.

[0084] System 200 includes a path planning development module 210, configured to generate candidate paths for the vehicle based on a path planning model and data indicating the vehicle's surrounding environment (such as perception data generated by the vehicle's perception system 201, for example). Accordingly, perception system 201 is configured to generate perception outputs based on sensor data acquired from one or more onboard sensors over a time period. Sensor data can be, for example, outputs from one or more of radar devices, lidar devices, cameras, and ultrasonic sensors. In other words, the "perception system" (i.e., the perception system of the production platform) will be understood in this context as a system responsible for acquiring raw sensor data from onboard sensors such as cameras, lidar and radar, and ultrasonic sensors, and transforming that raw data into scene understanding that includes state estimation and prediction.

[0085] Furthermore, the vehicle's production platform / system is configured to provide a reference frame for evaluating candidate paths. The reference frame is configured to indicate one or more risk values ​​associated with a candidate path when it is applied within the reference frame. Here, the reference frame is in the form of a risk map generated by the risk map compilation module 220.

[0086] The risk map compilation module 220 is configured to generate a risk map based on the vehicle's actuation capabilities and the location of free space regions in the surrounding environment. Actuation capabilities include an uncertainty estimate of the actuation capabilities, and the location of the free space regions includes an uncertainty estimate of the estimated location of the free space regions. In other words, the risk map compiler 220 is configured to compile a risk map based on the detection and prediction of uncertainties, including those from the perception system 201 and capabilities and uncertainties (e.g., steering capability, braking capability, etc.) reported by the vehicle platform 203. The risk map can be understood as a virtual map of the vehicle's surrounding environment with multiple defined segments, each segment associated with a corresponding risk parameter.

[0087] Furthermore, the risk map has a time component that indicates the temporal evolution of risk parameters for a region segment based on the predicted temporal evolution of the free space region, at least within a duration defined by the predicted duration of the candidate path. The predicted temporal evolution can be based, for example, on sensed data and one or more prediction models (e.g., trajectory prediction, statistical models, etc.).

[0088] According to the illustrative example, the obtained actuation capability may include braking capability and the type of uncertainty estimate or "error tolerance" of the obtained braking capability. For example, if the obtained braking capability indicates that the vehicle can come to a complete stop within a distance of 150 meters (assuming emergency braking actuation), the uncertainty estimate of that estimate may include an error tolerance of ±15 meters (i.e., ±10%). As previously mentioned, the actuation capability can be given from a statistical model of one or more actuation parameters, where the uncertainty estimate of each actuation parameter can be defined by the standard deviation in these statistical models.

[0089] Further, the risk map compiler 220 (e.g., from the vehicle's perception system 201) receives the location of a free space region in the vehicle's surrounding environment, wherein the obtained location of the free space region includes an estimate of the uncertainty of the estimated location of the free space region. As previously mentioned, a free space region in this context can be understood as an area in the surrounding environment of the vehicle where no objects (e.g., other vehicles, pedestrians, obstacles, animals, bicycles, static objects, etc.) are present. Therefore, the obtained location of the free space region may include a determined location of external objects (static and dynamic objects), a determined trajectory of dynamic objects, and an estimate of the uncertainty of that determination (i.e., the risk of the determined location of the free space region is not actually free). Further, in some embodiments, the location of the free space region further includes an estimate of the location of a "drivable area," wherein, in addition to an estimate of the area where no objects are present (and the associated uncertainty), the estimate also includes the location / presence (and associated uncertainty) of a road surface or "road-like" surface.

[0090] The compiled risk map, along with candidate paths, is provided to the risk assessment module 230, which is configured to assess the "safety" of the candidate paths within a corresponding time period based on the generated risk map. In other words, the risk assessment module 230 is configured to determine the risk value of the candidate paths based on risk parameters of the regions that will intersect with the candidate paths. As previously mentioned, the risk value can be assessed based on two metrics (i.e., aggregate risk value and total risk value), each determined by the aggregate risk estimator 231 and the total risk estimator 232, respectively. More specifically, estimators 231 and 232 are configured to assess the risk of the candidate paths according to one or more criteria, and if a candidate path meets the criteria, a first control signal is generated, indicating instructions for the ADS to execute the candidate path.

[0091] Therefore, safety can be determined through simple inductive variables. More specifically, if it is assumed that the vehicle is now in a sufficiently safe state and the provided path (according to the risk map) is assessed as having a sufficiently low risk, then the process will subsequently end with the state shown by the risk map as sufficiently safe, and thus it is possible to return to the beginning of the variables. Furthermore, the "risk map method" can also allow for capturing situations where the risk varies along consecutive paths (this should be captured in the overall risk estimate, and the evaluation of the path planning development module can be paused at least temporarily).

[0092] Continuing, once the candidate path has passed the risk / safety assessment, a signal is sent to the control block 202 of the ADS, which is configured to instruct the vehicle platform to execute the candidate path. It should be noted that the candidate path can be sent directly from the path planning development module to the control block 202, while the risk assessment module 230 simply provides a confirmation signal to the control block 202. However, in some example embodiments, the path development module 210, along with the "production path planner," is a "part" or "submodule" within the control block 202 of the ADS. However, for the purpose of clarifying the invention, some modules / functions such as the path development module 210 are illustrated herein. The perception module 201, control block 202, and vehicle platform 203 block, as illustrated and described herein, can be understood as parts / blocks / modules / systems of the "ADS" or "production platform / system."

[0093] When a path is being executed, or when a path has been executed, the path evaluation engine 240 is configured to obtain at least one quality parameter of the candidate path being executed, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort. Quality parameters may include, for example, external quality parameters and internal quality parameters. In this context, external quality parameters can be understood as parameters indicating the perceived quality of the path based on data obtained from the vehicle's perception system 201. This may include, for example, distances to external objects, distances to road edges, distances to lane markings, lane positions, driver monitoring system (DMS) data, and various threat assessment metrics. In this context, internal quality parameters can be understood as parameters indicating the perceived quality of the path based on data obtained from the vehicle's IMU or other "internal" sensors. This may include, for example, lateral acceleration, longitudinal acceleration, vertical acceleration, rate of change of acceleration, roll acceleration, pitch acceleration, yaw acceleration, and fuel / energy consumption.

[0094] The path evaluation engine 240 is further configured to determine a cost function indicative of the performance of the path planning development module based on at least one obtained quality parameter. The cost function is then provided to the learning engine 250, which is configured to update one or more parameters of the path planning model using an optimization algorithm configured to optimize the determined cost function. Furthermore, the learning engine 250 can be configured to send one or more updated parameters of the path planning model of the path planning module to a remote entity 2, and then receive a set of globally updated parameters of the path planning model of the path planning module from the remote entity. This set of globally updated parameters is based on information obtained from multiple vehicles including the corresponding path planning modules. Subsequently, the learning engine 75 can update the path planning model of the path planning module based on the received set of globally updated parameters. This can be interpreted as an "integration" process across the entire vehicle fleet.

[0095] The path planning model in the path planning development module can be updated until a predefined maturity level or learning saturation is reached, thus replacing the "production path planner". This can be indicated, for example, by new updates to the model parameters that cause them to oscillate around a certain resting point. Alternatively, or additionally, the path planning development module can be used to set a threshold for "sufficient mileage," and once this threshold is reached (possibly while one or more other activities and conditions are met), the model can be deemed mature enough to replace the currently used "production path planner".

[0096] Figure 3 and Figure 4These are two schematic top views of a risk map 40 at two consecutive time steps according to an embodiment of the invention, some of which are example components 41-46” contributing to the risk map 40. Furthermore, the planned execution paths 47a-47b of the ego vehicle are indicated in the risk map 40. More specifically, dashed arrows indicate “candidate paths” 47a from previous time instances / samples, and the current candidate path 47b is... Figure 4 The arrow pointing in front of the vehicle.

[0097] Furthermore, the risk map 40 includes information indicating the estimated braking capacity 43 of vehicle 41, including uncertainties 43' and 43''. Further, the risk map 40 includes the vehicle's geographic location 41 on the map, the uncertainty estimate 42 of geographic location 41, the locations of external objects 44 and 46, the uncertainties of the locations of external objects 44', 44'', and 46'', the trajectory 45 of dynamic object 44, and the uncertainties 45' and 45'' of trajectory 45. The estimated uncertainties can be calculated, for example, based on a model (predefined or self-learning / machine learning). This model defines the tolerances or error margins in measurements provided from the vehicle's sensors (e.g., cameras, radar, lidar, ultrasonic sensors, etc.). Consequently, the resulting risk map 40 also considers uncertainties inherent in such measurements, such as those arising from sensor manufacturing tolerances and noise, affecting the vehicle's own worldview. This makes the overall risk estimate more accurate and reliable, thus more precisely reflecting the actual risk exposure of the vehicle's ADS. However, in some embodiments, as will readily understand to those skilled in the art, the estimated uncertainties are inherently provided by the perception system of the production platform and are subsequently incorporated into the generated risk map 40.

[0098] Figure 5 It is relative to the embodiments of the present invention. Figure 4 The diagram depicts a schematic top view of the risk map at subsequent time steps / samples, where area segment 51 is indicated by grid 50. As previously described, to determine the total risk value of candidate path 47, the risk values ​​associated with the area segments intersecting with planned path 47 at 52 can be summed, and / or an average of the risk values ​​associated with the area segments intersecting with planned path 47 at 52 can be formed.

[0099] also, Figure 5The lower left corner shows a schematic top view of the risk map and serves to illustrate how the total risk value can be determined according to some embodiments of the invention. More specifically, candidate paths 47 extending through the risk map 40 are described, which further has overlapping grid frames 50 to illustrate how area segments are formed. As previously described, the risk map 40 has a plurality of area segments 51, each associated with a risk parameter indicating at least one of the probability of an incident event occurring if a path were to intersect the associated area segment and the probability of a violation of a predefined safety threshold. The “value” of the risk parameter is indicated in each box 51 by a pattern within the box 51. Accordingly, the risk map 40 depicts specific “high-risk” area segments 44, “low-risk” area segments 43, and values ​​in between.

[0100] The risk of candidate path 47 (i.e., the risk value associated with the candidate path) can be assessed using risk map 40 in one or more ways. For example, when candidate path 47 is executed on risk map 40, the total risk of candidate path 47 (i.e., the integral of the path on the risk map) can be used. Accordingly, the risk value of the region segment 51 that intersects with candidate path 47 52 can be totaled to derive the total risk or average risk of candidate path 47.

[0101] Regarding total risk, the total risk of driving can be obtained by summing the risks of each planned and executed individual path and normalizing them over driving time. Alternatively, if equidistant time steps are used throughout the driving process, the total risk of the planned paths can be normalized by the number of planned paths. The total risk can therefore be schematically described as:

[0102]

[0103] Here, T is the total driving time during the driving period considered (so far), and the integral is used to account for a more precise calculation of the risk for each planned path, rather than simply summing the risk parameters over the intersecting grid boxes as mentioned in the example above. It should be noted that the integral in Equation (1) is the path integral, where the parameter (x) is obtained along the executed path. Furthermore, by considering the number of samples included in the risk calculation itself, the system can provide a continuous estimate (e.g., per unit of time) of the total risk that the path planning development module has been exposed to. Therefore, the total risk, along with the number of samples used in previous calculations, is fed back for the next iteration to arrive at the total risk.

[0104] In other words, for each time sample (after a path for execution has been selected), the risk (i.e., the current risk value) for that time instance is calculated by summing the risk parameters of the areas intersecting with the planned path based on the risk map at that time instance. This "current risk value" is then summed with the previously calculated risk values ​​for each executed path to obtain the "current composite risk value." It should be noted that these previously calculated risk values ​​are determined based on the risk maps of these time samples. The total risk value is then calculated by dividing the "current composite risk value" by the total number of samples used up to and including the current time sample.

[0105] However, even if it may be implicit, because the total risk is fed back to the total risk estimator, there is an additional step of multiplying the previously calculated total risk by the number of samples used to calculate the total risk before the new “current risk value” is added to the number of samples used to calculate the total risk. This is presented more formally in the following equations (2) and (3).

[0106]

[0107]

[0108] Here, TR t CR represents the total risk value at time instance t; t This represents the current composite risk value, which is the sum of the risk parameters that the ADS has been "exposed to" up to and including time instance t; and T t Indicates that it is used for calculation until

[0109] And it includes the total number of samples of the current composite risk value up to time instance t. Equation (3) describes the previously mentioned process, where the total risk of previous time instances (i.e., the total risk of feedback) is added to the "current risk value" (PR). t+1 Before that, the total risk of previous time instances (i.e., the total risk of feedback) is multiplied by the number of samples used to calculate the total risk. PR t+1 This represents the risk value of the planned path at time instance t+1.

[0110] Accordingly, as the vehicle maneuvers through the resulting risk map, a periodically / continuously updated and quantitative estimate of the risks the vehicle has been exposed to during those maneuvers can be obtained. By continuously / periodically estimating the total risk throughout driving, it is possible to further determine whether the path planning development module has violated any of the safety requirements. This is particularly advantageous for ADS at Level 4 and above (according to SAE J3016 Level of Driving Automation), as it must accommodate the need to execute (unintended) "high-risk" paths, since the responsibility of the ADS is to drive the vehicle in all situations where the ADS is active. However, if the risk level continues to rise above an acceptable level, this should be acknowledged and appropriate actions should be taken, such as requiring the driver to withdraw from or disengage the path planning development module to avoid an accident.

[0111] Therefore, return to Figures 3 to 4 The following is a series of examples of how to determine the total risk value and how the risk value changes over time, according to embodiments of the present invention. Figure 3 The risk diagram 40 represents the risk at the first instance t. Accordingly, to calculate the total risk, the vehicle's control system is configured to calculate the current risk value of the ADS (or path planning development module) based on a set of area segments intersecting the current planned path 47a. In this example, the current risk value is assumed to be 10 based on a set of area segments intersecting the current planned path 47a. -8 Since there is no previously calculated total risk value or composite risk value, the current risk value is the total risk value at time t.

[0112] Continue to Figure 4 This represents the risk map 40 at the subsequent second time instance t+1. Here, a new path 47b has been planned and selected for execution. Similarly, the current risk value of ADS is determined based on a set of area segments intersecting with the currently planned path 47b. In this time instance, assume the current risk value is 10. -7 Accordingly, the current composite risk value is determined by summing the current risk value with the previously calculated composite risk value of the previous time sample, and the total risk value at time t+1 is determined by dividing the current composite risk value by the total number of samples used up to and including the current time sample. Therefore, the total risk value at time t+1 is given by the following equation (4).

[0113]

[0114] Furthermore, in a subsequent (undescribed scenario) time instance, risk map 40 is located at a third time instance t+2. Following the same procedure as above, it is assumed that the current risk value at the third time instance is 10.-7 Furthermore, using equation (3), the total risk value at time t+2 is given by equation (5) below.

[0115]

[0116] Accordingly, as previously illustrated, the “total risk value” provides a more or less continuous estimate of the risk exposure of the path planning development module throughout driving, and also provides how the risk exposure evolves over time. As previously mentioned, unlike other existing methods, this approach provides a continuous measurement between risk-free and collision-free periods to assess the performance of automated / autonomous vehicles. As presented in this paper, the estimation of total risk leads to a fine-grained measurement of the risks that the path planning development module (and by extension, the vehicle and its occupants) have been exposed to. This not only allows for the addition of accident-free hours but also enables the comparison of individual driving hours.

[0117] Figure 6 This is a schematic side view of a vehicle 1 including an automated development device 10 for a path planning development module for a vehicle equipped with ADS, according to an embodiment of the present invention. The vehicle 1 further includes a perception module / system 6 (i.e., the perception system of the production platform) and a positioning system 5. The positioning system 5 is configured to monitor the vehicle's geographic location and heading, and may be in the form of a Global Navigation Satellite System (GNSS) such as GPS. However, the positioning system may alternatively be implemented as a Real-Time Kinematic (RTK) GPS for improved accuracy.

[0118] More specifically, the perception module / system 6 can refer to any well-known system and / or function, such as that included in one or more electronic control modules and / or nodes of vehicle 1, adapted and / or configured to interpret driving-related sensor information of vehicle 1 to identify, for example, obstacles, vehicle lanes, relevant signs, appropriate navigation paths, etc. Thus, in conjunction with sensor information, the exemplary perception system 6 can be adapted to rely on and obtain input from multiple data sources (such as vehicle imaging, image processing, computer vision, and / or in-vehicle networking, etc.). This exemplary sensor information can be derived, for example, from one or more optional ambient environment detection sensors 6a-6c included in and / or disposed on vehicle 1. The ambient environment detection sensors 6a-6c can be represented by any arbitrary sensor adapted to sense and / or perceive the surrounding environment and / or location of vehicle 1, and can, for example, refer to one or more of radar, lidar, sonar, cameras, navigation systems (e.g., GPS), odometers, and / or inertial measurement units, or combinations thereof.

[0119] Device 10 includes one or more processors 11, memory 12, sensor interface 13, and communication interface 14. Processor 11 may also be referred to as control line 11 or control circuit 11. Control circuit 11 is configured to execute instructions stored in memory 12 to perform an automated development method f for a path planning module for vehicle 1 according to any embodiment disclosed herein. In other words, memory 12 of device 10 may include one or more (non-transitory) computer-readable storage media for storing computer-executable instructions, which, when executed by one or more computer processors 11, may cause the computer processors 11 to perform, for example, the techniques described herein. Memory 12 may optionally include high-speed random access memory, such as DRAM, SRAM, DDR RAM, or other random access solid-state memory devices; and may optionally include non-volatile memory, such as one or more disk storage devices, optical disk storage devices, flash memory devices, or other non-volatile solid-state memory devices.

[0120] The device 10 includes control circuitry 11 configured to obtain candidate paths from a path planning development module. The path planning development module is configured to generate candidate paths for the vehicle to execute based on a path planning model and data indicating the vehicle's surrounding environment.

[0121] The control circuit 11 is further configured to obtain a risk map of the vehicle's surrounding environment, wherein the risk map is formed based on the vehicle's actuation capability and the location of free space regions in the surrounding environment. The actuation capability includes an uncertainty estimate of the actuation capability, and the location of the free space regions includes an uncertainty estimate of the estimated location of the free space regions. The risk map includes risk parameters for each of a plurality of region segments included in the vehicle's surrounding environment. Furthermore, the risk map further has a time component that indicates the temporal evolution of the risk parameters of the region segments based on the predicted time evolution of the free space regions, at least within a duration defined by the predicted duration of the candidate path.

[0122] The control circuit 11 is further configured to determine the risk value of the candidate path based on the risk parameters of the area segment to which the candidate path intersects, and if the determined risk value meets at least one risk criterion, the control circuit 11 is further configured to:

[0123] • Generate a first control signal at the output that indicates the instructions for executing candidate paths in ADS.

[0124] • Obtain at least one quality parameter of the candidate path to be executed, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort of the executed candidate path.

[0125] • Determine the cost function that indicates the performance of the path planning development module based on at least one obtained quality parameter.

[0126] • Update one or more parameters of the path planning model using an optimization algorithm configured to optimize a determined cost function.

[0127] Furthermore, vehicle 1 can connect to an external network(s) 20 via a wireless link, for example (e.g., for retrieving map data). The same or other wireless links can be used to communicate with other vehicles 2 in the vicinity or with local infrastructure components. Cellular communication technologies can be used for long-distance communication, such as to external networks, and if the cellular communication technology used has low latency, it can also be used for communication between vehicles, between vehicles (V2V), and / or between vehicles and infrastructure (V2X). Examples of cellular radio technologies are GSM, GPRS, EDGE, LTE, 5G, and 5G NR, as well as future cellular solutions. However, some solutions use medium- to short-range communication technologies, such as wireless local area networks (LANs), for example, solutions based on IEEE 802.11. ETSI is working on cellular standards for vehicle communication, and 5G, for example, is considered a suitable solution due to its low latency, high bandwidth, and efficient handling of communication channels.

[0128] The present invention has been presented above with reference to specific embodiments. However, other embodiments besides those described above are also possible and within the scope of the invention. Within the scope of the invention, method steps different from those described above, performed by hardware or software, can be provided. Therefore, according to exemplary embodiments, a non-transitory computer-readable storage medium is provided that stores one or more programs configured to be executed by one or more processors of a vehicle control system, the programs including instructions for performing the method according to any of the embodiments discussed above. Alternatively, according to another exemplary embodiment, a cloud computing system can be configured to perform any of the methods presented herein. The cloud computing system may include distributed cloud computing resources that jointly perform the methods presented herein under the control of one or more computer program products.

[0129] Generally, computer-accessible media can include any tangible or non-transitory storage medium or memory medium, such as electronic, magnetic, or optical media (e.g., a disk or CD / DVD-ROM coupled to a computer system via a bus). As used herein, the terms “tangible” and “non-transitory” are intended to describe computer-readable storage media (or “memory”) excluding those that transmit electromagnetic signals, but are not intended to otherwise limit the types of physical computer-readable storage devices covered by the phrases “computer-readable medium” or “memory.” For example, the terms “non-transitory computer-readable medium” or “tangible memory” are intended to cover types of storage devices that do not necessarily store information permanently, including, for example, random access memory (RAM). Program instructions and data stored in a non-transitory form on a tangible computer-accessible storage medium can be further transmitted via a transmission medium or signals (such as electrical signals, electromagnetic signals, or digital signals) that can be transmitted via communication media (such as networks and / or wireless links).

[0130] One or more processors 11 (associated with device 10) may be or include any number of hardware components for performing data or signal processing or for executing computer code stored in memory 12. Device 10 has an associated memory 12, and memory 12 may be one or more devices for storing data and / or computer code used to perform or facilitate the various methods described herein. Memory may include volatile or non-volatile memory. Memory 12 may include database components, object code components, script components, or any other type of information structure for supporting the various activities described herein. According to exemplary embodiments, any distributed or local memory device may be used with the systems and methods described herein. According to exemplary embodiments, memory 12 (e.g., via circuitry or any other wired, wireless, or network connection) may be communicatively connected to processor 11 and includes computer code for performing one or more processes described herein.

[0131] It should be understood that sensor interface 13 can also provide the possibility of acquiring sensor data directly or via dedicated sensor control circuitry 6 in the vehicle. Communication / antenna interface 14 can further provide the possibility of transmitting output to a remote location (e.g., a remote operator or control center) via antenna 8. Furthermore, some sensors in the vehicle can communicate with system 10 using local network settings (such as CAN bus, I2C, Ethernet, and fiber optics). Communication interface 14 can be arranged to communicate with other control functions of the vehicle and can therefore also be considered a control interface; however, a separate control interface (not shown) can also be provided. Local communication within the vehicle can also be wireless, using protocols such as WiFi, LoRa, Zigbee, Bluetooth, or similar medium / short-range technologies.

[0132] Therefore, it should be understood that parts of the described solution can be implemented in a vehicle, in a system located outside the vehicle, or in a combination of inside and outside the vehicle; for example, in a server communicating with the vehicle, i.e., a so-called cloud solution. For example, data can be sent to an external system, and that system can perform steps for evaluating candidate paths. Different features and steps of the embodiments can be combined in combinations other than those described.

[0133] It should be noted that the word "comprising" does not exclude the presence of other elements or steps besides those listed, and the word "a" preceding an element does not exclude the presence of a plurality of such elements. It should be further noted that no reference numerals in the drawings limit the scope of the claims, the invention can be implemented at least in part by means of both hardware and software, and several "devices" or "units" can be represented by the same items in the hardware.

[0134] Although the accompanying drawings may show a specific order of method steps, the order of steps may differ from the depicted order. Furthermore, two or more steps may be performed simultaneously or partially simultaneously. For example, the steps of obtaining candidate paths and obtaining a risk graph may be interchanged based on a specific implementation. This variation will depend on the chosen software and hardware systems and the designer's choices. All these variations are within the scope of the invention. Similarly, software implementations can be accomplished using standard programming techniques with rule-based logic and other logic to perform various connection steps, processing steps, comparison steps, and decision steps. The embodiments mentioned and described above are given by way of example only and should not be construed as limiting the invention. Other solutions, uses, purposes, and functions within the scope of the invention claimed in the described patent embodiments will be apparent to those skilled in the art.

Claims

1. A method for automatically developing a path planning development module for a vehicle, executed by an onboard computing system, wherein, The vehicle is equipped with an automated driving system (ADS), and the method includes: Candidate paths are obtained from the path planning development module, wherein the path planning development module is configured to generate the candidate paths for the vehicle to execute based on a path planning model and data indicating the vehicle's surrounding environment; A risk map of the vehicle’s surrounding environment is obtained, wherein the risk map is formed based on the vehicle’s actuation capability and the location of a free space region in the surrounding environment, the actuation capability including an uncertainty estimate of the actuation capability, and the location of the free space region including an uncertainty estimate of the estimated location of the free space region. The risk map includes risk parameters for each of a plurality of area segments included in the surrounding environment of the vehicle. The risk map further has a time component, which indicates the time evolution of the risk parameter of the region segment based on the predicted time evolution of the free space region, at least within a duration defined by the predicted duration of the candidate path. The risk value of the candidate path is determined based on the risk parameters of the area segment to which the candidate path intersects; and If the determined risk value meets at least one risk criterion, the method further includes: At the output, a first control signal is generated that indicates the instructions for the ADS to execute the candidate path; Obtain at least one quality parameter of the executed candidate path, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort. A cost function indicative of the performance of the path planning development module is determined based on at least one obtained quality parameter; and One or more parameters of the path planning model are updated by an optimization algorithm configured to optimize the determined cost function.

2. The method according to claim 1, wherein, The at least one mass parameter includes at least one of lateral acceleration, longitudinal acceleration, vertical acceleration, rate of change of acceleration, path length, roll acceleration, pitch acceleration, yaw acceleration, and fuel consumption.

3. The method according to claim 1, wherein, The steps for determining the risk value include: The total risk value of the candidate path is determined based on the risk parameters of a set of area segments that intersect with the candidate path. The at least one risk criterion includes a total risk threshold, and the step of generating the first control signal at the output, indicating instructions for the ADS to execute the candidate path, is performed only when the total risk value is lower than the total risk threshold.

4. The method according to claim 3, wherein, The steps to determine the total risk value include: The risk parameters of the group of regional segments that intersect with the candidate path are summed.

5. The method according to claim 1, wherein, The step of determining the risk value further includes: The total risk value of the path planning development module is determined based on the risk parameters of a set of area segments that intersect with at least one previously executed path, wherein the set of area segments intersecting with at least one previously executed path is associated with a risk map previously obtained for the same period as the at least one previously executed path. The at least one risk criterion includes a total risk threshold, and the step of generating a first control signal at the output indicating instructions for the ADS to execute the candidate path is performed only when the total risk value is lower than the total risk threshold.

6. The method according to claim 5, wherein, The steps for determining the total risk value include: The risk parameters of the group of area segments intersecting with the at least one previously executed path are averaged to determine the average risk exposure of the path planning development module, the average risk exposure defining the total risk value.

7. The method according to claim 6, wherein, The step of averaging the risk parameters includes: For each time sample: The current risk value of the path planning development module is determined based on a set of regional segments that intersect with the most recently executed candidate path since previous time samples. The current composite risk value is determined by summing the current risk value with the previously calculated composite risk value of the previous time sample; and The average risk exposure is determined by dividing the current composite risk value by the total number of samples used up to and including the current time sample.

8. The method of claim 1, further comprising: An alternative route is obtained from the vehicle's production platform, which includes a validated route planning development module configured to generate the alternative route; and If the determined risk value fails to meet at least one of the risk criteria, the method further includes: A second control signal is generated at the output to execute the obtained alternative path; as well as Disable the path planning development module.

9. The method of claim 8, further comprising: If the determined risk value fails to meet at least one of the risk criteria: A second control signal is then generated at the output to enable the path planner of the production platform of the vehicle, for generating candidate paths to be executed by the vehicle.

10. The method of claim 1, further comprising: Transmit one or more updated parameters of the path planning model in the path planning development module to the remote entity; The remote entity receives a set of globally updated parameters of the path planning model of the path planning development module, wherein the set of globally updated parameters is based on information obtained from multiple vehicles including the path planning development module; and The path planning model of the path planning development module is updated based on the received set of globally updated parameters.

11. The method according to claim 1, wherein, Each risk parameter indicates at least one of the probability of an accident event occurring and the probability of violating a predefined safety threshold if the vehicle is to occupy the associated area segment.

12. The method according to claim 1, wherein, The location of the free space region includes: The location of an external object in the surrounding environment of the vehicle, wherein the estimated location of the external object includes the uncertainty of the location of the external object, the trajectory of any dynamic object in the external object, and the uncertainty of the estimated trajectory of the dynamic object.

13. A computer-readable storage medium storing one or more programs configured to be executed by one or more processors of an in-vehicle computing system, the one or more programs including instructions for performing the method according to any one of the preceding claims.

14. An apparatus for the automatic development of a path planning development module for vehicles, wherein, The vehicle is equipped with an automated driving system (ADS), the device including a control circuit configured to: Candidate paths are obtained from the path planning development module, wherein the path planning development module is configured to generate the candidate paths for the vehicle to execute based on a path planning model and data indicating the vehicle's surrounding environment; A risk map of the vehicle’s surrounding environment is obtained, wherein the risk map is formed based on the vehicle’s actuation capability and the location of a free space region in the surrounding environment, the actuation capability including an uncertainty estimate of the actuation capability, and the location of the free space region including an uncertainty estimate of the estimated location of the free space region. The risk map includes risk parameters for each of a plurality of area segments included in the surrounding environment of the vehicle. The risk map further has a time component, which indicates the time evolution of the risk parameter of the region segment based on the predicted time evolution of the free space region, at least within a duration defined by the predicted duration of the candidate path. The risk value of the candidate path is determined based on the risk parameters of the area segment to which the candidate path intersects; and If the determined risk value meets at least one risk criterion, the control circuit is further configured to: At the output, a first control signal is generated that indicates the instructions for the ADS to execute the candidate path; Obtain at least one quality parameter of the executed candidate path, wherein the at least one quality parameter indicates the performance of the candidate path in terms of quality and / or comfort. A cost function indicative of the performance of the path planning development module is determined based on at least one obtained quality parameter; and One or more parameters of the path planning model are updated by an optimization algorithm configured to optimize the determined cost function.

15. A vehicle comprising: A set of onboard sensors is configured to monitor the vehicle's surrounding environment; The autonomous driving system ADS includes: The perception system is configured to generate perception data based on sensor data obtained from one or more of the set of onboard sensors; and Production platform path planner; The path planning development module is configured to generate candidate paths for the vehicle to execute; and The apparatus according to claim 14.

Images