Interpretable trajectory prediction for autonomous and semi-autonomous systems and applications

Abstract

In various examples, a trajectory prediction model provides interpretable trajectory predictions for autonomous and semi-autonomous systems and applications using counterfactual game-theoretic reasoning. Model-based latent variables can be formulated through responsibility assessments. Responsibility can be broken down into several components, such as safety and courtesy. For example, responsibility can be quantified by answering a counterfactual question: Could an agent have acted differently to respect the safety of other agents and be more courteous to the intentions of others? The framework can be used to abstract computed responsibility sequences into different levels of responsibility and integrate latent levels into a trajectory prediction model capable of rendering interpretable and accurate inferences about the trajectory.

Description

BACKGROUND

[0001] Approaches for predicting the behavior of other agents (such as other vehicles in the vicinity of an ego vehicle) can be categorized as one of two types: phenomenological methods and ontological methods. Phenomenological methods infer the behavior of other agents through deep neural networks and a large amount of traffic data. However, the predicted trajectories generated by phenomenological methods are not explainable, making them difficult to use for ex post analysis or integration into downstream tasks, such as planning and control. Ontological methods construct explicit models of agents' internal decision-making processes or interaction structures.Ontological methods can provide interpretable explanations for their predictions, but they do not use traffic data efficiently, show poor scalability when the number of agents involved is large, cannot generalize across different traffic scenarios, and / or fail to encode a broad distribution of behaviors. SUMMARY OF THE INVENTION

[0002] The invention is defined by the claims. To illustrate the invention, aspects and embodiments that may or may not fall within the scope of the claims are described herein.

[0003] A trajectory prediction model is disclosed in various examples, providing interpretable trajectory predictions for autonomous and semi-autonomous systems and applications using counterfactual game theory reasoning. Model-based latent variables can be formulated using responsibility assessments. Responsibility can be divided into several components, such as safety and courtesy. Responsibility can be quantified, for example, by answering a counterfactual question: Could an agent have acted differently to respect the safety of other agents and be more courteous toward the intentions of others? The framework can be used to abstract computed responsibility sequences into different responsibility levels and integrate latent levels into a trajectory prediction model capable of rendering interpretable and accurate inferences about the trajectory.

[0004] Embodiments of the present disclosure relate to interpretable trajectory prediction for autonomous and semi-autonomous systems and applications via counterfactual game-theoretic reasoning. Systems and methods are disclosed that can provide a trajectory prediction model to generate a trajectory for an agent based on at least one or more movements. The trajectory prediction model can employ a responsibility formulation. Responsibility can be formulated as a measure of how much an agent considers the well-being of other agents through counterfactual reasoning. For example, responsibility can be formulated based on at least counterfactual metrics, such as a safety metric and a politeness metric.In contrast to conventional systems, the disclosed approach improves the accuracy and interpretability of agent behavior predictions, contributes to reducing excessive caution by autonomous and semi-autonomous systems in decision strategies, and leads to more efficient and high-performance behaviors by autonomous and semi-autonomous systems.

[0005] At least one aspect relates to a processor (e.g., one or more processors or processing units). The processor may be or may include one or more circuits. The one or more circuits may be configured to obtain sensor data corresponding to one or more movements of a query agent in proximity to an autonomous or semi-autonomous system. The one or more circuits may be configured to execute a trajectory prediction model to generate a trajectory for at least one of the query agent or the autonomous or semi-autonomous system. The trajectory may be generated for at least the one or more movements. The trajectory prediction model may include a responsibility formulation based on at least one or more counterfactual metrics.

[0006] In various embodiments, the responsibility formulation indicates a classification for at least one of the query agent or the autonomous or semi-autonomous system. In various embodiments, the one or more counterfactual metrics include one or more of a safety metric and / or a courtesy metric for the query agent. In various embodiments, the one or more counterfactual metrics include a safety metric for the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents besides the query agent.In various embodiments, the safety metric is indicative of one or more safety margins supported by the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents besides the query agent. In various embodiments, the one or more counterfactual metrics include a politeness metric for the query agent with respect to the autonomous or semi-autonomous system. In various embodiments, the politeness metric is indicative of an influence on the query agent of at least one of the autonomous or semi-autonomous system or one or more other agents besides the query agent. In various embodiments, the politeness metric is based at least on a Kullback-Leibler (KL) divergence between at least two distributions.In various embodiments, the at least two distributions correspond to movement of the query agent (1) with the autonomous or semi-autonomous system or with one or more other agents besides the query agent, and (2) without the autonomous or semi-autonomous system or without the one or more other agents besides the query agent. In various embodiments, the one or more counterfactual metrics include a safety metric and / or a courtesy metric for the query agent. In various embodiments, the responsibility formulation uses a reward function. In various embodiments, the one or more circuits are configured to execute a maneuver by the autonomous or semi-autonomous system based on at least the generated trajectory.

[0007] At least one further aspect relates to a system. The system may include or consist of one or more processing units to perform operations including: obtaining sensor data corresponding to one or more movements of a first agent in proximity to a second agent; and executing a trajectory prediction model to generate a trajectory for at least one of the first agent or the second agent based on at least the one or more movements. The trajectory prediction model may include a responsibility formulation based on at least one or more counterfactual metrics.

[0008] In various embodiments, the one or more counterfactual metrics include a safety metric for the first agent with respect to at least one of the second agent and / or one or more agents other than the first agent. In various embodiments, the safety metric is indicative of one or more safety margins supported by the first agent with respect to at least one of the second agent and / or one or more agents other than the first agent. In various embodiments, the one or more counterfactual metrics include a politeness metric for the first agent with respect to the second agent. In various embodiments, the politeness metric is indicative of an influence on the first agent of at least one of the second agent and / or one or more agents other than the first agent.In various embodiments, the one or more counterfactual metrics include a safety metric and a courtesy metric for the first agent. In various embodiments, the system is an autonomous or semi-autonomous vehicle. In various embodiments, the vehicle includes one or more sensors from which the sensor data, or a portion thereof, is received. In various embodiments, the vehicle includes a control system configured to execute a second trajectory for the vehicle based on the generated trajectory.

[0009] At least one further aspect relates to a method. The method may consist of or include: obtaining sensor data corresponding to one or more movements of a query agent in the vicinity of a dynamic system; and executing a trajectory prediction model to generate a trajectory for at least one of the query agent and / or the dynamic system based on at least the one or more movements, wherein the trajectory prediction model includes a responsibility formulation based on at least one safety metric and one courtesy metric.

[0010] In various embodiments, the processors, systems, and / or methods described herein may be implemented by or via or included in at least one of the following: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulation; a system for performing collaborative content creation for 3D assets; a system for performing deep learning operations; a system implemented using an edge device; a system implemented using a robot; a system for performing conversational AI; a system implementing one or more language models (LLMs); a system for performing one or more generative AI operations;a system for generating synthetic data; a system including one or more virtual machines (VMs); a system implemented at least in part in a data center; and / or a system implemented at least in part using cloud computing resources.

[0011] The disclosure extends to all novel aspects or features described and / or illustrated herein.

[0012] Further features of the disclosure are characterized by the independent and dependent claims.

[0013] Any feature of one aspect of the disclosure may be applied to other aspects of the disclosure in any suitable combination. In particular, method aspects may be applied to device or system aspects, and vice versa.

[0014] Furthermore, features implemented in hardware may be implemented in software, and vice versa. Any reference to software and hardware features herein should be construed accordingly.

[0015] Any system or device feature described herein may also be provided as a method feature, and vice versa. System and / or device aspects described functionally (including means plus functional features) may alternatively be expressed in terms of their corresponding structure, such as a suitably programmed processor and associated memory.

[0016] It is also to be understood that certain combinations of the various features described and defined in each aspect of the disclosure may be implemented and / or provided and / or used independently of one another.

[0017] The disclosure also provides computer programs and computer program products comprising software code configured to, when executed on a data processing device, perform any of the methods described herein and / or embody any of the device and system features described herein, including any or all component steps of a method.

[0018] The disclosure also includes a computer or computer system (including networked or distributed systems) having an operating system that supports a computer program for performing the methods described herein and / or for embodying the apparatus or system features described herein.

[0019] The disclosure also provides a computer-readable medium on which one or more of the aforementioned computer programs are stored.

[0020] The disclosure also provides a signal carrying one or more of the aforementioned computer programs.

[0021] The disclosure extends to methods and / or devices and / or systems as described herein with reference to the accompanying drawings.

[0022] Aspects and embodiments of the disclosure will now be described, by way of example only, with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present systems and methods for interpretable trajectory prediction for autonomous and semi-autonomous systems and applications are described in detail below with reference to the attached drawings. They show: Fig. 1A is an illustration of an exemplary autonomous vehicle, according to some embodiments of the present disclosure; Fig. 1B shows an example of camera locations and fields of view for the exemplary autonomous vehicle from Fig. 1A, according to some embodiments of the present disclosure; Fig. 1C is a block diagram of an exemplary system architecture for the exemplary autonomous vehicle of Fig. 1A, according to some embodiments of the present disclosure; Fig. 1D shows a system diagram for communication between one or more cloud-based servers and the exemplary autonomous vehicle of Fig. 1A, according to some embodiments of the present disclosure; Fig. 2 is a block diagram of an exemplary computing device suitable for use in implementing some embodiments of the present disclosure; and Fig. 3 is a block diagram of an exemplary data center suitable for use in implementing some embodiments of the present disclosure. Fig. 4 is an example flowchart illustrating a process by which a system may use a model to predict a trajectory, according to some embodiments of the present disclosure; Fig. 5 illustrates an exemplary interpretability framework for a wide range of phenomenological trajectory prediction methods through responsibility assessments, according to some embodiments of the present disclosure; Fig. 6 illustrates an exemplary distribution of security and civility responsibility of a training set (left) and a validation set (right) according to some embodiments of the present disclosure; Fig. 7 shows, on the left, an exemplary Bayesian information criterion under a different number of levels and, on the right, observation distributions of a Hidden Markov Model (HMM) with five responsibility levels, according to some embodiments of the present disclosure; Fig. 8, in the top row, that a polled agent exhibits irresponsible behavior when attempting to overtake a target agent, while the bottom row shows that the polled agent becomes responsible after overtaking, according to some embodiments of the present disclosure; Fig. 9, that an exemplary responsibility-dependent reward function according to some embodiments of the present disclosure yields a diverse and interpretable trajectory prediction; and Fig. 10 shows a comparison of ground truth trajectories and predicted trajectories illustrating that responsible prediction outperforms a heuristic reward-based model for Bilevel Imitation for Traffic Simulation (BITS) according to some embodiments of the present disclosure. DETAILED DESCRIPTION

[0024] Systems and methods related to interpretable trajectory prediction for autonomous and semi-autonomous systems and applications are disclosed. Although the present disclosure is made with respect to an exemplary autonomous vehicle 100 (alternatively referred to herein as "vehicle 100" or "ego machine 100," an example of which is disclosed with respect to Fig. 1A through 1D), this is not intended to be limiting. The systems and methods described herein may be used, for example, without limitation, by non-autonomous vehicles or machines, semi-autonomous vehicles or machines (e.g., in one or more adaptive driver assistance systems (ADAS)), autonomous vehicles or machines, guided and unguided robots or robotic platforms, warehouse vehicles, off-highway vehicles, vehicles coupled to one or more trailers, hydrofoils, boats, shuttles, emergency vehicles, motorcycles, electric or motorized bicycles, aircraft, construction vehicles, underwater vehicles, drones, and / or other types of vehicles.Furthermore, although the present disclosure may be described with respect to trajectory prediction for autonomous or semi-autonomous driving, this is not to be construed as limiting, and the systems and methods described herein may also be used in augmented reality, virtual reality, mixed reality, robotics, security and surveillance, autonomous or semi-autonomous machine applications, and / or any other technology areas in which trajectory prediction may be used.

[0025] Various autonomous and semi-autonomous systems, such as autonomous vehicles (AVs), evaluate the movements of agents (e.g., vehicles) in their environment to make predictions about the agents' trajectories, for example, to avoid collisions. A major challenge for AVs is future planning in interactive driving scenarios, as AVs must be responsible for the safety of road users (e.g., other drivers, pedestrians, and cyclists) while also complying with traffic laws and balancing the trade-off between caution and task performance (e.g., passenger comfort and mobility). In such interactive settings, an autonomous or semi-autonomous system can reason about the uncertainty and interaction dynamics of other agents to select its control strategies and tactics (e.g., changes in direction and / or speed).However, because safety is a priority, uncertainty and a lack of transparency in the behavior of other actors often lead to overly cautious system behavior, which impairs performance. Improving the accuracy and interpretability of models used to predict agent behavior can help achieve a level of caution that improves both safety and performance.

[0026] To combine the advantages of both ontological and phenomenological methods, the disclosed approach provides an interpretive approach for a wide range of phenomenological approaches using model-based latent variables. These latent variables can be formulated through responsibility assessments, since responsibility is inherent in human behavior. This is consistent with legal frameworks that aim to manage the risk of road accidents partly by enforcing responsibility or a duty of care for each road user. Responsibility can be divided into several components, such as the following two components: (1) How important does an agent consider its margin of error until colliding with other agents? and (2) How much does an agent influence the behavior of other agents?

[0027] Instead of relying on a specific model of agent decision-making, embodiments of the disclosed approach can use a phenomenological traffic model to generate counterfactual control strategies that an agent could have adopted and answer the above questions to quantify the agent's responsibility. The disclosed approach can thus provide an interpretable framework for a wide range of phenomenological methods for predicting trajectories based at least in part on responsibility assessments. This interpretable framework can be used to abstract computed responsibility sequences into different responsibility levels and integrate these latent levels into a trajectory prediction model capable of rendering interpretable and accurate inferences.Embodiments of the disclosed responsibility-based interpretability framework can be modular and easily integrated into a wide range of prediction models.

[0028] As further explained herein, embodiments of the disclosed approach can formulate responsibility by answering a counterfactual question: Could an agent have acted differently to respect the safety of other agents and be more courteous toward the intentions of others? A hierarchical framework can be used to categorize the intention behind interactions between agents into different levels of responsibility and link these levels to models for predicting trajectories. Simulation studies demonstrate that the disclosed responsibility metrics can be used to group human behavior into different groups and provide more interpretable and accurate trajectory predictions.

[0029] With reference to Fig. 1 is Fig. 1 illustrates an exemplary autonomous or semi-autonomous vehicle capable of navigating based on trajectory predictions generated using trajectory prediction models, according to some embodiments of the present disclosure. It should be understood that these and other arrangements described herein are presented only as examples. Other arrangements and elements (e.g., machines, interfaces, functions, arrangements, groupings of functions, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Furthermore, many of the elements described herein are functional entities that may be implemented as individual or distributed components, or in conjunction with other components, and in any suitable combination and location.Various functions described herein that are performed by entities may be performed by hardware, firmware, and / or software. For example, various functions may be performed by a processor executing instructions stored in memory. In some embodiments, the systems, methods, and processes described herein may be implemented using similar components, features, and / or functionality as the example autonomous vehicle 100 of FIG. Fig. 1A-1D, the exemplary computer device 200 of Fig. 2 and / or the exemplary data center 300 from Fig. 3 are executed.

[0030] The systems and methods described herein may be used, without limitation, by non-autonomous vehicles, semi-autonomous vehicles (e.g., in one or more adaptive driver assistance systems (ADAS)), guided and unguided robots or robotic platforms, warehouse vehicles, off-road vehicles, vehicles coupled to one or more trailers, hydrofoils, boats, shuttles, emergency vehicles, motorcycles, electric or motorized bicycles, aircraft, construction vehicles, underwater vehicles, drones, and / or other types of vehicles.Furthermore, the systems and methods described herein may be used for a variety of purposes, including, without limitation, machine control, machine locomotion, machine propulsion, synthetic data generation, model training, perception, augmented reality, virtual reality, mixed reality, robotics, security and surveillance, simulation and digital twinning, autonomous or semi-autonomous machine applications, deep learning, environmental simulation, object or actor simulation and / or digital twinning, data center processing, conversational AI, light transport simulation (e.g., ray tracing, path tracing, etc.), collaborative content creation for 3D assets, cloud computing, and / or other suitable applications.

[0031] The embodiments disclosed herein may include a variety of different systems, such as automotive systems (e.g., a control system for an autonomous or semi-autonomous machine, a perception system for an autonomous or semi-autonomous machine), systems implemented with a robot, aviation systems, media systems, boat systems, intelligent area monitoring systems, systems for performing deep learning operations, systems for performing simulation operations, systems for performing digital twin operations, systems implemented using an edge device, systems including one or more virtual machines (VMs), systems for performing synthetic data generation operations, systems implemented at least partially in a data center, systems for performing conversational AI operations, systems for hosting real-time streaming applications,Systems for presenting one or more virtual reality content, augmented reality content, or mixed reality content, systems for performing light transport simulations, systems for performing collaborative content creation for 3D assets, systems implemented at least partially using cloud computing resources, and / or other types of systems. EXEMPLARY AUTONOMOUS VEHICLE

[0032] Fig. 1A is an illustration of an exemplary autonomous vehicle 100, according to some embodiments of the present disclosure. The autonomous vehicle 100 (alternatively referred to herein as "vehicle 100") may include, without limitation, a passenger vehicle, such as a car, a truck, a bus, an emergency service vehicle, a shuttle, an electric or motorized bicycle, a motorcycle, a fire engine, a police vehicle, an ambulance, a boat, a construction vehicle, an underwater vehicle, a robotic vehicle, a drone, an aircraft, a vehicle coupled to a trailer (e.g., a semi-truck used to transport cargo), and / or another type of vehicle (e.g., that is unmanned and / or accommodates one or more passengers).Autonomous vehicles are generally described in terms of levels of automation defined by the National Highway Traffic Safety Administration (NHTSA), a division of the U.S. Department of Transportation, and the Society of Automotive Engineers (SAE) standard "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles" (Standard No. J3016-201806, published June 15, 2018, Standard No. J3016-201609, published September 30, 2016, and prior and future versions of this standard). Vehicle 100 may exhibit functionality consistent with one or more of the Level 3 through Level 5 autonomous driving levels.The vehicle 100 may have functionality according to one or more of Level 1 through Level 5 autonomous driving levels. For example, depending on the embodiment, the vehicle 100 may be capable of driver assistance (Level 1), partial automation (Level 2), conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5). The term "autonomous" as used herein may include any and / or all types of autonomy for the vehicle 100 or other machine, such as fully autonomous, highly autonomous, conditionally autonomous, partially autonomous, assisted autonomy, semi-autonomous, primarily autonomous, or another designation.

[0033] The vehicle 100 may include components such as a chassis, a vehicle body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other components of a vehicle. The vehicle 100 may include a propulsion system 150, such as an internal combustion engine, a hybrid electric power plant, a pure electric motor, and / or another type of propulsion system. The propulsion system 150 may be connected to a drivetrain of the vehicle 100, which may include a transmission to enable propulsion of the vehicle 100. The propulsion system 150 may be controlled in response to receiving signals from the throttle or accelerator 152.

[0034] A steering system 154, which may include a steering wheel, may be used to steer the vehicle 100 (e.g., along a desired path or route) when the propulsion system 150 is operating (e.g., when the vehicle is moving). The steering system 154 may receive signals from a steering actuator 156. The steering wheel may be optional for full automation (Level 5).

[0035] The brake sensor system 146 may be used to apply the vehicle brakes in response to receiving signals from the brake actuators 148 and / or the brake sensors.

[0036] The one or more controllers 136 that control one or more systems on chips (SoCs) 104 ( Fig. 1C) and / or GPUs may provide signals (e.g., representative of instructions) to one or more components and / or systems of the vehicle 100. For example, the one or more controllers may send signals to apply the vehicle brakes via one or more brake actuators 148, to apply the steering system 154 via one or more steering actuators 156, to apply the propulsion system 150 via one or more throttle / accelerator devices 152. The one or more controllers 136 may include one or more built-in (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals and issue operational commands (e.g., signals representing commands) to enable autonomous driving and / or to assist a human driver in driving the vehicle 100.The one or more controllers 136 may include a first controller 136 for autonomous driving functions, a second controller 136 for functional safety functions, a third controller 136 for artificial intelligence functions (e.g., computer vision), a fourth controller 136 for infotainment functions, a fifth controller 136 for emergency redundancy, and / or other controllers. In some examples, a single controller 136 may perform two or more of the above functionalities, two or more controllers 136 may perform a single functionality, and / or any combination thereof.

[0037] The one or more controllers 136 may provide the signals to control one or more components and / or systems of the vehicle 100 in response to sensor data received from one or more sensors (e.g., sensor inputs). The sensor data may be received, for example and without limitation, from one or more of the following: global navigation satellite systems (“GNSS”) sensor(s) 158 (e.g., global positioning system sensor(s)), RADAR sensor(s) 160, ultrasonic sensor(s) 162, LIDAR sensor(s) 164, inertial measurement unit (IMU) sensor(s) 166 (e.g., accelerometer(s), gyroscope(s), magnetic compass(es), magnetometer(s), etc.), microphone(s) 196, stereo camera(s) 168, wide-angle camera(s) 170 (e.g., fisheye cameras), infrared camera(s) 172, ambient camera(s) 174 (e.g.,360-degree cameras), long-range and / or medium-range camera(s) 198, speed sensor(s) 144 (e.g., for measuring the speed of the vehicle 100), vibration sensor(s) 142, steering sensor(s) 140, brake sensor(s) (e.g., as part of the brake sensor system 146), and / or other sensor types.

[0038] One or more of the controllers 136 may receive inputs (e.g., in the form of input data) from an instrument cluster 132 of the vehicle 100 and provide outputs (e.g., in the form of output data, display data, etc.) via a human-machine interface (HMI) display 134, an audible annunciator, a speaker, and / or via other components of the vehicle 100. The outputs may include information such as vehicle speed, RPM, time, map data (e.g., the high-definition (“HD”) map 122 of Fig. 1C), location data (e.g., the location of the vehicle 100, e.g., on a map), direction, location of other vehicles (e.g., an occupancy grid), information about objects and status of objects as perceived by the one or more controllers 136, etc. For example, the HMI display 134 may display information about the presence of one or more objects (e.g., a road sign, a warning sign, a changing traffic light, etc.) and / or information about maneuvers that the vehicle has performed, is currently performing, or will perform (e.g., change lanes now, take exit 34B in two miles, etc.).

[0039] The vehicle 100 further includes a network interface 124 that may utilize one or more wireless antennas 126 and / or modems for communication over one or more networks. The network interface 124 may, for example, be capable of communication over Long-Term Evolution (LTE), Wideband Code Division Multiple Access (WCDMA), Universal Mobile Telecommunications System (UMTS), Global System for Mobile Communications (GSM), IMT-CDMA Multi-Carrier (CDMA2000), etc. The one or more wireless antennas 126 may also enable communication between objects in the environment (e.g., vehicles, mobile devices, etc.) using local area networks such as Bluetooth, Bluetooth Low Energy (LE), Z-Wave, ZigBee, etc.and / or Low Power Wide Area Networks (LPWANs), such as LoRaWAN, SigFox, etc.

[0040] Fig. 1B is an example of camera locations and fields of view for the example autonomous vehicle 100 of Fig. 1A, according to some embodiments of the present disclosure. The cameras and respective fields of view are an example embodiment and are not intended to be limiting. For example, additional and / or alternative cameras may be included and / or the cameras may be located at various locations on the vehicle 100.

[0041] The camera types for the cameras may include, but are not limited to, digital cameras that may be configured for use with the components and / or systems of the vehicle 100. The one or more cameras may operate at Automotive Safety Integrity Level (ASIL) B and / or another ASIL. The camera types may be capable of any image capture rate, such as 60 frames per second (fps), 120 fps, 240 fps, etc., depending on the embodiment. The cameras may use rolling shutters, global shutters, another type of shutter, or a combination thereof.In some examples, the color filter array may include a red clear clear clear (RCCC) color filter array, a red clear clear blue (RCCB) color filter array, a red blue green clear (RBGC) color filter array, a Foveon X3 color filter array, a Bayer sensor color filter array (RGGB), a monochrome sensor color filter array, and / or another type of color filter array. In some embodiments, clear-pixel cameras, such as cameras with an RCCC, an RCCB, and / or an RBGC color filter array, may be used to increase light sensitivity.

[0042] In some examples, one or more of the cameras can be used to perform advanced driver assistance system (ADAS) functions (e.g., as part of a redundant or fail-safe design). For example, a multifunction mono camera can be installed to provide features including lane departure warning, traffic sign assist, and intelligent headlight control. One or more of the cameras (e.g., all cameras) can simultaneously record and provide image data (e.g., video).

[0043] One or more of the cameras may be mounted in a bracket, such as a specially designed (three-dimensional ("3D") printed) bracket, to eliminate stray light and reflections from inside the vehicle (e.g., dashboard reflections reflected in the windshield) that could interfere with the camera's image data acquisition. With respect to the bracket for exterior mirrors, the exterior mirrors may be custom 3D printed so that the camera mounting plate conforms to the shape of the exterior mirror. In some examples, the one or more cameras may be integrated into the exterior mirror. For side-mounted cameras, the one or more cameras may also be integrated into the four pillars at each corner of the cabin.

[0044] Cameras with a field of view that includes portions of the environment in front of the vehicle 100 (e.g., forward-facing cameras) can be used for the surrounding view to help identify forward paths and obstacles, and to provide information critical to establishing an occupancy grid and / or determining preferred vehicle paths with the assistance of one or more controllers 136 and / or control SoCs. Forward-facing cameras can be used to perform many of the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance. Forward-facing cameras can also be used for ADAS features and systems that include lane departure warnings (LDW), autonomous cruise control (ACC), and / or other features such as traffic sign detection.

[0045] A variety of cameras may be used in a forward-facing configuration, including, for example, a monocular camera platform containing a complementary metal oxide semiconductor (CMOS) color imager. Another example is the wide-angle cameras 170, which may be used to capture objects entering the field of view from the periphery (e.g., pedestrians, crossing vehicles, or bicycles). Although Fig. 1B illustrates only one wide-angle camera, the vehicle 100 may include any number (including zero) of wide-angle cameras 170. Furthermore, any number of long-range cameras 198 (e.g., a long-range stereo camera pair) may be used for depth-based object detection, particularly for objects for which a neural network has not yet been trained. The one or more long-range cameras 198 may also be used for object detection and classification, as well as basic object tracking.

[0046] Any number of stereo cameras 168 may also be included in a forward-facing configuration. In at least one embodiment, one or more of the stereo cameras 168 may include an integrated control unit comprising a scalable processing unit that can provide a programmable logic ("FPGA") and a multi-core microprocessor with an integrated controller area network ("CAN") or Ethernet interface on a single chip. Such a unit can be used to create a 3D map of the vehicle's surroundings that includes a distance estimate for all points in the image. One or more alternative stereo cameras 168 may include a compact stereo vision sensor that can include two camera lenses (one each on the left and right) and an image processing chip that can measure the distance between the vehicle and the target object and process the generated information (e.g.,metadata) to activate the autonomous emergency braking and lane departure warning functions. Other types of stereo cameras 168 may be used in addition to or alternatively to those described here.

[0047] Cameras with a field of view that includes portions of the environment to the side of the vehicle 100 (e.g., side cameras) may be used for the environment view and provide information used to create and update the occupancy grid and to generate collision warnings in the event of a side impact. For example, the one or more environment cameras 174 (e.g., four environment cameras 174, as in Fig. 1B) may be positioned on the vehicle 100. The one or more surround cameras 174 may include one or more wide-angle cameras 170, one or more fisheye cameras, one or more 360-degree cameras, and / or the like. For example, four fisheye cameras may be mounted on the front, rear, and sides of the vehicle. In an alternative arrangement, the vehicle may utilize three surround cameras 174 (e.g., left, right, and rear) and utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround camera.

[0048] Cameras with a field of view that includes portions of the environment behind the vehicle 100 (e.g., rearview cameras) may be used for parking assistance, surround view, rear impact warnings, and occupancy grid creation and updating. A variety of cameras may be used, including, but not limited to, cameras that are also suitable as one or more forward-facing cameras (e.g., one or more long-range and / or medium-range cameras 198, one or more stereo cameras 168, one or more infrared cameras 172, etc.), as described herein.

[0049] Fig. 1C is a block diagram of an example system architecture for the example autonomous vehicle 100 of Fig. 1A, according to some embodiments of the present disclosure. It should be understood that these and other arrangements described herein are set forth as examples only. Other arrangements and elements (e.g., engines, interfaces, functions, arrangements, groupings of functions, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Furthermore, many of the elements described herein are functional entities that may be implemented as individual or distributed components, or in conjunction with other components, and in any suitable combination and location. Various functions performed by entities described herein may be performed by hardware, firmware, and / or software. Various functions may be performed, for example, by a processor executing instructions stored in memory.

[0050] Each of the components, features and systems of the vehicle 100 in Fig. 1C is illustrated as being connected via bus 102. Bus 102 may include a controller area network (CAN) data interface (alternatively referred to herein as a "CAN bus"). A CAN may be a network within vehicle 100 that serves to support the control of various features and functions of vehicle 100, such as the application of brakes, acceleration, braking, steering, windshield wipers, etc. A CAN bus may be configured to have dozens or even hundreds of nodes, each with its own unique identifier (e.g., a CAN ID). The CAN bus may be read to determine steering wheel angle, vehicle speed, engine speed (RPM), button positions, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0051] Although bus 102 is described herein as a CAN bus, this is not intended to be a limitation. For example, FlexRay and / or Ethernet may be used in addition to or alternatively to the CAN bus. Furthermore, while a single wire is used to represent bus 102, this is not intended to be a limitation. For example, there may be any number of buses 102, which may include one or more CAN buses, one or more FlexRay buses, one or more Ethernet buses, and / or one or more other types of buses that use a different protocol. In some examples, two or more buses 102 may be used to perform different functions and / or may be used for redundancy. For example, a first bus 102 may be used for collision avoidance functionality and a second bus 102 may be used for actuation control.In each example, each bus 102 may communicate with one of the components of the vehicle 100, and two or more buses 102 may communicate with the same components. In some examples, each SoC 104, each controller 136, and / or each computer within the vehicle may have access to the same input data (e.g., inputs from sensors of the vehicle 100) and may be connected to a common bus, such as the CAN bus.

[0052] The vehicle 100 may include one or more controllers 136 as described herein with respect to Fig. 1A. The one or more controllers 136 may be used for a variety of functions. The one or more controllers 136 may be coupled to any of the various other components and systems of the vehicle 100 and may be used for control of the vehicle 100, artificial intelligence of the vehicle 100, infotainment for the vehicle 100, and / or the like.

[0053] The vehicle 100 may include one or more systems on a chip (SoC) 104. The SoC 104 may include one or more CPUs 106, one or more GPUs 108, one or more processors 110, one or more caches 112, one or more accelerators 114, one or more data memories 116, and / or other components and features not illustrated. The one or more SoCs 104 may be used to control the vehicle 100 in a variety of platforms and systems. For example, the one or more SoCs 104 may be combined in a system (e.g., the system of the vehicle 100) with an HD card 122 that may be accessed via a network interface 124 from one or more servers (e.g., the one or more servers 178 of Fig. 1D) can receive map refreshes and / or updates.

[0054] The one or more CPUs 106 may include a CPU cluster or CPU complex (alternatively referred to herein as a "CCPLEX"). The one or more CPUs 106 may include multiple cores and / or L2 caches. For example, in some embodiments, the one or more CPUs 106 may include eight cores in a coherent multiprocessor configuration. In some embodiments, the one or more CPUs 106 may include four dual-core clusters, each cluster having a dedicated L2 cache (e.g., a 2 MB L2 cache). The one or more CPUs 106 (e.g., the CCPLEX) may be configured to support concurrent operation of clusters, such that any combination of the clusters of the one or more CPUs 106 may be active at any given time.

[0055] The one or more CPUs 106 may implement power management features that include one or more of the following: individual hardware blocks may be automatically clocked when idle to conserve dynamic power; each core clock may be controlled when the core is not actively executing instructions due to the execution of WFI / WFE instructions; each core may be independently power controlled; each core cluster may be independently clock controlled if all cores are clock controlled or power controlled; and / or each core cluster may be independently power controlled if all cores are power controlled. The one or more CPUs 106 may further implement an enhanced power state management algorithm in which allowable power states and expected wake-up times are established, and the hardware / microcode determines the best power state to enter for the core, cluster, and CCPLEX.The processing cores can support simplified sequences for inputting the energy state into the software, offloading the work to the microcode.

[0056] The one or more GPUs 108 may include an integrated GPU (alternatively referred to herein as an "iGPU"). The one or more GPUs 108 may be programmable and may be efficient for parallel workloads. The one or more GPUs 108 may, in some examples, use an extended Tensor instruction set. The one or more GPUs 108 may include one or more streaming microprocessors, where each streaming microprocessor may include an L1 cache (e.g., an L1 cache with at least 96 KB of memory capacity), and two or more of the streaming microprocessors may share an L2 cache (e.g., an L2 cache with 512 KB of memory capacity). In some embodiments, the one or more GPUs 108 may include at least eight streaming microprocessors. The one or more GPUs 108 may use one or more application programming interfaces (APIs) for computations.Additionally, the one or more GPUs 108 may utilize one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0057] The one or more GPUs 108 may be power-optimized for best performance in automotive and embedded use cases. For example, the one or more GPUs 108 may be fabricated on a fin field-effect transistor (FinFET). However, this is not a limitation, and the one or more GPUs 108 may also be fabricated using other semiconductor manufacturing processes. Each streaming microprocessor may include a number of mixed-precision processing cores divided into multiple blocks. For example, and without limitation, 64 PF32 cores and 32 PF64 cores may be divided into four processing blocks. In such an example, each processing block can be assigned 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two NVIDIA TENSOR COREs with mixed precision for deep learning matrix arithmetic, an L0 instruction cache, a warp scheduler, a dispatch unit, and / or a 64 KB register file.In addition, the streaming microprocessors can include independent parallel integer and floating-point datapaths to enable efficient execution of workloads with a mix of computations and addressing calculations. The streaming microprocessors can include independent thread scheduling to enable fine-grained synchronization and cooperation between parallel threads. The streaming microprocessors can include a combined L1 data cache and a shared memory unit to improve performance while simplifying programming.

[0058] The one or more GPUs 108 may include high-bandwidth memory (HBM) and / or a 16 GB HBM2 memory subsystem to provide, in some examples, a peak memory bandwidth of approximately 900 GB / second. In some examples, synchronous graphics random-access memory (SGRAM), such as double data rate synchronous graphics random-access memory (GDDR5), may be used in addition to or as an alternative to the HBM memory.

[0059] The one or more GPUs 108 may include unified memory technology that includes access counters to enable more accurate migration of memory pages to the processor that accesses them most frequently, thereby improving efficiency for processor-shared memory areas. In some examples, address translation services (ATS) support may be used to allow the one or more GPUs 108 to directly access the page tables of the one or more CPUs 106. In such examples, if the memory management unit (MMU) of the one or more GPUs 108 fails, an address translation request may be sent to the one or more CPUs 106.In response, the one or more CPUs 106 may look up the virtual-to-physical mapping for the address in their page tables and send the translation back to the one or more GPUs 108. Thus, the unified memory technology may enable a single unified virtual address space for the memory of both the one or more CPUs 106 and the one or more GPUs 108, thereby simplifying the programming of the one or more GPUs 108 and the porting of applications to the one or more GPUs 108.

[0060] Additionally, the one or more GPUs 108 may include an access counter that can track the frequency of access by the one or more GPUs 108 to memory of other processors. The access counter can help ensure that memory pages are moved to the physical memory of the processor that accesses the pages most frequently.

[0061] The one or more SoCs 104 may include any number of caches 112, including those described herein. For example, the one or more caches 112 may include an L3 cache available to both the one or more CPUs 106 and the one or more GPUs 108 (e.g., connected to both the one or more CPUs 106 and the one or more GPUs 108). The one or more caches 112 may include a write-back cache that may track line states, e.g., by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). The L3 cache may include 4 MB or more, depending on the embodiment, although smaller cache sizes may be used.

[0062] The one or more SoCs 104 may include one or more arithmetic logic units (ALU(s)) that may be utilized in performing processing related to any of the various tasks or operations of the vehicle 100, such as processing DNNs. Additionally, the one or more SoCs 104 may include one or more floating-point units (FPU(s))—or other mathematical co-processors or numerical co-processors—for performing mathematical operations within the system. For example, the one or more SoCs 104 may include one or more FPUs integrated as execution units within one or more CPUs 106 and / or one or more GPUs 108.

[0063] The one or more SoCs 104 may include one or more accelerators 114 (e.g., hardware accelerators, software accelerators, or a combination thereof). For example, the one or more SoCs 104 may include a hardware acceleration cluster, which may include optimized hardware accelerators and / or large on-chip memory. The large on-chip memory (e.g., 4MB SRAM) may enable the hardware acceleration cluster to accelerate neural networks and other computations. The hardware acceleration cluster may be used to complement the one or more GPUs 108 and offload some of the tasks of the one or more GPUs 108 (e.g., to free up more cycles of the one or more GPUs 108 to perform other tasks). The one or more accelerators 114 may, for example, be used for targeted workloads (e.g.,Perception, convolutional neural networks (CNNs), etc.) that are robust enough to be suitable for acceleration can be used. The term "CNN" as used here can include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and fast RCNNs (e.g., for object detection).

[0064] The one or more accelerators 114 (e.g., the hardware acceleration cluster) may include a deep learning accelerator (DLA). The one or more DLAs may include one or more tensor processing units (TPUs) configured to provide an additional tens of trillion operations per second for deep learning applications and inferencing. The TPUs may be accelerators configured and optimized to perform image processing functions (e.g., for CNNs, RCNNs, etc.). The one or more DLAs may also be optimized for a specific set of neural network types and floating-point operations, as well as for inferencing. The design of the one or more DLAs may provide more performance per millimeter than a general-purpose GPU, far exceeding the performance of a CPU.The one or more TPUs can perform multiple functions, including a single-instance convolution function that supports, for example, INT8, INT16, and FP16 data types for both features and weights, as well as post-processing functions.

[0065] The one or more DLAs can quickly and efficiently execute neural networks, particularly CNNs, on processed or unprocessed data for a variety of functions, including, for example and without limitation: a CNN for object identification and detection using data from camera sensors; a CNN for distance estimation using data from camera sensors; a CNN for detection and identification of emergency vehicles using data from microphones; a CNN for facial recognition and identification of vehicle owners using data from camera sensors; and / or a CNN for safety and / or security events.

[0066] The one or more DLAs can perform any function of the one or more GPUs 108, and by using an inference accelerator, for example, a developer can dedicate either the one or more DLAs or the one or more GPUs 108 to each function. For example, the developer can focus the processing of CNNs and floating-point operations on the one or more DLAs and leave other functions to the one or more GPUs 108 and / or other accelerators 114.

[0067] The one or more accelerators 114 (e.g., the hardware acceleration cluster) may include a programmable vision accelerator (PVA), which may also be referred to herein as a computer vision accelerator. The one or more PVAs may be designed and configured to accelerate computer vision algorithms for advanced driver assistance systems (ADAS), autonomous driving, and / or augmented reality (AR) and / or virtual reality (VR) applications. The one or more PVAs may provide a balance between performance and flexibility. For example, and without limitation, each PVA may include any number of reduced instruction set computer (RISC) cores, direct memory access (DMA) cores, and / or any number of vector processors.

[0068] The RISC cores may interact with image sensors (e.g., the image sensors of one of the cameras described herein), image signal processors, and / or the like. Each of the RISC cores may include any amount of memory. The RISC cores may use any number of protocols, depending on the embodiment. In some examples, the RISC cores may execute a real-time operating system (RTOS). The RISC cores may be implemented with one or more integrated circuits, application-specific integrated circuits (ASICs), and / or memory devices. The RISC cores may include, for example, an instruction cache and / or tightly coupled RAM.

[0069] The DMA may enable components of the PVA(s) to access the system's memory independently of the one or more CPUs 106. The DMA may support any number of features designed to optimize the PVA, including, but not limited to, support for multi-dimensional addressing and / or circular addressing. In some examples, the DMA may support up to six or more dimensions of addressing, which may include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0070] The vector processors may be programmable processors that can be designed to efficiently and flexibly execute programming for computer vision algorithms and provide signal processing functions. In some examples, the PVA may include a PVA core and two vector processing subsystem partitions. The PVA core may include a processor subsystem, one or more DMA engines (e.g., two DMA engines), and / or other peripheral devices. The vector processing subsystem may operate as the primary processing engine of the PVA and may include a vector processing unit (VPU), an instruction cache, and / or memory (e.g., VMEM).A VPU core can contain a digital signal processor, such as a single instruction multiple data (SIMD) and very long instruction word (VLIW) digital signal processor. Combining SIMD and VLIW can increase throughput and speed.

[0071] Each of the vector processors may include an instruction cache and may be coupled to dedicated memory. Therefore, in some examples, each of the vector processors may be configured to operate independently of the other vector processors. In other examples, the vector processors included in a particular PVA may be configured to use data parallelism. For example, in some embodiments, the multiple vector processors included in a single PVA may execute the same computer vision algorithm, but on different regions of an image. In other examples, the vector processors included in a particular PVA may concurrently execute different computer vision algorithms on the same image, or even execute different algorithms on consecutive images or portions of an image.Among other things, any number of PVAs can be included in the hardware acceleration cluster, and any number of vector processors can be included in each of the PVAs. Furthermore, the one or more PVAs can contain additional memory for error correcting code (ECC) to increase the overall security of the system.

[0072] The one or more accelerators 114 (e.g., the hardware acceleration cluster) may include an on-chip computer vision network and SRAM to provide high-bandwidth, low-latency SRAM to the one or more accelerators 114. In some examples, the on-chip memory may include at least 4 MB of SRAM, consisting of, for example, and without limitation, eight field-configurable memory blocks accessible by both the PVA and the DLA. Each pair of memory blocks may include an Advanced Peripheral Bus (APB) interface, configuration circuitry, a controller, and a multiplexer. Any type of memory may be used. The PVA and the DLA may access the memory through a backbone that provides high-speed access to the memory to the PVA and the DLA.The backbone may include an on-chip computer vision network connecting the PVA and DLA to the memory (e.g., using the APB).

[0073] The on-chip computer vision network can include an interface that determines that both the PVA and the DLA are delivering ready and valid signals before transmitting control signals / addresses / data. Such an interface can provide separate phases and channels for transmitting control signals / addresses / data, as well as bursty communication for continuous data transmission. This type of interface can conform to ISO 26262 or IEC 61508, although other standards and protocols can also be used.

[0074] In some examples, the one or more SoCs 104 may include a real-time ray tracing hardware accelerator, as described in U.S. Patent Application No. 16 / 101,232, filed August 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the positions and extents of objects (e.g., within a world model) to generate real-time visualization simulations, for RADAR signal interpretation, for sound propagation synthesis and / or analysis, for simulation of SONAR systems, for general wave propagation simulation, for comparison with LIDAR data for localization, and / or for other functions, and / or for other purposes. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more operations related to ray tracing.

[0075] The one or more accelerators 114 (e.g., the hardware accelerator cluster) have a wide range of uses for autonomous driving. The PVA can be a programmable vision accelerator that can be used for critical processing steps in ADAS and autonomous vehicles. The capabilities of the PVA are well suited to algorithmic areas that require predictable processing with low power and low latency. In other words, the PVA is well suited for semi-dense or dense regular computations, even on small datasets, that require predictable runtimes with low latency and low power. In the context of autonomous vehicle platforms, the PVAs are therefore designed to execute classical computer vision algorithms because they are efficient at object detection and operate on integer mathematics.

[0076] According to one embodiment of the technology, the PVA is used, for example, to perform computer stereovision. In some examples, a semi-global matching-based algorithm may be used, although this is not intended as a limitation. Many applications for Level 3-5 autonomous driving require on-the-fly motion estimation or stereo matching (e.g., structure from motion, pedestrian detection, lane detection, etc.). The PVA can perform a computer stereovision function on inputs from two monocular cameras.

[0077] In some examples, the PVA can be used to perform dense optical flow, such as processing raw radar data (e.g., using a 4D Fast Fourier Transform) to provide processed radar. In other examples, the PVA is used for time-of-flight depth processing, such as processing raw time-of-flight data to provide processed time-of-flight data.

[0078] The DLA can be used to power any type of network to improve control and driving safety; for example, this includes a neural network that outputs a confidence measure for each object detection. Such a confidence value can be interpreted as a probability or as providing a relative "weight" to each detection compared to other detections. This confidence value allows the system to make further decisions about which detections should be considered true positives rather than false positives. For example, the system can set a confidence threshold and consider only those detections that exceed the threshold as true positives.In an automatic emergency braking (AEB) system, false positive detections would cause the vehicle to automatically perform emergency braking, which is obviously undesirable. Therefore, only the most certain detections should be considered as triggers for AEB. The DLA may employ a neural network to regress the confidence value. The neural network may take as input at least a subset of parameters, such as, but not limited to, the bounding box dimensions, the ground plane estimate obtained (e.g., from another subsystem), the output of the inertial measurement unit (IMU) sensor 166 correlated with the orientation of the vehicle 100, distance, and 3D position estimates of the object obtained by the neural network and / or other sensors (e.g., one or more LIDAR sensors 164 or one or more RADAR sensors 160).

[0079] The one or more SoCs 104 may include one or more data stores 116 (e.g., memory). The one or more data stores 116 may be on-chip memory on the one or more SoCs 104 in which neural networks to be executed on the GPU and / or the DLA may be stored. In some examples, the one or more data stores 116 may be large enough to store multiple instances of neural networks for redundancy and security. The one or more data stores 112 may include one or more L2 or L3 caches 112. The reference to the one or more data stores 116 may include a reference to the memory associated with the PVA, the DLA, and / or one or more other accelerators 114, as described herein.

[0080] The one or more SoCs 104 may include one or more processors 110 (e.g., embedded processors). The one or more processors 110 may include a boot and power management processor, which may be a dedicated processor and subsystem to handle the boot power and management functions and associated security enforcement. The boot and power management processor may be part of the boot sequence of the one or more SoCs 104 and may provide runtime power management services. The boot power and power management processor may provide clock and voltage programming, assisting with system transitions to a low-power state, managing the thermals and temperature sensors of the one or more SoCs 104, and / or managing the one or more SoCs 104 power states.Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the one or more SoCs 104 may use the ring oscillators to sense the temperatures of the one or more CPUs 106, the one or more GPUs 108, and / or the one or more accelerators 114. If it is determined that the temperatures exceed a threshold, the boot and power management processor may enter a temperature fault routine and place the one or more SoCs 104 into a lower-power state and / or place the vehicle 100 into a chauffeur-to-safe-stop mode (e.g., bring the vehicle 100 to a safe stop).

[0081] The one or more processors 110 may also include a set of embedded processors that can serve as an audio processing engine. The audio processing engine may be an audio subsystem that enables full hardware support for multi-channel audio across multiple interfaces and a wide and flexible range of audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core with a digital signal processor with dedicated RAM.

[0082] The one or more processors 110 may also include an always-on processor engine that can provide the necessary hardware functions to support low-power sensor management and wake-up use cases. The always-on processor engine may include a processor core, tightly coupled RAM, supporting peripherals (e.g., timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0083] The one or more processors 110 may also include a safety cluster engine containing a dedicated processor subsystem for safety management of automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (e.g., timers, an interrupt controller, etc.), and / or routing logic. In a safety mode, the two or more cores may operate in a lockstep mode, functioning as a single core with comparison logic that captures any differences between their operations.

[0084] The one or more processors 110 may also include a real-time camera engine, which may include a dedicated processor subsystem for managing the real-time camera.

[0085] The one or more processors 110 may further include a high dynamic range signal processor, which may include an image signal processor, which is a hardware engine that is part of the camera processing pipeline.

[0086] The one or more processors 110 may include a video image compositor, which may be a processing block (e.g., implemented on a microprocessor) that implements video post-processing functions required by a video playback application to generate the final image for the player window. The video image compositor may perform lens distortion correction on the one or more wide-angle cameras 170, the one or more surround cameras 174, and / or the in-cabin surveillance camera sensors. The in-cabin surveillance camera sensor is preferably monitored by a neural network running on another instance of the enhanced SoC and configured to detect events in the cabin and respond accordingly.An in-cabin system can perform lip reading to activate cellular service and make a call, dictate emails, change the destination, activate or change the vehicle's infotainment system and settings, or enable voice-activated web browsing. Certain functions are available to the driver only when the vehicle is operating in autonomous mode and are disabled otherwise.

[0087] The video compositor can incorporate enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, when motion occurs in a video, the noise reduction weights the spatial information accordingly, reducing the weight of information provided by neighboring frames. If a frame or section of a frame contains no motion, the temporal noise reduction performed by the video compositor can use information from the previous frame to reduce noise in the current frame.

[0088] The video compositor may also be configured to perform stereo distortion correction on the input stereo lens images. The video compositor may also be used for user interface design when the operating system's desktop is in use and the GPU(s) 108 do not need to constantly render new surfaces. Even when the GPU(s) 108 are powered on and actively performing 3D rendering, the video compositor may be used to offload the GPU(s) 108, thereby improving performance and responsiveness.

[0089] The one or more SoCs 104 may also include a Mobile Industry Processor Interface (MIPI) serial camera interface for receiving video and input from cameras, a high-speed interface, and / or a video input block that may be used for camera and related pixel input functions. The one or more SoCs 104 may also include one or more input / output controllers, one or more of which may be software-controlled and used to receive I / O signals that are not assigned to a specific role.

[0090] The one or more SoCs 104 may further include a wide range of peripheral interfaces to enable communication with peripheral devices, audio codecs, power management, and / or other devices. The one or more SoCs 104 may be used to process data from cameras (e.g., via Gigabit Multimedia Serial Link and Ethernet), sensors (e.g., one or more LIDAR sensors 164, one or more RADAR sensors 160, etc., which may be connected via Ethernet), data from bus 102 (e.g., speed of vehicle 100, steering wheel position, etc.), and data from one or more GNSS sensors 158 (e.g., connected via Ethernet or CAN bus).The one or more SoCs 104 may further include dedicated high-performance mass storage controllers, which may include their own DMA engines and which may be used to offload routine data management tasks from the one or more CPUs 106.

[0091] The one or more SoCs 104 may be an end-to-end platform with a flexible architecture spanning automation levels 3-5, thereby providing a comprehensive functional safety architecture that supports and efficiently utilizes computer vision and ADAS techniques for diversity and redundancy, and provides a platform for a flexible, reliable driving software stack along with deep learning tools. The one or more SoCs 104 may be faster, more reliable, and even more energy and space efficient than conventional systems. For example, the one or more accelerators 114, in combination with the one or more CPUs 106, the one or more GPUs 108, and the one or more data memories 116, may form a fast, efficient platform for Level 3-5 autonomous vehicles.

[0092] The technology thus offers capabilities and functions that cannot be achieved by conventional systems. For example, computer vision algorithms can be executed on CPUs that can be configured using a high-level programming language, such as the C programming language, to execute a variety of processing algorithms on a wide variety of visual data. However, CPUs are often unable to meet the performance requirements of many computer vision applications, such as execution time and power consumption. In particular, many CPUs are unable to execute complex object detection algorithms in real time, which is a prerequisite for in-vehicle ADAS applications and a requirement for practical Level 3-5 autonomous vehicles.

[0093] In contrast to conventional systems, the technology described herein enables the simultaneous and / or sequential execution of multiple neural networks and the combination of the results to enable Level 3-5 autonomous driving functionality by providing a CPU complex, a GPU complex, and a hardware acceleration cluster. For example, a CNN running on the DLA or dGPU (e.g., the one or more GPUs 120) may include text and word recognition, allowing the supercomputer to read and understand traffic signs, including signs for which the neural network has not been specifically trained. The DLA may further include a neural network capable of identifying, interpreting, and providing semantic understanding of the sign, and passing this semantic understanding to the path planning modules running on the CPU complex.

[0094] Another example is that multiple neural networks can run simultaneously, as required for driving at Level 3, 4, or 5. For example, a warning sign reading "Caution: Flashing lights indicate black ice," along with an electric light, can be interpreted independently or jointly by multiple neural networks. The sign itself can be identified as a traffic sign by a first deployed neural network (e.g., a trained neural network), and the text "Flashing lights indicate black ice" can be interpreted by a second deployed neural network, which informs the vehicle's path-planning software (preferably running on the CPU complex) that if flashing lights are detected, black ice is present.The turn signal can be identified across multiple images by a third neural network, which informs the vehicle's path planning software of the presence (or absence) of turn signals. All three neural networks can run simultaneously, e.g., within the DLA and / or on the one or more GPUs 108.

[0095] In some examples, a facial recognition and vehicle owner identification CNN may use data from camera sensors to identify the presence of an authorized driver and / or owner of the vehicle 100. The always-on sensor processing engine may be used to unlock the vehicle and turn on the lights when the owner approaches the driver's door, and to disable the vehicle in security mode when the owner exits the vehicle. In this way, the one or more SoCs 104 provide security against theft and / or carjacking.

[0096] In another example, a CNN for detecting and identifying emergency vehicles may use data from microphones 196 to detect and identify emergency vehicle sirens. Unlike conventional systems that use general classifiers to detect sirens and manually extract features, the one or more SoCs 104 use the CNN to classify environmental and urban sounds, as well as visual data. In a preferred embodiment, the CNN running on the DLA is trained to detect the relative approach speed of the emergency vehicle (e.g., by using the Doppler effect). The CNN may also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by one or more GNSS sensors 158.For example, if the CNN is operating in Europe, it will attempt to detect European sirens, and if it is operating in the United States, the CNN will attempt to identify only North American sirens. Once an emergency vehicle is detected, a controller can be used to execute an emergency vehicle safety routine, slow the vehicle, pull over to the side of the road, park the vehicle, and / or idle the vehicle, using ultrasonic sensors 162, until the one or more emergency vehicles pass by.

[0097] The vehicle may include one or more CPUs 118 (e.g., one or more discrete CPUs or one or more dCPUs) that may be coupled to the one or more SoCs 104 via a high-speed connection (e.g., PCIe). The CPUs 118 may include, for example, an x86 processor. The CPUs 118 may be used, for example, to perform a variety of functions, including reconciling potentially inconsistent results between ADAS sensors and the one or more SoCs 104 and / or monitoring the status and health of the one or more controllers 136 and / or the infotainment SoC 130.

[0098] The vehicle 100 may include one or more GPUs 120 (e.g., one or more discrete GPUs or one or more dGPUs) that may be coupled to the one or more SoCs 104 via a high-speed interconnect (e.g., NVIDIA's NVLINK). The one or more GPUs 120 may provide additional artificial intelligence capabilities, e.g., by executing redundant and / or distinct neural networks, and may be used to train and / or update neural networks based on inputs (e.g., sensor data) from sensors of the vehicle 100.

[0099] The vehicle 100 may further include the network interface 124, which may include one or more wireless antennas 126 (e.g., one or more wireless antennas for various communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). The network interface 124 may be used to enable a wireless connection via the internet to the cloud (e.g., to the one or more servers 178 and / or other network devices), to other vehicles, and / or to computing devices (e.g., passenger client devices). To communicate with other vehicles, a direct connection between the two vehicles and / or an indirect connection may be established (e.g., via networks and the internet). Direct connections may be established via vehicle-to-vehicle communication.Vehicle-to-vehicle communication may provide vehicle 100 with information about vehicles in the vicinity of vehicle 100 (e.g., vehicles in front of, beside, and / or behind vehicle 100). This functionality may be part of a cooperative adaptive cruise control feature of vehicle 100.

[0100] The network interface 124 may include an SoC that provides modulation and demodulation functions and enables the one and more controllers 136 to communicate over wireless networks. The network interface 124 may include a radio frequency front end for upconverting from baseband to radio frequency and downconverting from radio frequency to baseband. The frequency conversions may be performed using known methods and / or super-heterodyne methods. In some examples, the radio frequency front end functionality may be provided by a separate chip. The network interface may include wireless functionality for communicating over LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-Wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0101] The vehicle 100 may further include one or more data stores 128, which may be located off-chip (e.g., outside the SoCs 104). The one or more data stores 128 may include one or more memory elements, including RAM, SRAM, DRAM, VRAM, flash, hard drives, and / or other components and / or devices capable of storing at least one bit of data.

[0102] The vehicle 100 may further include one or more GNSS sensors 158. The one or more GNSS sensors 158 (e.g., GPS, assisted GPS sensors, differential GPS (DGPS) sensors, etc.) assist in mapping, sensing, occupancy grid creation, and / or path planning. Any number of GNSS sensors 158 may be used, including, for example, and without limitation, a GPS using a USB port with an Ethernet-to-serial (RS-232) bridge.

[0103] The vehicle 100 may further include one or more RADAR sensors 160. The one or more RADAR sensors 160 may be used by the vehicle 100 for long-range vehicle detection, even in darkness and / or adverse weather conditions. The functional safety level of the RADAR may be ASIL B. The one or more RADAR sensors 160 may utilize the CAN and / or bus 102 (e.g., to transmit the data generated by the one or more RADAR sensors 160) for control and access to object tracking data, with access to the raw data occurring via Ethernet in some examples. A variety of RADAR sensor types may be used. The one or more RADAR sensors 160 may be suitable for, for example, front-, rear-, and side-facing RADAR, without limitation. In some examples, one or more Pulse Doppler RADAR sensors are used.

[0104] The one or more radar sensors 160 may include various configurations, such as long range with a narrow field of view, short range with a wide field of view, short range side coverage, etc. In some examples, long range radar may be used for the adaptive cruise control function. Long range radar systems may provide a wide field of view realized by two or more independent scans, e.g., within a range of 250 m. The one or more radar sensors 160 may assist in distinguishing between static and moving objects and may be used by ADAS systems for emergency braking assistance and forward collision warning. Long range radar sensors may include a monostatic multimodal radar with multiple (e.g., six or more) fixed radar antennas and a high-speed CAN and FlexRay interface.In a six-antenna example, the middle four antennas can create a focused beam pattern designed to detect the surroundings of vehicle 100 at higher speeds with minimal interference from traffic in adjacent lanes. The other two antennas can expand the field of view so that vehicles entering or exiting the lane of vehicle 100 can be quickly detected.

[0105] Medium-range radar systems, for example, can have a range of up to 160 m (front) or 80 m (rear) and a field of view of up to 42 degrees (front) or 150 degrees (rear). Short-range radar systems can include, among other features, radar sensors designed for installation at both ends of the rear bumper. When installed at both ends of the rear bumper, such a radar sensor system can generate two beams that continuously monitor the blind spot area to the rear and side of the vehicle.

[0106] Short-range radar systems can be used in an ADAS system to monitor blind spots and / or assist with lane change.

[0107] The vehicle 100 may also include one or more ultrasonic sensors 162. The one or more ultrasonic sensors 162, which may be mounted on the front, rear, and / or sides of the vehicle 100, may be used for parking assistance and / or for creating and updating an occupancy grid. A plurality of ultrasonic sensors 162 may be used, and different ultrasonic sensors 162 may be used for different detection ranges (e.g., 2.5 m, 4 m). The one or more ultrasonic sensors 162 may operate at functional safety levels of ASIL B.

[0108] The vehicle 100 may include one or more LIDAR sensors 164. The one or more LIDAR sensors 164 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. The one or more LIDAR sensors 164 may conform to ASIL B functional safety level. In some examples, the vehicle 100 may include multiple LIDAR sensors 164 (e.g., two, four, six, etc.) that may use Ethernet (e.g., to deliver data to a Gigabit Ethernet switch).

[0109] In some examples, the one or more LIDAR sensors 164 may be capable of providing a list of objects and their distances for a 360-degree field of view. For example, commercially available LIDAR sensors 164 may have an advertised range of approximately 100 m, with an accuracy of 2 cm to 3 cm, and support for a 100 Mbps Ethernet connection. In some examples, one or more non-prominent LIDAR sensors 164 may be used. In such examples, the one or more LIDAR sensors 164 may be implemented as a small device that may be embedded in the front, rear, sides, and / or corners of the vehicle 100. In such examples, the one or more LIDAR sensors 164 may provide a horizontal field of view of up to 120 degrees and a vertical field of view of up to 35 degrees, with a range of 200 m, even for objects with low reflectivity.The one or more front-mounted LIDAR sensors 164 may be configured for a horizontal field of view between 45 degrees and 135 degrees.

[0110] In some examples, LIDAR technologies, such as 3D flash LIDAR, may also be used. 3D flash LIDAR uses a laser flash as the transmitting source to illuminate the vehicle's surroundings up to approximately 200 m. A flash LIDAR unit contains a receptor that records the time of flight of the laser pulse and the reflected light at each pixel, which in turn corresponds to the distance between the vehicle and the objects. Flash LIDAR can enable highly accurate and distortion-free images of the surroundings to be created with each laser flash. In some examples, four flash LIDAR sensors may be deployed, one on each side of the vehicle. Available 3D flash LIDAR systems include a solid-state 3D focal plane array LIDAR camera that contains no moving parts other than a fan (e.g., a non-scanning LIDAR device).The flash lidar device can use a 5-nanosecond pulse of a Class I (eye-safe) laser per frame and collect the reflected laser light as 3D range point clouds and co-registered intensity data. By using flash lidar, and because flash lidar is a solid-state device with no moving parts, the one or more lidar sensors 164 can be less susceptible to motion blur, vibration, and / or shock.

[0111] The vehicle may also include one or more IMU sensors 166. The one or more IMU sensors 166 may, in some examples, be located at the center of the rear axle of the vehicle 100. The one or more IMU sensors 166 may, for example and without limitation, include one or more accelerometers, one or more magnetometers, one or more gyroscopes, one or more magnetic compasses, and / or other types of sensors. In some examples, such as in six-axis applications, the one or more IMU sensors 166 may include accelerometers and gyroscopes, while in nine-axis applications, the one or more IMU sensors 166 may include accelerometers, gyroscopes, and magnetometers.

[0112] In some embodiments, the one or more IMU sensors 166 may be implemented as a miniaturized, high-performance GPS-aided inertial navigation system (GPS / INS) that combines microelectromechanical system (MEMS) inertial sensors, a high-sensitivity GPS receiver, and advanced Kalman filter algorithms to provide estimates of position, velocity, and attitude. Thus, in some examples, the one or more IMU sensors 166 may enable the vehicle 100 to estimate heading without requiring input from a magnetic sensor by directly observing and correlating velocity changes from the GPS with the one or more IMU sensors 166. In some examples, the one or more IMU sensors 166 and the one or more GNSS sensors 158 may be combined into a single integrated unit.

[0113] The vehicle may include one or more microphones 196 mounted in and / or around the vehicle 100. The one or more microphones 196 may be used, among other things, to detect and identify emergency vehicles.

[0114] The vehicle may further include any number of camera types, including one or more stereo cameras 168, one or more wide-angle cameras 170, one or more infrared cameras 172, one or more surround cameras 174, one or more long-range and / or medium-range cameras 198, and / or other camera types. The cameras may be used to capture image data around the entire periphery of the vehicle 100. The types of cameras used depend on the embodiments and requirements of the vehicle 100, and any combination of camera types may be used to provide the necessary coverage around the vehicle 100. Furthermore, the number of cameras may vary depending on the embodiment. For example, the vehicle may include six cameras, seven cameras, ten cameras, twelve cameras, and / or a different number of cameras.The cameras may support, by way of example and without limitation, Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Each of the one or more cameras is referred to herein with reference to . Fig. 1A and Fig. 1B is described in more detail.

[0115] The vehicle 100 may further include one or more vibration sensors 142. The one or more vibration sensors 142 may measure vibrations from components of the vehicle, such as the one or more axles. For example, changes in vibrations may indicate a change in the road surface. In another example, when two or more vibration sensors 142 are used, the differences between the vibrations may be used to determine friction or slippage on the road surface (e.g., when the difference in vibration is between a driven axle and a free-spinning axle).

[0116] The vehicle 100 may include an ADAS system 138. The ADAS system 138 may include an SoC in some examples. The ADAS system 138 may include autonomous / adaptive / automatic cruise control (ACC), cooperative adaptive cruise control (CACC), forward crash warning (FCW), automatic emergency braking (AEB), lane departure warning (LDW), lane keep assist (LKA), blind spot warning (BSW), rear cross-traffic warning (RCTW), collision warning systems (CWS), lane centering (LC), and / or other features and functions.

[0117] The ACC systems may use one or more radar sensors 160, one or more lidar sensors 164, and / or one or more cameras. The ACC systems may include longitudinal ACC and / or lateral ACC. Longitudinal ACC monitors and controls the distance to the vehicle immediately in front of the vehicle 100 and automatically adjusts the vehicle speed to maintain a safe distance from preceding vehicles. Lateral ACC performs follow-through and advises the vehicle 100 to change lanes if necessary. Lateral ACC is associated with other ADAS applications, such as LCA and CWS.

[0118] The CACC utilizes information from other vehicles, which may be received via the network interface 124 and / or the one or more wireless antennas 126 from other vehicles over a wireless connection or indirectly via a network connection (e.g., over the Internet). Direct connections may be provided via a vehicle-to-vehicle (V2V) communication link, while indirect connections may be an infrastructure-to-vehicle (I2V) communication link. In general, the V2V communication concept provides information about the immediately preceding vehicles (e.g., vehicles immediately in front of the vehicle 100 and in the same lane as it), while the I2V communication concept provides information about traffic further ahead. CACC systems may include one or both I2V and V2V information sources.Given the information about the vehicles ahead of vehicle 100, CACC can be more reliable and has the potential to improve traffic flow and reduce congestion on the road.

[0119] FCW systems are designed to warn the driver of a hazard so they can take corrective action. FCW systems utilize a forward-facing camera and / or one or more radar sensors 160 coupled with a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to feedback to the driver, such as a display, speaker, and / or vibrating component. FCW systems can provide a warning, such as a sound, a visual warning, a vibration, and / or a rapid braking pulse.

[0120] AEB systems detect an impending frontal collision with another vehicle or object and can automatically apply the brakes if the driver does not take corrective action within a specified time or distance parameter. AEB systems may utilize one or more forward-facing cameras and / or one or more radar sensors 160 coupled with a dedicated processor, DSP, FPGA, and / or ASIC. When the AEB system detects a hazard, it typically first alerts the driver to take corrective action to avoid the collision; if the driver does not take corrective action, the AEB system can automatically apply the brakes to prevent or at least mitigate the effects of the predicted collision. AEB systems may incorporate techniques such as dynamic brake support and / or impending crash braking.

[0121] LDW systems provide visual, audible, and / or tactile warnings, such as steering wheel or seat vibrations, to alert the driver when the vehicle 100 crosses lane markings. An LDW system will not activate if the driver indicates an intentional lane departure by activating a turn signal. LDW systems may utilize forward-facing cameras coupled with a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to feedback to the driver, such as a display, speaker, and / or vibrating component.

[0122] LKA systems are a variant of LDW systems. LKA systems provide steering inputs or braking to correct the vehicle 100 when the vehicle 100 begins to depart from its lane.

[0123] BSW systems detect and warn the driver of vehicles in the car's blind spot. BSW systems may provide a visual, audible, and / or tactile warning signal to indicate that merging into or changing lanes is unsafe. The system may provide an additional warning when the driver activates a turn signal. BSW systems may utilize one or more rear-facing cameras and / or radar sensors 160 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to feedback to the driver, such as a display, speaker, and / or vibrating component.

[0124] RCTW systems can provide visual, audible, and / or tactile notification when an object outside the range of the rearview camera is detected when the vehicle 100 is reversing. Some RCTW systems include AEB to ensure the vehicle brakes are applied to avoid a crash. RCTW systems can utilize one or more rear-facing RADAR sensors 160 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to feedback to the driver, such as a display, speaker, and / or vibrating component.

[0125] Conventional ADAS systems can produce false positives, which can be annoying and distracting for the driver, but are typically not catastrophic because ADAS systems warn the driver and give them the opportunity to decide whether a safety issue truly exists and act accordingly. However, in an autonomous vehicle 100, in the event of conflicting results, the vehicle 100 must decide for itself whether to consider the result of a primary computer or a secondary computer (e.g., a first controller 136 or a second controller 136). In some embodiments, the ADAS system 138 may, for example, be a backup and / or secondary computer that provides perception information to a rationality module of the backup computer.The backup computer rationality monitor can run redundant, diverse software on hardware components to detect errors in perception and dynamic driving tasks. The outputs of the ADAS system 138 can be provided to a supervising MCU. If the outputs of the primary computer and the secondary computer conflict, the supervising MCU must determine how to resolve the conflict to ensure safe operation.

[0126] In some examples, the primary computer may be configured to provide the monitoring MCU with a confidence value indicating the primary computer's confidence in the chosen outcome. If the confidence value exceeds a threshold, the monitoring MCU may follow the primary computer's instruction regardless of whether the secondary computer provides a conflicting or inconsistent result. If the confidence value does not meet the threshold and the primary and secondary computers indicate different results (e.g., a conflict), the monitoring MCU may arbitrate between the computers to determine the appropriate outcome.

[0127] The monitoring MCU may be configured to run one or more neural networks trained and configured to determine, based on the outputs of the primary and secondary computers, the conditions under which the secondary computer triggers false alarms. This allows the one or more neural networks in the monitoring MCU to learn when the output of the secondary computer can and cannot be trusted. For example, if the secondary computer is a radar-based FCW system, a neural network in the monitoring MCU can learn when the FCW system identifies metallic objects that are not actually hazardous, such as a drain grate or manhole cover, which triggers an alarm.Similarly, if the secondary computer is a camera-based LDW system, a neural network in the monitoring MCU can learn to override the LDW system when cyclists or pedestrians are present and lane departure is indeed the safest maneuver. In embodiments including one or more neural networks running on the monitoring MCU, the monitoring MCU can include at least one DLA or GPU suitable for executing the one or more neural networks with associated memory. In preferred embodiments, the monitoring MCU can comprise and / or be included as a component of the one or more SoCs 104.

[0128] In other examples, ADAS system 138 may include a secondary computer that executes ADAS functionality according to classic computer vision rules. Thus, the secondary computer may use classic computer vision rules (if-then), and the presence of one or more neural networks in the supervising MCU may improve reliability, safety, and performance. For example, diverse implementation and intentional non-identity make the overall system more fault-tolerant, especially against errors caused by software (or software-hardware interfaces).For example, if a software bug or error occurs in the software on the primary computer and the non-identical software code on the secondary computer produces the same overall result, the monitoring MCU can have greater confidence that the overall result is correct and the bug in the software or hardware on the primary computer does not cause a significant error.

[0129] In some examples, the output of the ADAS system 138 may be fed into the perception block of the primary computer and / or the dynamic driving task block of the primary computer. For example, if the ADAS system 138 displays a forward collision warning due to an object immediately in front of the vehicle, the perception block may use this information in identifying objects. In other examples, the secondary computer may have its own neural network trained to reduce the risk of false positives, as described herein.

[0130] The vehicle 100 may further include the infotainment SoC 130 (e.g., an in-vehicle infotainment (IVI) system). Although illustrated and described as an SoC, the infotainment system may not be an SoC and may include two or more discrete components. The infotainment SoC 130 may include a combination of hardware and software that can be used to provide the vehicle 100 with audio (e.g., music, a personal digital assistant, navigation directions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), phone (e.g., hands-free calling), network connectivity (e.g., LTE, Wi-Fi, etc.), and / or information services (e.g., navigation systems, rear parking assist, a radio data system, vehicle-related information such as fuel level, total distance traveled, brake fuel level, oil level, door open / close, air filter information, etc.).The infotainment SoC 130 may include, for example, radios, record players, navigation systems, video players, USB and Bluetooth connectivity, car computers, in-car entertainment, Wi-Fi, steering wheel audio controls, hands-free calling, a head-up display (HUD), an HMI display 134, a telematics device, a control panel (e.g., for controlling and / or interacting with various components, functions, and / or systems), and / or other components. The infotainment SoC 130 may further be used to provide information (e.g., visual and / or audible) to one or more users of the vehicle, such as information from the ADAS system 138, autonomous driving information such as planned vehicle maneuvers, road layouts, environmental information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0131] The infotainment SoC 130 may include GPU functionality. The infotainment SoC 130 may communicate with other devices, systems, and / or components of the vehicle 100 via the bus 102 (e.g., CAN bus, Ethernet, etc.). In some examples, the infotainment SoC 130 may be coupled to a supervisory MCU so that the infotainment system's GPU may perform some self-driving functions if the one or more primary controllers 136 (e.g., the primary and / or backup computers of the vehicle 100) fail. In such an example, the infotainment SoC 130 may place the vehicle 100 into a chauffeur-to-safe-stop mode, as described herein.

[0132] The vehicle 100 may further include an instrument cluster 132 (e.g., a digital instrument panel, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 132 may include a controller and / or supercomputer (e.g., a discrete controller or supercomputer). The instrument cluster 132 may include a variety of instruments, such as a speedometer, fuel level, oil pressure, tachometer, odometer, turn signals, shift position indicator, seat belt warning light(s), parking brake warning light(s), engine malfunction light(s), airbag system (SRS) information, lighting controls, safety system controls, navigation information, etc. In some examples, information from the infotainment SoC 130 and the instrument cluster 132 may be displayed and / or shared. In other words, the instrument cluster 132 may be included as part of the infotainment SoC 130, or vice versa.

[0133] Fig. 1D is a system diagram for communication between the one or more cloud-based servers and the example autonomous vehicle 100 of Fig. 1A, according to some embodiments of the present disclosure. The system 176 may include the one or more servers 178, the one or more networks 190, and the vehicles, including the vehicle 100. The server(s) 178 may include a plurality of GPUs 184(A)-184(H) (collectively referred to herein as GPUs 184), PCIe switches 182(A)-182(H) (collectively referred to herein as PCIe switches 182), and / or CPUs 180(A)-180(B) (collectively referred to herein as CPUs 180). The GPUs 184, the CPUs 180, and the PCIe switches may be interconnected with high-speed interconnects, such as, without limitation, the NVLink interfaces 188 and / or PCIe interconnects 186 developed by NVIDIA. In some examples, the GPUs 184 are connected via NVLink and / or NVSwitch SoC, and the GPUs 184 and the PCIe switches 182 are connected via PCIe links.Although eight GPUs 184, two CPUs 180, and two PCIe switches are illustrated, this is not intended to be limiting. Depending on the embodiment, each of the servers 178 may include any number of GPUs 184, CPUs 180, and / or PCIe switches. For example, the one or more servers 178 may each include eight, sixteen, thirty-two, and / or more GPUs 184.

[0134] The one or more servers 178 may receive, via the one or more networks 190 and from the vehicles, image data representative of images depicting unexpected or changed road conditions, such as recently commenced roadwork. The one or more servers 178 may transmit, via the one or more networks 190 and to the vehicles, neural networks 192, updated neural networks 192, and / or map information 194 containing information about traffic and road conditions. The updates to the map information 194 may include updates to the HD map 122, such as information about construction, potholes, detours, flooding, and / or other obstacles.In some examples, the neural networks 192, the updated neural networks 192, and / or the map information 194 may result from new training and / or experience represented in the data received from any number of vehicles in the environment and / or may be based on training performed in a data center (e.g., using the one or more servers 178 and / or other servers).

[0135] The one or more servers 178 may be used to train machine learning models (e.g., neural networks) based on training data. The training data may be generated by the vehicles and / or in a simulation (e.g., using a game engine). In some examples, the training data is tagged (e.g., when the neural network benefits from supervised learning) and / or undergoes other preprocessing, while in other examples, the training data is not tagged and / or preprocessed (e.g., when the neural network does not require supervised learning).Training may be performed using one or more classes of machine learning techniques, including, without limitation, supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, federated learning, transfer learning, feature learning (including principal component and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including replacement dictionary learning), rule-based machine learning, anomaly detection, and any variations or combinations thereof. Once the machine learning models are trained, the machine learning models may be used by the vehicles (e.g., transmitted to the vehicles via the one or more networks 190) and / or the machine learning models may be used by the one or more servers 178 to remotely monitor the vehicles.

[0136] In some examples, the one or more servers 178 may receive data from the vehicles and apply the data to real-time, real-time neural networks for intelligent inference. The one or more servers 178 may include deep learning supercomputers and / or dedicated AI computers powered by GPUs 184, such as the DGX and DGX Station machines developed by NVIDIA. However, in some examples, the one or more servers 178 may include a deep learning infrastructure using only CPU-powered data centers.

[0137] The deep learning infrastructure of the one or more servers 178 may be capable of performing rapid inference in real time and may utilize this capability to evaluate and verify the state of the processors, software, and / or associated hardware in the vehicle 100. For example, the deep learning infrastructure may receive periodic updates from the vehicle 100, such as a sequence of images and / or objects that the vehicle 100 has located in that sequence of images (e.g., via computer vision and / or other machine learning object classification techniques).The deep learning infrastructure may run its own neural network to identify the objects and compare them to the objects identified by the vehicle 100, and if the results do not match and the infrastructure concludes that the AI ​​in the vehicle 100 is not functioning properly, the one or more servers 178 may send a signal to the vehicle 100 instructing a fail-safe computer of the vehicle 100 to take control, notify passengers, and perform a safe parking maneuver.

[0138] For inferencing, the one or more servers 178 may include GPUs 184 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT). The combination of GPU-driven servers and inference accelerators may enable real-time responsiveness. In other examples, such as when performance is less critical, servers powered by CPUs, FPGAs, and other processors may be used for inferencing. EXAMPLE COMPUTER DEVICE

[0139] Fig. 2 is a block diagram of an exemplary computing device 200 suitable for use in implementing some embodiments of the present disclosure. Computing device 200 may include an interconnect system 202 that directly or indirectly couples the following devices: memory 204, one or more central processing units (CPUs) 206, one or more graphics processing units (GPUs) 208, a communications interface 210, input / output (I / O) ports 212, input / output components 214, a power supply 216, one or more presentation components 218 (e.g., display(s)), and one or more logic units 220. In at least one embodiment, the one or more computing devices 200 may include one or more virtual machines (VMs), and / or each of the components thereof may include virtual components (e.g., virtual hardware components).As non-limiting examples, one or more of the GPUs 208 may include one or more vGPUs, one or more of the CPUs 206 may include one or more vCPUs, and / or one or more of the logic units 220 may include one or more virtual logic units. Thus, a computing device 200 may include discrete components (e.g., a full GPU associated with the computing device 200), virtual components (e.g., a portion of a GPU associated with the computing device 200), or a combination thereof.

[0140] Although the different blocks of Fig. 2 as being connected to wires via the interconnect system 202, this is not intended as a limitation and is for clarity only. For example, in some embodiments, a presentation component 218, such as a display device, may be considered an I / O component 214 (e.g., if the display is a touchscreen). As another example, the CPUs 206 and / or GPUs 208 may include memory (e.g., the memory 204 may represent a storage device in addition to the memory of the GPUs 208, the CPUs 206, and / or other components). In other words, the computing device of Fig. 2 is merely illustrative. It does not distinguish between categories such as “workstation,” “server,” “laptop,” “desktop,” “tablet,” “client device,” “mobile device,” “handheld device,” “game console,” “electronic control unit (ECU),” “virtual reality system,” and / or other device or system types, since all fall within the scope of the computing device of Fig. 2 may be considered.

[0141] The interconnect system 202 may represent one or more connections or buses, such as an address bus, a data bus, a control bus, or a combination thereof. The interconnect system 202 may include one or more bus or connection types, such as an Industry Standard Architecture (ISA) bus, an Extended ISA bus, a Video Electronics Standards Association (VESA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI Express (PCIe) bus, and / or another type of bus or connection. In some embodiments, there are direct connections between components. For example, the CPU 206 may be directly connected to the memory 204. Further, the CPU 206 may be directly connected to the GPU 208.For a direct or point-to-point connection between components, interconnect system 202 may include a PCIe link to establish the connection. In these examples, a PCI bus need not be included in computing device 200.

[0142] Memory 204 may include a variety of computer-readable media. The computer-readable media may be any available media accessible by the computing device 200. The computer-readable media may include both volatile and non-volatile media, as well as removable and non-removable media. By way of example and without limitation, the computer-readable media may include computer storage media and communication media.

[0143] The computer storage media may include both volatile and non-volatile media and / or removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, and / or other data types. For example, the memory 204 may store computer-readable instructions (e.g., representing one or more programs) and / or one or more program elements, such as an operating system. Computer storage media may include, but is not limited to, RAM, ROM, EEPROM, flash memory or other storage technologies, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by the computing device 200.As used herein, computer storage media does not per se include signals.

[0144] The computer storage media may embody computer-readable instructions, data structures, program modules, and / or other data types in a modulated data signal, such as a carrier wave or other transport mechanism, and may include any media for conveying information. The term "modulated data signal" may refer to a signal having one or more of its characteristics adjusted or altered to encode information in the signal. The computer storage media may include, for example, and is not limited to, wired media, such as a wired network or a direct-wired connection, and wireless media, such as acoustic, RF, infrared, and other wireless media. Combinations of the above should also be within the scope of computer-readable media.

[0145] The one or more CPUs 206 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 200 to perform one or more of the methods and / or processes described herein. The one or more CPUs 206 may each include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) capable of concurrently executing a plurality of software threads. The one or more CPUs 206 may include any type of processor and may include different types of processors depending on the type of computing device 200 implemented (e.g., processors with fewer cores for mobile devices and processors with more cores for servers).Depending on the type of computing device 200, the processor may be, for example, an Advanced RISC Machines (ARM) processor implemented with reduced instruction set computing (RISC) or an x86 processor implemented with complex instruction set computing (CISC). Computing device 200 may include one or more CPUs 206, in addition to one or more microprocessors or additional coprocessors, such as math coprocessors.

[0146] In addition or alternatively to the one or more CPUs 206, the one or more GPUs 208 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 200 to perform one or more of the methods and / or processes described herein. One or more of the GPUs 208 may be an integrated GPU (e.g., with one or more of the CPUs 206) and / or one or more of the GPUs 208 may be a discrete GPU. In embodiments, one or more of the GPUs 208 may be a co-processor of one or more of the CPUs 206. The one or more GPUs 208 may be used by the computing device 200 to render graphics (e.g., 3D graphics) or to perform general-purpose computations. The one or more GPUs 208 may be used, for example, for general-purpose computing on GPUs (GPGPU).The one or more GPUs 208 may include hundreds or thousands of cores capable of processing hundreds or thousands of software threads simultaneously. The one or more GPUs 208 may generate pixel data for output images in response to rendering commands (e.g., rendering commands from the one or more CPUs 206 received via a host interface). The one or more GPUs 208 may include graphics memory, such as display memory, for storing pixel data or other suitable data, such as GPGPU data. The display memory may be included as part of the memory 204. The one or more GPUs 208 may include two or more GPUs operating in parallel (e.g., via a link). The link may connect the GPUs directly (e.g., using NVLINK) or connect the GPUs via a switch (e.g., using NVSwitch).When combined, each GPU 208 can generate pixel data or GPGPU data for different sections of an output or for different outputs (e.g., a first GPU for a first image and a second GPU for a second image). Each GPU can contain its own memory or share memory with other GPUs.

[0147] In addition to or alternatively to the one or more CPUs 206 and / or the one or more GPUs 208, the one or more logic units 220 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 200 to perform one or more of the methods and / or processes described herein. In embodiments, the one or more CPUs 206, the one or more GPUs 208, and / or the one or more logic units 220 may discretely or jointly execute any combination of the methods, processes, and / or portions thereof. One or more of the logic units 220 may be part of and / or integrated with one or more of the CPUs 206 and / or one or more of the GPUs 208, and / or one or more of the logic units 220 may be discrete components or otherwise external to the CPUs 206 and / or the GPUs 208.In embodiments, one or more of the logic units 220 may be a co-processor of one or more of the CPUs 206 and / or one or more of the GPUs 208.

[0148] Examples of the one or more logic units 220 include one or more processing cores and / or components thereof, such as data processing units (DPUs), tensor cores (TCs), tensor processing units (TPUs), pixel visual cores (PVCs), vision processing units (VPUs), graphics processing clusters (GPCs), texture processing clusters (TPCs), streaming multiprocessors (SMs), tree traversal units (TTUs), artificial intelligence accelerators (AIAs), deep learning accelerators (DLAs), arithmetic logic units (ALUs), application-specific integrated circuits (Application-Specific Integrated Circuits, ASICs), Floating Point Units (FPUs),Input / output (I / O) elements, peripheral component interconnect (PCI) or PCI Express (PCIe) elements, and / or similar.

[0149] The communication interface 210 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 200 to communicate with other computers over an electronic network, including wired and / or wireless communication. The communication interface 210 may include components and functions that enable communication over a variety of networks, such as wireless networks (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), wired networks (e.g., communication over Ethernet or InfiniBand), low-power wide-area networks (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.In one or more embodiments, the one or more logic units 220 and / or the communication interface 210 may include one or more data processing units (DPUs) to transfer data received over a network and / or via the interconnect system 202 directly to one or more GPUs 208 (e.g., a memory thereof).

[0150] The I / O ports 212 may enable the computing device 200 to be logically coupled to other devices, including the I / O components 214, the one or more presentation components 218, and / or other components, some of which may be built into (e.g., integrated) the computing device 200. Illustrative I / O components 214 include a microphone, a mouse, a keyboard, a joystick, a gamepad, a game controller, a satellite dish, a scanner, a printer, a wireless device, etc. The I / O components 214 may provide a natural user interface (NUI) that processes air gestures, speech, or other physiological inputs generated by a user. In some cases, the inputs may be communicated to a suitable network element for further processing.An NUI may implement any combination of speech capture, stylus capture, facial capture, biometric capture, both on-screen and off-screen gesture capture, air gestures, head and eye tracking, and touch capture (as further described below) associated with a display of computing device 200. Computing device 200 may include depth cameras, such as stereoscopic camera systems, infrared camera systems, RGB camera systems, touchscreen technology, and combinations thereof, for gesture capture and recognition. Additionally, computing device 200 may include accelerometers or gyroscopes (e.g., as part of an inertial measurement unit (IMU)) that enable motion capture. In some examples, the output of the accelerometers or gyroscopes from computing device 200 may be used to present immersive augmented reality or virtual reality.

[0151] Power supply 216 may include a hardwired power supply, a battery power supply, or a combination thereof. Power supply 216 may supply power to computing device 200 to enable operation of the components of computing device 200.

[0152] The one or more presentation components 218 may include a display (e.g., a monitor, a touchscreen, a television monitor, a head-up display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. The one or more presentation components 218 may receive data from other components (e.g., the one or more GPUs 208, the one or more CPUs 206, DPUs, etc.) and output the data (e.g., as an image, video, audio, etc.). EXEMPLARY DATA CENTER

[0153] Fig. Figure 3 illustrates an exemplary data center 300 that may be used in at least one embodiment of the present disclosure. Data center 300 may include a data center infrastructure layer 310, a framework layer 320, a software layer 330, and / or an application layer 340.

[0154] As in Fig. 3, the data center infrastructure layer 310 may include a resource orchestrator 312, clustered computing resources 314, and node computing resources (“node CRs”) 316(1)-316(N), where “N” represents any positive integer. In at least one embodiment, the node CRs 316(1)-316(N) may include, but are not limited to, any number of central processing units (CPUs) or other processors (including DPUs, accelerators, field programmable gate arrays (FPGAs), graphics processors or graphics processing units (GPUs), etc.), memory devices (e.g., dynamic read-only memory), storage devices (e.g., solid-state or disk drives), network input / output (NW I / O) devices, network switches, virtual machines (VMs), power modules and / or cooling modules, etc.In some embodiments, one or more Node CRs among Node CRs 316(1)-316(N) may correspond to a server having one or more of the computing resources mentioned above. Furthermore, in some embodiments, Node CRs 316(1)-316(N) may include one or more virtual components, such as vGPUs, vCPUs, and / or the like, and / or one or more of Node CRs 316(1)-316(N) may correspond to a virtual machine (VM).

[0155] In at least one embodiment, the grouped computing resources 314 may include separate groupings of node CRs 316 housed in one or more racks (not shown) or in many racks in data centers in different geographic locations (also not shown). Separate groupings of node CRs 316 within grouped computing resources 314 may include grouped computing, network, memory, or storage resources that may be configured or allocated to support one or more workloads. In at least one embodiment, multiple node CRs 316, including CPUs, GPUs, DPUs, and / or other processors, may be grouped in one or more racks to provide computing resources to support one or more workloads.The one or more racks may also contain any number of power modules, cooling modules, and / or network switches in any combination.

[0156] Resource orchestrator 312 may configure or otherwise control one or more node CRs 316(1)-316(N) and / or grouped computing resources 314. In at least one embodiment, resource orchestrator 312 may include an entity for managing the software design infrastructure (SDI) for data center 300. Resource orchestrator 312 may include hardware, software, or a combination thereof.

[0157] In at least one embodiment, as in Fig. 3, the framework layer 320 may include a job scheduler 333, a configuration manager 334, a resource manager 336, and / or a distributed file system 338. The framework layer 320 may include a framework that supports the software 332 of the software layer 330 and / or one or more applications 342 of the application layer 340. The software 332 or the one or more applications 342 may each include web-based service software or applications such as those provided by Amazon Web Services, Google Cloud, and Microsoft Azure. The framework layer 320 may be some type of free and open source software web application framework, such as, but not limited to, Apache Spark™ (hereinafter "Spark"), which may utilize a distributed file system 338 for processing large amounts of data (e.g., "Big Data").In at least one embodiment, job scheduler 333 may include a Spark driver to facilitate scheduling workloads supported by various layers of data center 300. Configuration manager 334 may be capable of configuring various layers, such as software layer 330 and framework layer 320, which includes Spark and distributed file system 338, to support processing large amounts of data. Resource manager 336 may be capable of managing clustered or grouped computing resources allocated or assigned to support distributed file system 338 and job scheduler 333. In at least one embodiment, the clustered or grouped computing resources may include clustered computing resource 314 at infrastructure layer 310 of the data center.The resource manager 336 may coordinate with the resource orchestrator 312 to manage these allocated or assigned computing resources.

[0158] In at least one embodiment, the software 332 included in software layer 330 may include software used by at least portions of node CRs 316(1)-316(N), clustered computing resources 314, and / or distributed file system 338 of framework layer 320. One or more types of software may include, among others, web page searching software, email virus scanning software, database software, and streaming video content software.

[0159] In at least one embodiment, the applications 342 included in the application layer 340 may include one or more types of applications used by at least portions of the node CRs 316(1)-316(N), the clustered computing resources 314, and / or the distributed file system 338 of the framework layer 320. One or more types of applications may include, but are not limited to, any number of genomic applications, cognitive computation, and machine learning applications, including training or inference software, machine learning framework software (e.g., PyTorch, TensorFlow, Caffe, etc.), and / or other machine learning applications used in connection with one or more embodiments.

[0160] In at least one embodiment, the configuration manager 334, the resource manager 336, and / or the resource orchestrator 312 may implement any number and type of self-modifying actions based on any amount and type of data collected in any technically feasible manner. Self-modifying actions may relieve a data center operator of the data center 300 from making potentially poor configuration decisions and potentially avoid underutilized and / or poorly performing sections of a data center.

[0161] Data center 300 may include tools, services, software, or other resources to train one or more machine learning models or to predict or infer information using one or more machine learning models according to one or more embodiments described herein. For example, one or more machine learning models may be trained by calculating weighting parameters according to a neural network architecture using software and / or computing resources described above with respect to data center 300.In at least one embodiment, trained or deployed machine learning models corresponding to one or more neural networks may be used to infer or predict information using the resources described above with reference to data center 300 using weighting parameters calculated by one or more training techniques, such as, but not limited to, those described herein.

[0162] In at least one embodiment, data center 300 may utilize CPUs, application-specific integrated circuits (ASICs), GPUs, FPGAs, and / or other hardware (or corresponding virtual computing resources) to perform training and / or inferencing using the resources described above. Furthermore, one or more of the software and / or hardware resources described above may be configured as a service to enable users to train or infer information, such as image capture, speech capture, or other artificial intelligence services. EXAMPLE NETWORK ENVIRONMENTS

[0163] Network environments suitable for implementing embodiments of the disclosure may include one or more client devices, servers, network attached storage (NAS), other backend devices, and / or other device types. The client devices, servers, and / or other device types (e.g., each device) may be implemented on one or more instances of the one or more computing devices 200 of Fig. 2 - for example, each device may include similar components, features, and / or functionality of the one or more computing devices 200. If backend devices (e.g., servers, NAS, etc.) are implemented, the backend devices may also be included as part of a data center 300, an example of which is described herein with reference to Fig. 3 is described in more detail.

[0164] The components of a network environment can communicate with each other over one or more networks, which can be wired, wireless, or both. The network can contain multiple networks or a network of networks. For example, the network can contain one or more wide area networks (WANs), one or more local area networks (LANs), one or more public networks such as the Internet and / or a public switched telephone network (PSTN), and / or one or more private networks. If the network contains a wireless telecommunications network, components such as a base station, a communications tower, or even access points (as well as other components) can provide wireless connectivity.

[0165] Compatible network environments may include one or more peer-to-peer network environments—in which case, a server may not be included in a network environment—and one or more client-server network environments—in which case, one or more servers may be included in a network environment. In peer-to-peer network environments, the functionality described herein may be implemented with reference to one or more servers on any number of client devices.

[0166] In at least one embodiment, a network environment may include one or more cloud-based network environments, a distributed computing environment, a combination thereof, etc. A cloud-based network environment may include a framework layer, a job scheduler, a resource manager, and a distributed file system implemented on one or more servers, which may include one or more core network servers and / or edge servers. A framework layer may include a framework for supporting software of a software layer and / or one or more applications of an application layer. The software or the one or more applications may each include web-based service software or applications. In embodiments, one or more of the client devices may use the web-based service software or applications (e.g.,by accessing the service software and / or applications through one or more application programming interfaces (APIs). The framework layer may be some type of free and open-source software web application framework, e.g., using, but not limited to, a distributed file system for processing large amounts of data (e.g., "Big Data").

[0167] A cloud-based network environment may provide cloud computing and / or cloud storage that performs any combination of the computing and / or data storage functions described herein (or one or more portions thereof). Each of these various functions may be distributed across multiple locations of central or core servers (e.g., one or more data centers that may be located across a state, region, country, globe, etc.). When a connection to a user (e.g., a client device) is relatively close to one or more edge servers, one or more core servers may offload at least some functionality to the one or more edge servers. A cloud-based network environment may be private (e.g., limited to a single organization), public (e.g., available to many organizations), and / or a combination thereof (e.g., a hybrid cloud environment).

[0168] The one or more client devices may include at least some of the components, features, and functions of the one or more described herein with reference to Fig. 2. By way of example, and not limitation, a client device may be embodied as a personal computer (PC), a laptop, a mobile device, a smartphone, a tablet computer, a smart watch, a wearable computer, a personal digital assistant (PDA), an MP3 player, a virtual reality headset, a global positioning system (GPS) or global positioning device, a video player, a video camera, a surveillance device or system, a vehicle, a boat, a hydrofoil, a virtual machine, a drone, a robot, a handheld communication device, a hospital device, a gaming device or gaming system, an entertainment system, a vehicle computing system, an embedded system controller, a remote control, an appliance, a consumer electronics device, a workstation, an edge device,any combination of these described devices or any other suitable device.

[0169] Referring to Fig. 4, each block of the method 400 described herein comprises a computational process that may be performed using any combination of hardware, firmware, and / or software. Various functions may be performed, for example, by a processor executing instructions stored in memory. The methods may also be embodied as computer-usable instructions stored on computer storage media. The methods may be provided by a standalone application, a service or hosted service (standalone or in combination with another hosted service), or a plug-in for another product, to name a few.Although method 400 may be described with reference to an autonomous vehicle by way of example, these methods may additionally or alternatively be performed by any system or combination of systems, including, but not limited to, the systems described herein.

[0170] Fig. 4 is a flowchart illustrating a method 400 for acquiring an interpretable trajectory prediction model, according to some embodiments of the present disclosure. The method 400 includes, at block 402, obtaining sensor data. The sensor data may be based on and / or derived from measurements acquired using one or more sensors. One or more of the sensors may be part of, under the control of, and / or in communication with an autonomous or semi-autonomous system (used interchangeably with "system"). Additionally or alternatively, one or more of the sensors may be separate from the system but otherwise capable of making measurements related to the system or the system's environment. The sensor data may be used to determine the presence, position, and / or movement of agents and objects in the system's environment.The method 400 includes, at block 404, determining one or more states (e.g., a location, a speed, and / or a movement of a vehicle in the environment of the system) of an agent in the environment of the system. For example, a sensor may provide a sequence of measurements indicating, based on changes in position, that a vehicle in the vicinity of the system is traveling in a particular direction and / or at a particular speed.

[0171] The method 400 includes, at block 406, executing a trajectory prediction model to generate a trajectory prediction. The generated trajectory prediction may be a prediction of a trajectory of an agent that is close enough to the system to be relevant to the system (e.g., because the agent could potentially collide with the system depending on the agent's behavior). The trajectory prediction model, as further explained herein, may (at block 408) quantify a responsibility of one or more agents in the vicinity of the system. The responsibility may be quantified, for example, based on counterfactual metrics such as safety and civility, as further discussed below. The method 400 may include, at block 410, executing one or more maneuvers and / or one or more changes in one or more maneuvers based on the generated trajectory prediction.For example, the system may change course to safely avoid a collision with the agent based on a prediction that the agent's trajectory will bring the agent too close to the system.

[0172] Fig. 5 illustrates an exemplary interpretability framework for a wide range of phenomenological trajectory prediction methods through responsibility assessments according to some embodiments. In the upper part of Fig. Figure 5 illustrates that responsibility can be formulated by using counterfactual reasoning to assess how much an agent (vehicle 502) prioritizes the well-being of other road users (vehicle 504). This includes assessing the agent's consideration for the safety of other drivers, as indicated by the deviation at which it adheres to safety margins compared to alternative strategies. Furthermore, the extent to which the agent's actions may disrupt the intentions of other drivers is considered, which is indicated by the probability of strategies (numbers next to the trajectories). In the lower panel of Fig. Figure 5 illustrates that the hierarchical framework projects these responsibility metrics onto latent features and integrates these features into a learning-based trajectory prediction model to obtain accurate but interpretable inferences.

[0173] The concept of responsibility is inherent in human action and enforced by legal systems. Therefore, responsibility can be a powerful indicator for an interpretable prediction of the potential movements of neighboring agents. Responsibility can be formalized through counterfactual reasoning. Embodiments of the disclosed approach consider whether an agent could have taken alternative movements to be more considerate toward its neighbors. In various embodiments, a method for determining alternative movements based on a phenomenological traffic model is used to generate realistic and diverse movements. A set of potential movements can be created using predictions from the traffic model.Responsibility metrics can then be calculated by analyzing how important a particular agent considers the welfare of its neighbors by comparing the executed movement with other possible actions from the traffic model.

[0174] Motion Series Synthesis: Embodiments construct a series of open-loop motions using a data-driven traffic model π θ , as it can generate realistic and multimodal behaviors. The traffic model can be trained with real traffic protocols in a supervised learning manner. Given each time step k, embodiments have a context variable xk:=(mk,sk−Th:k(0:N)) where m k is a local semantic map and sk−Th:k(0:N):=(sk−Th:k(0),…,sk−Th:k(N)) the state progression with the horizon T h of the agent (A(i)) and its neighbors (A−i:=A(0,⋯,i−1,i+1,⋯,N)) The output of the exemplary traffic model is a conditional distribution πθ(uk:k+Tf(i)|xk,i) about the future movement with the horizon T f of the agent.

[0175] The distribution over movements enables the generation of dynamically feasible trajectories by determining the dynamics f (i) the agent is integrated into the movement. Then, embodiments can determine the series of movements in the open loop and their corresponding trajectory by sampling movements from the traffic model π θ construct: Uk(i)(xk):={uk:k+Tf(i)∼πθ(⋅|xk,i)}, Ξk(i)(xk):={ξsk(i),u,f(i)|∀u∈Uk(i)(xk)}, where ξ s,u,f :{0,1,···,T f} → ℝ 2xy-positions of the trajectory starting at state s after the dynamics f under the motion u. Note that motion and trajectory are sometimes used interchangeably, since a motion is sampled from the output of the traffic model and the dynamics can be integrated with the motion in an open-loop manner to obtain a trajectory.

[0176] Responsibility formulation: Responsible agents should consider the well-being of their surrounding agents. In particular, they exercise reasonable intentions to avoid collisions and dramatically influencing the behavior of others. Thus, exemplary embodiments formulate responsibility by answering two counterfactual questions: (1) Safety responsibility: Maintains the movement intention of the queried agent A(a) a larger distance margin towards the target agent Away) compared to other movements in the sequence sampled by the traffic model as in (1)? and (2) Politeness responsibility: Would the target agent's movements differ significantly if the sampled agent were not present in the scene? Thus, an AV or other autonomous system can estimate the responsibility of the sampled agent and use this additional information for more accurate and interpretable trajectory prediction, which is described below.

[0177] To answer the first question, embodiments construct a safety result function: Dg(ξ,ξ'):=mink∈{1,2,⋯,Tf}‖ξk−ξ'k‖2, which is the closest distance between two trajectories ξ and ξ' in the L 2 -standard. The implementation examples then calculate the distribution of the decrease in the safety value based on Ξk(a)(xk):C(a,b,xk):={Dg(ξ,ξ(b))−Dg(ξ(a),ξ(b))|ξ∈Ξk(a)(xk)}, where ξ (i) the agent's real trajectory A(i) represented from the traffic protocol.

[0178] Embodiments use conditional value-at-risk (CvaR) to quantify how important the queried agent considers safety when it has a more substantial safety outcome D g CVaR, or expected shortfall, is a risk measure used in portfolio optimization in the financial market. CVaR at level α is the expected cost in the worst α percentile of C under p(C≤c)=|{c'∈C|c'≤c}||c| and formally defined as: CVaRα(C):=∫1−α1VaR1−γ(C)dγα, where VaR α (C) := inf{c ∈ ℝ p(C ≤ c) ≥ 1 - α} is the Value-at-Risk at level α such that the probability of a loss is greater than (at most) α VaR α(C) while the probability of loss is less than VaR α (C) (at least) 1 - α.

[0179] In particular, the security responsibility of the queried agent A(a) with respect to the target agent Away) with scene context x k defined as follows: βkSafety(a,b,xk):=CVaRα(C(a,b,xk))

[0180] In the extreme case, when α → 1, the safety responsibility is the maximum decrease in the safety outcome, i.e. βkSafety(a,b,xk)=maxξ∈Ξk(a)(xk)Dg(ξ,ξ(b))−Dg(ξ(a),ξ(b)).

[0181] The upper left part of the figure in Fig. Figure 5 shows that the executed plan (502) is irresponsible, since there are other control strategies that can maintain a greater margin of safety towards the road user (504) (e.g., a high decrease in safety, illustrated by the numbers corresponding to different trajectories).

[0182] On the other hand, implementation examples model the politeness responsibility as the difference in the target agent's movement distributions with and without the queried agent. The implementation examples calculate the politeness responsibility as follows: βkCourtesy(a,b,xk):=DKL(πθ(⋅|xk,b)‖πθ(⋅|xk−a,b)), where the exponent xk−a the removal of the trajectory history of the queried agent and D KLis the Kullback-Leibler (KL) divergence. This metric quantifies the extent to which the queried agent influences the possible movements of the target agent. The upper right part of Fig. Figure 5 shows that road user 502 is irresponsible because his intention largely influences the distribution of the intentions of road user 504. The probabilities are displayed near the trajectories.

[0183] In implementation examples, the responsibility of the queried agent is defined as the worst responsibility for all its neighbors: βkSecurity(a,xk)=maxA(b)∈A−qβkSecurity(a,b,xk),βkCourtesy(a,xk)=maxA(b)∈A−qβkCourtesy(a,b,xk).

[0184] Interpretable trajectory prediction through responsibility integration: With the two responsibility metrics defined in equations (3) and (4), implementation examples employ a hierarchical framework to render interpretable trajectory predictions. The lower part of Fig. Figure 5 shows the overview of the proposed hierarchical framework according to some embodiments. In the first stage, embodiments show the abstracting sequences of computed responsibility metrics into different responsibility levels. Each responsibility level signifies different interaction intentions, making them valuable for ex-post traffic analysis. In the second stage, embodiments integrate these interpretable latent levels into trajectory prediction models by learning responsibility-dependent reward functions. To simplify the notation, the discussion will focus on a single queried agent and the dependence on A(a), ie βk:=[βksecurity,βkcourtesy]τ, omitted.

[0185] Interpretable Latent Abstraction: In the first stage of the framework, embodiments simplify the sequences of computed responsibility metrics into different levels, assuming that the agents' responsibility can be represented in discrete levels for controllability. This representation can circumvent noisy observations in the traffic log and can be a more natural complement to more complex reasoning and planning. Furthermore, the responsibility metrics may exhibit long-term dependence over time. Therefore, embodiments employ a hidden Markov model (HMM) to treat responsibility metrics as noisy observations and mitigate the long-term dependence with discrete latent responsibility levels. First, embodiments characterize the traffic logs with responsibility observation sequences by (3) and (4).Then, implementation examples learn the parameters of HMM from H-responsibility levels (ρ, A, B),. ρ={ρi}∈ℝH,ρi=p(z0=i) A={aij}∈ℝH×H,aij=p(zk+1=j‖zk=i) B={bh(β)},h∈{0,1,…H−1}, bh(β)=p(βk=β|zk=h) where ρ is the distribution of the initial hidden layer, A is the transition matrix, where a ij is the probability of moving from state i to state j at the next time step, and B is the set of level-conditional emission distributions, where b h (β) is the probability of the observation considering the mode h. After the HMM parameters are learned, the distribution of the hidden responsibility level at time step k can be calculated by the forward algorithm, and the most probable hidden responsibility level can be obtained by Bayes' rule: zk=argmaxz∈{0,1,⋯,H−1}p(βk|z)p(z|β1:k−1)∑z'∈{0,1,⋯,H−1}p(βk|z')p(z'|β1:k−1)=argmaxz∈{0,1,⋯,H−1}p(z|β1:k).

[0186] This distribution and the most probable level can then provide a quantitative measure of how responsible a driver is compared to the collected traffic record.

[0187] Responsible trajectory prediction: In the subsequent phase of the disclosed framework, implementations combine the discrete responsibility levels with a trajectory prediction model to generate interpretable yet accurate predictions. The derived responsibility level serves as a condensed representation of the entire observation history, which can encode longer history information than the context variable, making it a valuable input for trajectory prediction. The noisy Boltzmann (or softmax) rationality model is used in psychology and econometrics to analyze human decisions, where the probability of a choice is exponentially higher if its reward increases relative to other options.Therefore, embodiments assume that for each responsibility level, there is a reward function that assigns higher rewards to behaviors within that level compared to behaviors at other levels. Embodiments use these responsibility-dependent reward functions to fine-tune traffic models, as shown below.

[0188] Implementation examples begin with the construction of a responsible traffic dataset. For each data point (x k , ξ (i) ) in the original data set, implementation examples can increase it by one level of responsibility z k -T f because the responsibility metrics in (3) can be estimated from the ground truth. In addition, embodiments extend the data point with “artificial” trajectories from the traffic model by using the same context variable x kThen the implementation examples group the data set into H-groups {D h | h = 0,1 ···,H - 1} with the HMM learned in the previous subsection.

[0189] For each group D h train examples of a reward function R ϕh , which is parameterized by a neural network to distinguish the real trajectory from the artificial trajectories using inverse reinforcement learning. Specifically, this reward function aims to minimize the following binary cross-entropy loss: L(ϕh):=−∑(ξ,y)∈Dhy log(Rϕh(ξ))+(1−y)log(1−Rϕh(ξ)). where y = 1 if the trajectory is real and y = 0 if the trajectory is artificial.

[0190] After obtaining the responsibility-dependent reward functions, embodiments can retrain the traffic model by supplementing the original reconstruction loss with this additional reward function term. On the other hand, if the traffic model consists of a prediction and planning hierarchy, embodiments can replace the planning cost function with the trained reward function. As discussed below, this simple cost function replacement can significantly improve prediction accuracy without burdensome additional training.

[0191] Simulation Dataset: To illustrate the improvements of embodiments of the disclosed approach, the "nuScenes" dataset (a public autonomous driving dataset supporting the study of urban driving situations), consisting of 1,000 scenes in Singapore and Boston, was used. The scenes are each 20 s long, and the map has 11 semantic layers. Embodiments train the traffic model and the HMM model on 800 scenes from the train split and evaluate 150 scenes from the validation split. For each scene, embodiments create snapshots at 0.1 s intervals. The history of 10 time steps and the future 20 time steps are considered.

[0192] Traffic model: Embodiments build on the BITS architecture, as it enables the generation of accurate and diverse open-loop movements (trajectories) based on goal-conditioning and planning-cost-aware selection. However, the pipeline disclosed here is suitable for any trajectory prediction models that can predict a multimodal distribution across future trajectories, such as transformer-based methods or methods based on conditional variational autoencoders (CVAEs). Here, embodiments focus on vehicle-to-vehicle responsibility and assume that all vehicles share the same single-wheel dynamics. It should be noted that embodiments of the disclosed framework can operate with heterogeneous dynamics and are not limited to vehicle-to-vehicle responsibility, as different embodiments require different dynamics per agent.(i) as in (1). Thus, the output of the traffic model can be a sequence of linear acceleration and heading angular velocity.

[0193] Offline ex-post traffic analysis: Embodiments of the disclosed framework can analyze drivers' responsibility and group their behavior into different levels. For example, if 40 movements were sampled from the traffic model to evaluate safety responsibility, and CVaR was calculated with α = 0.1. Regarding courtesy responsibility, in some embodiments, the KL divergence can be approximated using the trajectories in (1).

[0194] To approach the politeness responsibility, some embodiments first probe possible movements (e.g. 20) of the target agent Away) from the traffic model with and without the requested agent A(a), Ξk(b)'(xk):=Ξk(b)(xk)UΞk(b)(xk−a) In addition, some embodiments may include a reward function R(ξ; Ξ) = -10c coll (ξ; Ξ) - c dev (ξ; Ξ) that measures the cost of a collision and the cost of deviating from the road of the target agent's trajectory ξ among its neighbor's trajectories Ξ. Note that this reward function can be learned from the traffic log. Some embodiments can then compute two sets of the following utility functions: J1={R(ξ; Ξ)|∀ξ∈Ξk(b)'(xk)}, J2={R(ξ; Ξ−a)|∀ξ∈Ξk(b)'(xk)}, where Ξ -a Trajectories of the target agent's neighbor other than the queried agent A(a) designated.

[0195] Some embodiments may use the noisy Boltzmann rationality model to estimate two distributions over possible trajectories Ξk(b)'(xk) with two sets of utility functions J1 or J2 Here, the probability that the target agent selects a trajectory is exponentially proportional to the reward for executing that trajectory, defined as ∀ξ∈Ξk(b)'(xk). P1(ξ)=exp(R(ξ; Ξ))∑ξ'∈Ξk(b)'(xk)exp(R(ξ', Ξ)),P2(ξ)=exp(R(ξ; Ξ−a))∑ξ'∈Ξk(b)'(xk)exp(R(ξ'; Ξ−a)).

[0196] The responsibility of courtesy is βkCourtesy(a, b, xk)≈DKL(P1‖P2) approximated.

[0197] The distribution of raw responsibility metrics Fig. 6. It is observed that most of the data falls into the region near the origin, indicating that most scenarios in the dataset do not exhibit interaction between agents. After obtaining responsibility observation sequences, HMMs can be trained by assuming that the emission distribution is a Gaussian distribution with a diagonal covariance matrix. (bh(⋅)=N(⋅|μh, ∑h)) and is implemented (e.g., with Python's PyHHMM). The different number of levels {3, 5, 7, 9, 11} can be considered, and the model with the lowest Bayesian information criterion (BIC=2NLL +m log n) can be selected, where NLL is the negative log-likelihood of the observed data given the model, m is the number of model parameters, and n is the number of observed data. Fig. Figure 7 shows the BIC under the different number of planes and the plane-dependent emission distributions, where the ellipses are depicted with semiaxes of half standard deviations.

[0198] Fig. Figure 8 provides results of an evaluation of certain embodiments of safety and courtesy responsibility using the nuScenes dataset. The left column shows the movements that the sampled agent could have extracted from the traffic model, as shown in the magenta trajectories. The safety value decrease can be calculated for each sampled trajectory by comparing it with the actual trajectory of the sampled agent and the safety responsibility obtained by calculating the CVaR. The middle and right columns illustrate the distributions of the target agent's movement with and without the sampled agent. The courtesy responsibility can be obtained by calculating the KL divergence between these two distributions. The top row shows that the first agent ("a") could have maintained a larger margin over the target agent ("b") during overtaking.Furthermore, if the first agent had been removed, the trajectory distribution of the first agent would have changed in the first place. These two observations show that the first agent is irresponsible with respect to the target agent (z = 1 by HMM). However, after overtaking, the first agent exhibits more responsible behavior (z = 0 by HMM).

[0199] Responsibility-aware online trajectory prediction: In various embodiments, the latent responsibility levels in the framework can be used to render interpretable yet accurate trajectory predictions. Using inverse reinforcement learning, a reward function can be learned for each responsibility level that best describes driver behavior within that level. A certain number of moves (e.g., 24 moves) can be sampled from the traffic model, and the corresponding trajectories can be combined with the ground-truth trajectory to form a training set. Some embodiments can retrieve four features for each trajectory, such as the cost of a collision (with other agents), the cost of leaving the road, the final displacement (moving forward), and the predicted probability from the traffic model.The reward function can be defined using fully connected neural networks with, for example, two layers of 32 neurons.

[0200] In the prediction phase, some embodiments calculate the responsibility level h for all traffic data, sample a number of movements (e.g., 25 movements) from the traffic model, and select the predicted trajectory with argmax ξ ∈Ξ R ϕh (ξ). On the other hand, BITS can use the heuristic reward function introduced above and select the possible trajectory based on this reward function for all data. Fig. Figure 9 shows the predicted trajectories using the correct level-conditional reward function (z = 4 for top and z = 2 for bottom) compared to the most responsible level (z = 3). Fig. Figure 10 shows responsibility-based trajectory prediction and the heuristic cost-based prediction from BITS. Table 1 shows that the disclosed embodiments outperform BITS in both the average displacement error (ADE) and the final displacement error (FDE) on both the training and validation sets. It should also be noted that the trajectories of z = 3 are close to those of BITS, suggesting that in their implementation, they construct the heuristic reward to match most of the traffic log. However, these embodiments can differentiate responsibility levels and assign different selection criteria.

[0201] TABLE 1: Responsible selection (according to the disclosed approach) outperforms heuristic rule-based costs in both average and final relocation error. Trainierter ADE Trainierter FDE Validierter ADE Validierter FDE Heuristik (BITS) 1,02 m 0,39 m 1,64 m 0,65 m Verantwortung 0,75 m 0,26 m 1,45 m 0,57 m

[0202] Responsibility formulation: In some embodiments, responsibility is based on a series of open-loop moves. Agents can be assumed to make their decisions based on the current state and interaction history, so their controls do not change even after observing the future states of other agents (the predicted future horizon). Responsibility based on closed-loop policies is an extension, as this can capture an approximation of interaction, as in the politeness-responsibility calculation. In some embodiments, incorporating the execution order of agents can lead to different definitions of responsibility (e.g., different sequences of the leader-follower structure).It should also be noted that in alternative embodiments, responsibility concepts can be formulated in a similar pipeline, such as violating traffic rules or respecting the utility of other agents.

[0203] Integration into additional modules in the AV software stack: The disclosed hierarchical framework may include performing ex-post traffic analysis and providing an interpretive approach for trajectory prediction through responsibility-dependent reward functions. In some embodiments, responsibility may be integrated into other modules in the AV software stack. For example, in some embodiments, the disclosed framework may be implemented as (or as part of) a data mining tool to group traffic logs into interactive and naive lane-following scenarios. In other embodiments, the disclosed framework may be incorporated into a traffic simulation to mimic different behaviors using different responsibility-dependent reward functions for each affected agent.Furthermore, some embodiments may treat the continuous responsibility metrics or discrete levels as attribute code and use controllable generation tools to render controls that correspond to the desired responsibility. Furthermore, responsible policies may be helpful to the downstream planner, as they can filter out unrealistic control policies or highlight the significance of each scenario.

[0204] Various embodiments of the interpretability framework and responsibility assessments disclosed herein are applicable to a wide range of trajectory prediction methods. In some embodiments, responsibility is formulated using counterfactual reasoning, splitting responsibility into two components: safety responsibility, which quantifies how an agent could have responded to allow a greater safety margin for its surrounding agents; and courtesy responsibility, which measures how an agent's control strategy affects the intentions of its neighbors. In some embodiments, other metrics may be used instead of and / or in addition to safety and courtesy.A hierarchical framework can be used to abstract interactions into discrete responsibility levels and integrate these low-dimensional and interpretable latent features into a learning-based trajectory prediction model. It has been experimentally demonstrated that responsibility metrics of various embodiments are useful in analyzing real-world traffic logs and that incorporating responsibility into trajectory prediction improves the accuracy of complex traffic data.

[0205] The disclosure of this application also includes the following numbered clauses: Clause 1 Processor, comprising: one or more circuits for: Obtaining sensor data corresponding to one or more movements of a query agent in the vicinity of an autonomous or semi-autonomous system, and Executing a trajectory prediction model to generate a trajectory for at least one of the query agent or the autonomous or semi-autonomous system based at least on the one or more movements, wherein the trajectory prediction model includes a responsibility formulation based on at least one or more counterfactual metrics. Clause 2 Processor according to Clause 1, wherein the responsibility formulation is indicative of a classification for at least one of the query agent or the autonomous or semi-autonomous system. Clause 3 The processor of Clause 1 or Clause 2, wherein the one or more counterfactual metrics include one or more of a security metric or a politeness metric for the query agent. Clause 4 A processor according to any preceding clause, wherein the one or more counterfactual metrics include a security metric for the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. Clause 5 A processor according to Clause 4, wherein the security metric is indicative of one or more security margins supported by the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. Clause 6 A processor according to any preceding clause, wherein the one or more counterfactual metrics include a politeness metric for the querying agent with respect to the autonomous or semi-autonomous system. Clause 7 A processor according to Clause 6, wherein the politeness metric is indicative of an influence on the query agent of at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. Clause 8 A processor according to Clause 6 or Clause 7, wherein the politeness metric is based on at least a Kullback-Leibler (KL) divergence between at least two distributions. Clause 9 A processor according to Clause 8, wherein the at least two distributions correspond to movement of the query agent (1) with the autonomous or semi-autonomous system or with one or more other agents other than the query agent and (2) without the autonomous or semi-autonomous system or without the one or more other agents other than the query agent. Clause 10 The processor of any preceding clause, wherein the one or more counterfactual metrics include a security metric and a politeness metric for the query agent. Clause 11 Processor according to any of the preceding clauses, wherein the responsibility formulation uses a reward function. Clause 12 The processor of any preceding clause, wherein the one or more circuits are further configured to execute a maneuver by the autonomous or semi-autonomous system based at least on the generated trajectory. Clause 13 Processor according to any of the preceding clauses, wherein the processor comprises at least one of the following: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulations; a system for performing collaborative content creation for 3D assets; a system for performing deep learning operations; a system implemented using an edge device; a system implemented using a robot; a system for performing operations using conversational AI; a system that implements one or more language models; a system that implements one or more large language models (LLMs); a system for performing one or more operations using generative AI; a system for generating synthetic data; a system for generating content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system for rendering content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system that combines one or more virtual machines (VMs); a system that is at least partially implemented in a data center; or a system implemented at least in part using cloud computing resources. Clause 14 System comprehensive: one or more processing units for performing operations, comprising: Obtaining sensor data corresponding to one or more movements of a first agent in proximity to a second agent, and Executing a trajectory prediction model to generate a trajectory for at least one of the first agent or the second agent based at least on the one or more movements, wherein the trajectory prediction model includes a responsibility formulation based on at least one or more counterfactual metrics. Clause 15 A system according to Clause 14, wherein the one or more counterfactual metrics include a safety metric for the first agent with respect to at least one of the second agent or one or more agents other than the first agent, the safety metric being indicative of one or more safety margins maintained by the first agent with respect to at least one of the second agent or one or more agents other than the first agent. Clause 16 A system according to clause 14 or clause 15, wherein the one or more counterfactual metrics include a politeness metric for the first agent with respect to the second agent, the politeness metric being indicative of an influence on the first agent of at least one of the second agent and / or one or more other agents apart from the first agent. Clause 17 A system according to any one of clauses 14 to 16, wherein the one or more counterfactual metrics include a safety metric and a politeness metric for the first agent. Clause 18 A system according to any one of clauses 14 to 17, wherein the system is an autonomous or semi-autonomous vehicle, the vehicle comprising one or more sensors from which the sensor data or a portion thereof is received, and the vehicle further comprising a control system configured to execute a second trajectory for the vehicle based on the generated trajectory. Clause 19 A system as defined in any of Clauses 14 to 18, which system includes at least one of the following: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulations; a system for performing collaborative content creation for 3D assets; a system for performing deep learning operations; a system implemented using an edge device; a system implemented using a robot; a system for performing operations using conversational AI; a system that implements one or more language models; a system that implements one or more large language models (LLMs); a system for performing one or more operations using generative AI; a system for generating synthetic data; a system for generating content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system for rendering content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system that combines one or more virtual machines (VMs); a system that is at least partially implemented in a data center; or a system implemented at least in part using cloud computing resources. Clause 20 Procedure, comprising: Obtaining sensor data corresponding to one or more movements of a query agent in the vicinity of a dynamic system, and Executing a trajectory prediction model to generate a trajectory for at least one of the query agent or the dynamic system based at least on the one or more movements, wherein the trajectory prediction model comprises a responsibility formulation based on at least a safety metric and a politeness metric.

[0206] The disclosure may be described in the general context of computer code or machine-usable instructions, including computer-executable instructions, such as program modules, executed by a computer or other machine, such as a personal data assistant or other handheld device. In general, program modules, which include routines, programs, objects, components, data structures, etc., refer to code that performs specific tasks or implements specific abstract data types. The disclosure may be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialized computing devices, etc.The disclosure may also be practiced in distributed computing environments in which tasks are performed by remote processing devices that are interconnected via a network for communication.

[0207] As used herein, any reference to "and / or" in reference to two or more elements should be interpreted to mean only one element or a combination of elements. For example, "Element A, Element B, and / or Element C" may include only Element A, only Element B, only Element C, Element A and Element B, Element A and Element C, Element B and Element C, or Elements A, B, and C. Furthermore, "at least one of Element A or Element B" may include at least one of Element A, at least one of Element B, or at least one of Element A and at least one of Element B. Further, "at least one of Element A and Element B" may include at least one of Element A, at least one of Element B, or at least one of Element A and at least one of Element B.

[0208] The subject matter of the present disclosure is specifically described herein to satisfy legal requirements. However, the description itself is not intended to limit the scope of the present disclosure. Rather, the inventors contemplated that the claimed subject matter may be embodied in other ways to include various steps or combinations of steps similar to those described herein, in conjunction with other present or future technologies. Although the terms "step" and / or "block" may be used herein to refer to various elements of the methods employed, the terms should not be interpreted to imply any particular ordering among or between the various steps disclosed herein, unless and except that the order of the individual steps is expressly described.

[0209] It is to be understood that aspects and embodiments described above are purely exemplary and that modifications of details may be made within the scope of the claims.

[0210] Each device, method, and feature disclosed in the description, and (where appropriate) the claims and drawings may be provided independently or in any suitable combination.

[0211] Reference signs appearing in the claims are for illustrative purposes only and do not limit the scope of the claims. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] US 16 / 101,232

[0074] Cited non-patent literature

[0000] Society of Automotive Engineers (SAE) (Standard No. J3016-201806, published on June 15, 2018, Standard No. J3016-201609, published on September 30, 2016

[0032]

Claims

[1] Processor comprising: one or more circuits for: Obtaining sensor data corresponding to one or more movements of a query agent in the vicinity of an autonomous or semi-autonomous system, and Executing a trajectory prediction model to generate a trajectory for at least one of the query agent or the autonomous or semi-autonomous system based at least on the one or more movements, wherein the trajectory prediction model includes a responsibility formulation based on at least one or more counterfactual metrics. [2] The processor of claim 1, wherein the responsibility formulation is indicative of a classification for at least one of the query agent or the autonomous or semi-autonomous system. [3] The processor of claim 1 or claim 2, wherein the one or more counterfactual metrics include one or more of a security metric or a politeness metric for the query agent. [4] A processor according to any preceding claim, wherein the one or more counterfactual metrics include a security metric for the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. [5] The processor of claim 4, wherein the security metric is indicative of one or more security margins supported by the query agent with respect to at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. [6] A processor according to any preceding claim, wherein the one or more counterfactual metrics include a politeness metric for the query agent with respect to the autonomous or semi-autonomous system. [7] The processor of claim 6, wherein the politeness metric is indicative of an influence on the query agent of at least one of the autonomous or semi-autonomous system or one or more other agents other than the query agent. [8] The processor of claim 6 or claim 7, wherein the politeness metric is based on at least a Kullback-Leibler (KL) divergence between at least two distributions. [9] The processor of claim 8, wherein the at least two distributions correspond to movement of the query agent (1) with the autonomous or semi-autonomous system or with one or more other agents apart from the query agent and (2) without the autonomous or semi-autonomous system or without the one or more other agents apart from the query agent. [10] A processor according to any preceding claim, wherein the one or more counterfactual metrics include a security metric and a politeness metric for the query agent. [11] A processor according to any preceding claim, wherein the responsibility formulation uses a reward function. [12] The processor of any preceding claim, wherein the one or more circuits are further configured to execute a maneuver by the autonomous or semi-autonomous system based at least on the generated trajectory. [13] Processor according to one of the preceding claims, wherein the processor comprises at least one of the following: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulations; a system for performing collaborative content creation for 3D assets; a system for performing deep learning operations; a system implemented using an edge device; a system implemented using a robot; a system for performing operations using conversational AI; a system that implements one or more language models; a system that implements one or more large language models (LLMs); a system for performing one or more operations using generative AI; a system for generating synthetic data; a system for generating content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system for rendering content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system that combines one or more virtual machines (VMs); a system that is at least partially implemented in a data center; or a system implemented at least in part using cloud computing resources. [14] System comprising: one or more processing units for performing operations, comprising: Obtaining sensor data corresponding to one or more movements of a first agent in proximity to a second agent, and Executing a trajectory prediction model to generate a trajectory for at least one of the first agent or the second agent based at least on the one or more movements, wherein the trajectory prediction model includes a responsibility formulation based on at least one or more counterfactual metrics. [15] The system of claim 14, wherein the one or more counterfactual metrics include a safety metric for the first agent with respect to at least one of the second agent or one or more agents other than the first agent, the safety metric being indicative of one or more safety margins maintained by the first agent with respect to at least one of the second agent or one or more agents other than the first agent. [16] The system of claim 14 or claim 15, wherein the one or more counterfactual metrics include a politeness metric for the first agent with respect to the second agent, the politeness metric being indicative of an influence on the first agent of at least one of the second agent and / or one or more other agents apart from the first agent. [17] The system of any of claims 14 to 16, wherein the one or more counterfactual metrics include a safety metric and a politeness metric for the first agent. [18] The system of any one of claims 14 to 17, wherein the system is an autonomous or semi-autonomous vehicle, the vehicle comprising one or more sensors from which the sensor data or a portion thereof is received, and the vehicle further comprising a control system configured to execute a second trajectory for the vehicle based on the generated trajectory. [19] A system according to any one of claims 14 to 18, wherein the system comprises at least one of the following: a control system for an autonomous or semi-autonomous machine; a perception system for an autonomous or semi-autonomous machine; a system for performing simulation operations; a system for performing digital twin operations; a system for performing light transport simulations; a system for performing collaborative content creation for 3D assets; a system for performing deep learning operations; a system implemented using an edge device; a system implemented using a robot; a system for performing operations using conversational AI; a system that implements one or more language models; a system that implements one or more large language models (LLMs); a system for performing one or more operations using generative AI; a system for generating synthetic data; a system for generating content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system for rendering content for a virtual reality (VR), an augmented reality (AR) or a mixed reality (MR) system; a system that combines one or more virtual machines (VMs); a system that is at least partially implemented in a data center; or a system implemented at least in part using cloud computing resources. [20] Method comprising: Obtaining sensor data corresponding to one or more movements of a query agent in the vicinity of a dynamic system, and Executing a trajectory prediction model to generate a trajectory for at least one of the query agent or the dynamic system based at least on the one or more movements, wherein the trajectory prediction model comprises a responsibility formulation based on at least a safety metric and a politeness metric.

Images