Belief Propagation for Range Image Mapping in Autonomous Machine Applications

Abstract

The present disclosure relates to belief propagation for range image mapping in autonomous machine applications. In various examples, systems and methods are described for generating scene flow in 3D space by reducing 3D lidar data to a "2.5D" optical flow space (e.g., x, y, and depth flow). For example, a lidar range image can be used to generate a 2.5D representation of depth flow information between frames of lidar data, and two or more range images can be compared to generate the depth flow information, and messages can be passed (e.g., using a belief propagation algorithm) to update pixel values in the 2.5D representation. The resulting image can then be used to generate 2.5D motion vectors, and the 2.5D motion vectors can be converted back to 3D space to generate a 3D scene flow representation of the autonomous machine's surroundings.

Description

Background Art

[0001] To operate safely and efficiently, autonomous and semi-autonomous vehicles need to identify moving obstacles or objects at varying distances so they can take appropriate action in a timely manner. For example, appropriate actions may include turning, changing lanes, braking, and / or accelerating to avoid or move away from an identified obstacle. Furthermore, accurately identifying obstacles and their movement early on allows time to determine and execute appropriate actions. Therefore, object detection and tracking must be performed in real time or near real time, even when the vehicle is traveling at highway speeds or in the presence of cross traffic.

[0002] The representation of static and / or moving objects in three-dimensional (3D) space can be referred to as scene flow. Existing scene flow systems are often designed for static and / or limited fields of view. Due to the complexity and computational requirements of performing scene flow in 3D space, existing scene flow systems may not operate in real time, or may only operate in real time at limited resolution and / or accuracy. Consequently, due to these limitations, existing autonomous vehicles rely on alternative obstacle detection methods and data sources, such as camera-based neural networks, lidar point clouds, and / or radar data. However, while camera-based neural networks are accurate at short distances, they are less accurate at greater distances due to the limited number of pixels representing any given object or obstacle. Furthermore, lidar and radar sensors often include missing data points and may be noisy. These existing methods often rely on analyzing a single frame of data, which may lack important context that can only be obtained by analyzing multiple frames across time. Summary of the Invention

[0003] Embodiments of the present disclosure relate to belief propagation for range image mapping in autonomous machine applications. The present disclosure relates to systems and methods for providing lidar-based scene flow in real time, at least in part, by focusing on time-varying motion rather than single-frame detections. Compared to conventional systems as described above, the current systems and methods generate scene flow in 3D space by simplifying 3D lidar data into an enhanced two-dimensional ("2.5D") depth flow space (e.g., x, y, and depth flows - indicating changes in depth values between lidar range images), passing messages between pixels (e.g., between nodes of a pixel representation, such as a matrix, grid, table, etc., corresponding to the pixel) - via, for example, a belief propagation algorithm - by converting the 2.5D information back into 3D space (e.g., based on a known association between a 2.5D image space position and a 3D world space position), and computing a 3D motion vector for the pixel. Thus, message passing can be used (e.g., in 2D or 2.5D space) to estimate noise or missing information from the lidar data to generate a denser representation of the scene, which can then be converted back into 3D space for use in detecting, identifying, and / or tracking dynamic objects in the environment. The current system can be configured for use on a mobile platform, such as an autonomous machine (e.g., a vehicle, a robot, etc.), and can include a rotating lidar sensor or other depth sensor that provides a wider field of view (e.g., up to 360 degrees) than existing systems. In addition, because most of the computation is performed on a projected image (e.g., a lidar range image) rather than in three-dimensional space, the current system is also able to operate in real time to identify moving obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] The following describes in detail the disclosed systems and methods for belief propagation of range image mapping in autonomous machine applications with reference to the accompanying drawings, wherein:

[0005] Figure 1A is a data flow diagram of a scene flow estimation system according to some embodiments of the present disclosure;

[0006] Figure 1B is a time flow diagram of a scene flow generation process according to some embodiments of the present disclosure;

[0007] Figure 2 is a schematic diagram illustrating labels on example pixel volumes of image displacement and depth streams according to some embodiments of the present disclosure;

[0008] Figure 3 is a diagram illustrating an example cost function for message passing with respect to a depth stream according to some embodiments of the present disclosure;

[0009] Figure 4is a schematic diagram illustrating message transmission between a pixel node and adjacent pixel nodes according to some embodiments of the present disclosure;

[0010] Figure 5 is a schematic diagram illustrating two consecutive lidar range images and a composite depth stream image according to some embodiments of the present disclosure;

[0011] Figure 6A and 6B is a schematic diagram illustrating a synthetic scene flow generated from a depth flow image according to some embodiments of the present disclosure;

[0012] Figure 7 is a flowchart illustrating a method for scene flow generation using belief propagation according to some embodiments of the present disclosure;

[0013] Figure 8A is an illustration of an example autonomous vehicle according to some embodiments of the present disclosure;

[0014] Figure 8B According to some embodiments of the present disclosure Figure 8A Examples of camera positions and fields of view for autonomous vehicles;

[0015] Figure 8C According to some embodiments of the present disclosure Figure 8A a block diagram of an example system architecture for an example autonomous vehicle;

[0016] Figure 8D According to some embodiments of the present disclosure, a method for Figure 8A System diagram of an example of communication between autonomous vehicles;

[0017] Figure 9 is a block diagram of an example computing device suitable for implementing some embodiments of the present disclosure; and

[0018] Figure 10 is a block diagram of an example data center suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0019] The present disclosure relates to systems and methods related to belief propagation for range image mapping in autonomous machine applications. Although the present disclosure may be made with respect to an exemplary autonomous vehicle 800 (alternatively referred to herein as "vehicle 800" or "ego vehicle 800"), examples of which may be found in connection with FIG. Figures 8A-8D), but this is not intended to limit the present disclosure. For example, the systems and methods described herein may be used by, but are not limited to, non-autonomous vehicles, semi-autonomous vehicles (e.g., in one or more adaptive driver assistance systems (ADAS)), manned and unmanned robots or robotic platforms, warehouse vehicles, off-road vehicles, vehicles connected to one or more trailers, flying ships, boats, shuttles, fire trucks, motorcycles, electric or motorized bicycles, aircraft, construction vehicles, underwater vehicles, drones, and / or other vehicle or autonomous machine types. Furthermore, while the present invention may be described as generating scene flows for autonomous driving, this is not intended to be limiting, and the systems and methods described herein may be used for augmented reality, virtual reality, mixed reality, robotics, security and surveillance, autonomous or semi-autonomous machine applications, and / or any other technology space where scene flows may be used.

[0020] Embodiments of the present disclosure relate to determining scene flow using a depth sensor (such as a lidar sensor used in an autonomous vehicle) by creating and analyzing a range image generated from a lidar range (or other projection) image. For example, a lidar sensor may generate three-dimensional lidar data representing its field of view or sensory field, for example, the lidar data may represent a standard spherical projection or other projection type. The lidar sensor may generate lidar data continuously, thereby generating multiple sets of lidar data - each indicating a single scan or rotation of the lidar sensor. These sets of lidar data may be separated by a time interval - for example, corresponding to the time interval required to complete a lap - and embodiments of the present invention may use a first set (e.g., current) of lidar data and compare it to a second, previous set (e.g., previous set) of lidar data to determine various information from the data, as discussed herein. In some embodiments, two consecutive sets of lidar data may be analyzed; however, this is not intended to limit the present disclosure, and the analyzed sets of lidar data may correspond to different data arrangements other than consecutive frames.

[0021] Complex 3D lidar data (e.g., representing a 3D point cloud) can be simplified into a 2D lidar range image (or other projected image type) and can also be encoded with additional information (e.g., depth flow) to generate a 2.5D depth flow image (or referred to herein as a "depth flow image"), allowing for faster analysis. In some embodiments, a pyramid or multi-scale approach can also be used to reduce the resolution of the depth flow image to help identify optical flow information more quickly, especially when there is significant motion between frames. A depth flow image can include a set of pixels, each pixel having a set of two-dimensional (2D) coordinates and a depth flow value (indicating the change in depth or distance across frames). In some embodiments, the depth flow can be simplified into a set of labels (e.g., where a label-based algorithm such as belief propagation is employed) indicating motion toward the sensor, motion away from the sensor, and / or no motion toward or away (e.g., stationary). For example, these depth flow values can be represented as -1 (towards), +1 (away), and 0 (stationary), respectively. In some embodiments, the depth flow can be more fine-grained and can represent fast motion toward the sensor, slow motion toward the sensor, no motion toward the sensor, slow motion away from the sensor, and / or fast motion away from the sensor. This simplified depth flow information can be used for the analysis discussed herein.

[0022] When comparing two (e.g., consecutive) range images, one range image (e.g., the previous image) can be transformed or rectified to the coordinate system of the subsequent or current range image (or vice versa) to compensate for ego motion. By compensating for ego motion, subsequent analysis only needs to consider the motion of the object relative to the ego vehicle because the motion of the ego vehicle has been factored out, allowing the two range images to be aligned in the same coordinate system.

[0023] In some embodiments, two or more range images are compared to determine a three-dimensional motion vector associated with each pixel therein. The two or more depth images are analyzed using a message passing belief propagation algorithm to generate a three-dimensional motion vector. For example, the belief propagation algorithm can be used to propagate pixel data to neighboring pixels (e.g., pixel nodes represented in a matrix, table, grid, etc.), and the data received at the neighboring pixels (e.g., neighboring nodes represented by the pixel) is analyzed and used to determine the updated value of the pixel. Additional iterations are then performed to optimize the result.

[0024] Generally, belief propagation is a message passing algorithm used to infer information from a graphical model. While belief propagation can be exact in some cases, it is often used for approximations. The belief propagation algorithm may require defining two constraints: the first is to determine a data term; the second is to determine a smoothness term. The data term can be used to identify similar or identical pixels on two range images so that the movement of objects represented by the pixels can be determined. The smoothness term can be used to adjust the values between adjacent pixels so that identified objects move together, or to treat boundaries between objects or surfaces as such. Messages can be sent between pixels of the same depth stream image (e.g., pixel nodes corresponding to pixels) and carry information about the data term and the smoothness term. Specifically, the messages can include depth stream information (to help identify the general direction of entry and exit to identify the correct pixel between images), 2D pixel flow information (pixel movement between frames), and / or cost (to help identify the extent to which a pixel affects its neighboring pixels).

[0025] The data item may identify which pixels in two or more range images correspond to the same object (or portion of the same object). In embodiments, the data item may be useful because the physical object and / or the depth sensor moves between images. Variables such as depth flow, reflectivity, color values, intensity, time of flight (ToF), texture, return behavior, and / or other information (such as that represented by lidar data or in a lidar range image) may be analyzed in the data item.

[0026] Messages cannot be passed between pixels (e.g., pixel nodes) that correspond to different objects. For example, adjacent pixels that are calculated not to correspond to the same physical object in the physical environment (e.g., because the depth difference or reflectivity between the two pixels detected exceeds a certain threshold) will not pass messages between each other. This is because these pixels should not affect their neighboring pixels because they correspond to different physical objects. Therefore, the differences between these neighboring pixels may not be overly affected by the belief propagation algorithm.

[0027] The smoothness term can determine which value of the depth stream is approximately correct for pixels corresponding to the same physical object. Typically, physical objects moving in space move as a single unit. Therefore, the incomplete and imprecise data of some pixels and the influence of neighboring pixels can be adjusted to account for this information. For example, a cost is assigned to each pixel based on the reliability of the data associated with it. For pixels with incomplete data (for example, not returned to the lidar sensor), a low cost can be assigned, making them more likely to be influenced by their neighboring pixels through a belief propagation algorithm. This can produce a dense motion field despite the lack of original lidar scan data.

[0028] By passing and analyzing these messages between pixels of the same depth stream image, the accuracy of various data at each pixel can be improved to an optimized 2D (or 2.5D) motion vector. Missing or incomplete data for other pixels can also be estimated and optimized.

[0029] The 2D (or 2.5D) vectors can then be converted back to 3D space using the known correspondence between the 2D image space and the 3D world space position. As a result, a 3D motion vector for a point in the lidar point cloud can be determined, where the 3D motion vector represents the difference (or movement) of the point, and therefore the difference in the object or obstacle, between the lidar data frames. The message passing belief propagation algorithm therefore allows this complex analysis to be simplified and completed before returning to 3D space, thereby reducing the computational requirements and runtime of the system by avoiding analysis in a 3D coordinate system. Ultimately, the scene flow shows the motion of various physical objects around the depth sensor and can be used to detect and track dynamic objects around an autonomous or semi-autonomous machine so that the machine can take various appropriate actions to account for the objects.

[0030] refer to Figure 1A , Figure 1A is an example scene flow generation system 100 according to some embodiments of the present disclosure. It should be understood that this arrangement and other arrangements described herein are set forth as examples only. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, functional groupings, etc.) may be used in addition to or in place of the arrangements and elements shown, and some elements may be omitted entirely. In addition, many of the elements described herein are functional entities that may be implemented as discrete or distributed components, or in combination with other components, and may be implemented in any suitable combination and location. The various functions performed by the entities described herein may be performed by hardware, firmware, and / or software. For example, instructions stored in a memory may be executed by a processor to perform the various functions. In some embodiments, the system 100 may include the same as described herein. Figures 8A-8D Similar components, features and / or functions of the related vehicle 800. Figure 9 The example computing device 900 of Figure 10 Example data center 1000.

[0031] like Figure 1AAs shown, scene flow system 100 may include one or more depth perception sensors 102 (e.g., a lidar sensor, a radar sensor, an ultrasonic sensor, etc.). Depth perception sensor 102 may generally include a transmitter and a receiver, and may include any suitable field of view or sensing field, such as a wide field of view (e.g., 180 to 360 degrees), and in embodiments may be movable (e.g., rotated) to obtain a view of a larger area relative to depth perception sensor 102. For example, lidar signals may reflect off objects near depth perception sensor 102. Objects may move relative to depth perception sensor 102, and depth perception sensor 102 may move relative to an underlying surface (e.g., a road on which a vehicle is traveling). A receiver may receive (directly or indirectly) an indication of these different reflected signals, and this indication may be stored and / or transmitted as data for later analysis.

[0032] The depth perception sensor 102 may have a sensor controller 104 that may be used to control the operation of the depth perception sensor 102 and interpret the results. For example, the sensor controller 104 or other processor may receive sensor data 106 and process, analyze, or otherwise perform calculations related to the sensor data 106. In some embodiments, the depth perception sensor 102 and / or the sensor controller 104 may be similar to the one described with respect to FIG. Figures 8A-8C The lidar sensor 864 is depicted, or may be another type of depth perception sensor 102 .

[0033] The sensor controller 104 may output the sensor data 106 to a computing system 108, such as in the vehicle 800 and / or Figure 9 The computing system executed within the example computing device 900 of FIG. The sensor data 106 may be in any of various forms, such as, but not limited to, a 3D lidar point cloud. The sensor data 106 may be analyzed to perform various functions related thereto and may be used in conjunction with other sensor data 106 (e.g., Figures 8A-8C In embodiments using a lidar sensor, sensor data 106 may be referred to as lidar data; however, in other embodiments of the present disclosure, sensor data 106 may be another type of depth data (e.g., from radar, ultrasonic, etc.).

[0034] The system 100 may include a coordinate converter 110 that may convert the sensor data 106 into different formats, reference frames (e.g., from 3D to 2D or 2.5D, etc.), and / or types (e.g., from a point cloud to a projected or range image). As a non-limiting example, the coordinate converter 110 may convert lidar data representing a 3D point cloud into a 2D range image or other projected image type, and the depth stream generator may analyze the depth information between frames to generate a "2.5D" depth stream image using the depth stream image generator 112. In some embodiments, the coordinate converter 110 may convert one or more range images into the same coordinate system (based on the known or tracked motion of the vehicle 800 (e.g., ego-motion)), and after analysis of optical flow, message passing, etc. in the 2.5D space, the coordinate converter 110 may convert the resulting 2.5D depth stream image into a three-dimensional scene stream representation.

[0035] The system 100 may also include a depth stream generator 116 that may create one or more depth stream images using one or more sets of sensor data 106. Figure 5 As shown, one or more lidar sensors can be used to generate a range image from lidar data. Figure 5 , a first set of lidar data 502 captured at time T1 and a second set of lidar data 504 captured at time T2 are shown. For example, the lidar data sets may represent spherical projections or other types of projections extending from the depth perception sensor 102. The lidar sensor may generate lidar data continuously, thereby generating multiple sets of lidar data. Each such set may indicate a single scan, rotation, or other capture of the lidar sensor, and the lidar data sets may be separated by time intervals. In some embodiments, to generate an optical or depth stream image, the depth stream image generator 112 may use the second set (e.g., current or most recent) lidar data 504 and compare it to a previous set of lidar data 502 to determine various information from the data, as discussed herein. In some embodiments, two consecutive sets of lidar data may be analyzed; however, this is not intended to limit the present disclosure, and the analyzed lidar data sets may correspond to different arrangements other than consecutive frames.

[0036] In some embodiments, the optical or depth flow image can be a simplified "2.5D" rendering of the 3D information, where pixels (e.g., each pixel) have coordinates (e.g., (x, y) coordinates, representing azimuth and elevation, respectively) and one or more associated variables, such as depth and / or depth flow. By reducing the 3D representation to a 2D or 2.5D representation, computation and runtime can be reduced.

[0037] In some embodiments, the depth stream image generator 112 can generate a first range image based at least in part on first lidar data generated at time T1 using one or more lidar sensors. The first range image can be generated so that pixels of the image include depth stream values—for example, indicating changes in depth values across frames. Similarly, the depth stream image generator 112 can generate a second range image based at least in part on second lidar data generated at time T2, subsequently to time T1, using one or more lidar sensors. The first range image and the second range image can then be stored, analyzed, and / or compared to each other (as well as to the third depth stream image, the fourth depth stream image, etc.), as described herein.

[0038] In embodiments of the present disclosure, a range image may be generated using a lidar range image generated using data representing one or more lidar point clouds. As described herein, a lidar point cloud may correspond to raw data output by a lidar sensor (and / or lidar controller). Lidar data may represent any combination of various variables associated with transmitted and received signals, such as, but not limited to, reflectivity information, texture information, time of flight (ToF) information, color information, and / or intensity information.

[0039] Complex 3D lidar data (e.g., representing a 3D point cloud) can be simplified into a 2D lidar range image (or other projected image type) or encoded with additional information (e.g., a depth stream) to generate a 2.5D depth stream image that allows for rapid analysis. In some embodiments, a pyramid or multi-scale approach can also be used to reduce the resolution of the depth stream image to help identify optical flow information more quickly, especially in situations where there is significant motion between frames (e.g., significant motion of the depth perception sensor 102 on an autonomous vehicle operating at highway speeds). A depth stream image can include a set of pixels, each pixel having a set of 2D coordinates and a depth stream. In embodiments, the depth stream can be stored as a label on the pixel.

[0040] The simplified depth flow representation can be used to process scene flow instead of more complex 3D lidar point cloud data. The simplified calculations can allow the analysis discussed in this article to be performed in real time, and the results can be projected back into 3D space to generate accurate 3D scene flow information with less computation and processing time.

[0041] The system 100 may include a belief propagator 114 that may identify or assign pixels having one or more attributes, such as Figure 2 For example, Figure 3As shown, the belief propagator 114 can pass messages between pixels of the depth stream image (e.g., pixel nodes of a pixel representation, such as a matrix, graph, table, etc.) to refine and analyze the information therein. In some embodiments, a cost can be assigned to the pixel, such as Figure 4 shown.

[0042] Thus, embodiments of the present invention may use a belief propagation algorithm to pass messages between adjacent pixels (e.g., pixel nodes) of a depth stream image. Generally, belief propagation may include a message passing algorithm for inferring information from a graphical model. While belief propagation can be exact in some cases, it is typically used for approximation. For example, the exact information of lidar data is simplified to a depth stream image, message passing approximates the values of adjacent pixels, and the approximation is returned to the coordinate space of the generated lidar data. The simplification, message passing, and return are performed faster than performing a full analysis in full 3D while allowing for sufficient certainty in obstacle detection, identification, and / or tracking.

[0043] To help the belief propagation algorithm pass messages between adjacent pixels (e.g., pixel nodes of pixel representations), labels can be associated with pixels that include information related to the adjacent pixels. It should be understood that "pixel" as used herein can refer to any subdivision of the depth flow image and can correspond to a node (e.g., a matrix, table, graph, etc.) that represents a pixel or its subdivision in a representation of the depth flow image. The resolution of individual pixels can be reduced to allow for faster processing. Similarly, groups of pixels can be grouped together - for example, in an embodiment, messages can be passed between adjacent groups of pixels rather than between pixels.

[0044] refer to Figure 2 , Figure 2 Example labels that can be associated with pixels are shown. For example, a belief propagation label can correspond to a 2.5D rectangular block. The labels shown can represent image displacement and depth flow, where the label volume can correspond to movement in the Z direction (e.g., toward and away from the depth perception sensor 102).

[0045] In some embodiments, the depth flow can be simplified to a set of labels indicating motion toward the sensor, motion away from the sensor, and no motion toward or away from the sensor (e.g., stationary). For example, these depth flow values ( Figure 2In some embodiments, the depth flow may be represented by a-1, +1, and 0, respectively. In some non-limiting embodiments, these simplified data flow labels may be referred to as three-value labels using balanced three-value logic. In some embodiments, the depth flow may be a set of simplified labels indicating fast motion toward the sensor, slow motion toward the sensor, no motion relative to the sensor, slow motion away from the sensor, and fast motion away from the sensor. For example, the depth flow values may be represented by a-2, -1, 0, +1, and +2, respectively (not shown). However, these labels are for example purposes only, and in other embodiments, different label types may be used to represent depth flow or depth changes between frames.

[0046] refer to Figure 3 , Figure 3 shows another type of label that can be associated with a pixel. For example, Figure 3 An example diagram depicting the calculation of a cost function for a depth stream data item is shown. Figure 3 The x-axis of the graph in corresponds to the depth difference detected for that pixel, while Figure 3 The y-axis of the graph in corresponds to the associated cost. As a first example, if the depth difference is 0.0 and the depth flow value is 0, a cost of 0 will be assigned, which means that if the corresponding physical object has not moved and the two adjacent pixels are at the same distance, the cost will be very low. As a second example, if the depth difference is 0.125 (in Figure 3 ), the lowest cost depth stream label will be associated with Z = 1, the second lowest cost will be associated with Z = 0, and the highest cost will be associated with Z = -1. The cost will be selected as the y value at the appropriate depth stream value for that pixel and the corresponding depth stream for that pixel. It should be understood that the y values shown on the graph and the slope of the corresponding plot are for illustrative purposes only, and the actual cost value and depth label function may vary depending on any number of factors.

[0047] Once assigned labels, pixels can pass messages between them via a belief propagation algorithm, such as Figure 4As shown - for example, a node corresponding to a pixel can pass a message to a neighboring node corresponding to an adjacent or neighboring pixel. For example, the belief propagator 114 can pass one or more messages between pixels of a second set of pixels that at least partially indicate the respective depth flow values of the pixels and the respective costs of the pixels. In some embodiments, one or more messages are passed using a belief propagation algorithm. Typically, a message that at least partially indicates the label discussed herein can be sent to adjacent pixels. The received message can then be analyzed, specifically, the image displacement and depth flow with the associated cost can be compared with the corresponding value of the pixel. Then, a new message with the optimized value can be sent and analyzed to further optimize the value of the label. After a certain threshold, the iteration of messages and analysis may stop. In some examples, the threshold can be a specific number of iterations, a specific time interval, reaching a specific threshold confidence level, a threshold associated with the processor and / or memory capacity, and / or another threshold.

[0048] The use of the belief propagation algorithm can satisfy various constraints. For example, the belief propagation algorithm can identify which adjacent pixels correspond to the same physical object (e.g., data terms), and the belief propagation algorithm can "smooth" adjacent values so that adjacent pixels corresponding to the same physical object initially have similar resulting values (e.g., smoothness terms). Therefore, the belief propagation algorithm can determine which adjacent pixels correspond to the same physical object through the data terms, and can determine an approximation of the pixels associated with the physical object through the smoothness terms. By iterating the message passing and analysis multiple times on the same optical or depth stream image, physical objects can be determined, and an approximation of their motion can be determined. By iterating multiple times on multiple consecutive depth stream images, new physical objects can be identified (e.g., objects that were previously occluded or came into range of the depth perception sensor 102), and changes in previously identified physical objects (e.g., changes in direction or speed) can be determined and tracked.

[0049] In some embodiments, messages can be sent between pixels of the same depth stream image (e.g., nodes corresponding to pixels) and carry information about a data item and a smoothness item. In some embodiments, the data item can be determined during a first set of iterations, and the smoothness item can be determined during subsequent iterations of the same depth stream image. In other embodiments, the data item and the smoothness item can be determined simultaneously. Specifically, the message can include depth stream information (to help identify the general direction of ingress and egress to identify the correct pixel between images) and / or cost (to help identify the degree of influence of a pixel on its neighboring pixels).

[0050] Figure 4 An example of message passing is shown. Figure 4An example target pixel "X" is shown in the center of FIG. The target pixel is adjacent to the neighboring pixels in the horizontal and vertical directions. In some embodiments, the target pixel may also be adjacent to the neighboring pixels in the horizontal direction or other designated direction (not shown). Figure 4 In the , the vertical neighboring pixels are labeled as v+ (vertically above the target pixel) and v- (vertically below the target pixel). The horizontal neighboring pixels are labeled as h+ (horizontally to the right of the target pixel) and h- (horizontally to the left of the target pixel). It should be understood that any pixel in the depth stream image can be designated as a target pixel together with the corresponding neighboring pixels. For example, Figure 4 The pixel h+ in can be designated as the target pixel, which will have the horizontal neighboring pixel as pixel “x” and another pixel further to the right ( Figure 4 ). Furthermore, although referred to as pixels, message passing can be between nodes or other representations of pixels in a graph, matrix, table, or other representation of an image.

[0051] Messages can be passed between pixels. For example, a target pixel can send messages to its neighboring pixels, and each message can include at least one tag or a portion thereof. In embodiments, a message can correspond to a simple packet of information associated with one or more tags about each pixel. By sharing information with neighboring pixels, various aspects of data can be determined, as discussed herein.

[0052] like Figure 4 As shown, outgoing messages can be labeled with lowercase lambda and the subscript of the pixel from which the outgoing message comes, with the recipient of the message added in parentheses (this convention is only used for example purposes). For example, an outgoing message from a target pixel x to its vertically adjacent pixel can be designated as λ x (v+). Therefore, the destination pixel can create and send four outgoing messages: x (v+), λ x (v-), λ x (h+) and λ x (h-). As discussed herein, in embodiments of the present invention, other messages may also be sent (e.g., messages to diagonally adjacent pixels). The outgoing message may provide information about the current label or labels associated with the target pixel. In some examples, the outgoing messages from the target pixel (e.g., the node corresponding to the target pixel) to each adjacent pixel (e.g., the nodes corresponding to adjacent pixels) may be similar or identical because the one or more labels associated with the target pixel are the same. In other embodiments, the content of at least one outgoing message may be selected based on one or more features of the adjacent pixels.

[0053] exist Figure 4In the , incoming messages are labeled with lowercase pi and the subscript of the pixel to which the incoming message is to go, with the sender of the message added in parentheses (this convention is only used for example purposes). For example, an incoming message from the horizontal right neighbor of the object pixel x to the target pixel x can be designated as π x (h+). Therefore, the destination pixel receives four incoming messages: x (h+), π x (h-), π x (v+) and π x (v-). As discussed herein, in embodiments of the present invention, other messages (e.g., messages from diagonally adjacent pixels) may also be received. Incoming messages provide information about the current label or labels associated with adjacent pixels. In some embodiments, the content of one or more outgoing messages is different because the one or more labels associated with adjacent pixels are independent.

[0054] It should also be appreciated that a single message can be designated as a different message based on the perspective of the pixels involved. For example, from the perspective of pixel v+, the outgoing message λ x (v+) is also an incoming message. Similarly, from the perspective of pixel h+, the incoming message π x (h+) is also an outgoing message. Incoming message π x (h+) may also be similar or identical to other messages sent by pixel h+ to its neighboring pixels.

[0055] As discussed herein, the data item identifies which adjacent pixels correspond to the same physical object, and the smoothing item adjusts adjacent values so that adjacent pixels corresponding to the same physical object initially have similar resulting values. In an embodiment, the belief propagator 114 may calculate one or more data items indicating one or more of a first group of pixels in a first range image, which first group of pixels corresponds to one or more of a second group of pixels in a second range image, based on one or more messages. The data items may be used because the physical object and / or the depth sensor moves between images. Variables such as depth flow, reflectivity, color value, intensity, time of flight (ToF), texture, return behavior, and / or other information (e.g., information represented by lidar data or represented in a lidar range image) may be analyzed in the data items. In an example, the physical object may be relatively constant for one or more of these variables. For example, the physical object may move with a similar depth flow because the physical object moves as a single unit. As another example, reflectivity may vary based on the physical object (e.g., a painted vehicle may have a higher reflectivity value than a pedestrian).

[0056] In an embodiment, the belief propagator 114 calculates one or more smoothness terms based at least in part on the one or more messages, which smoothness terms indicate revised depth stream values for groups of pixels in the depth stream image. The smoothness terms can identify which value of the depth stream is approximately correct for pixels corresponding to the same physical object. Typically, physical objects moving in space move as a single unit. Therefore, the incomplete and imprecise data of some pixels and the influence of neighboring pixels can be adjusted to account for this information. For example, a cost is assigned to each pixel based on the reliability of the data therein. For pixels with incomplete data (e.g., not returned to the lidar sensor), a low cost can be assigned so that they are more likely to be affected by the belief propagation algorithm for their neighboring pixels. In this way, a dense motion field can be created even if the original lidar scan data is lost.

[0057] In some embodiments, the belief propagator 114 may prevent messages from being passed across boundaries corresponding to different physical objects. Whether any two adjacent pixels correspond to the same object can be calculated by comparing the relative depths and / or depth flows associated with the two corresponding pixels. Adjacent pixels that are calculated not to correspond to the same physical object in the physical environment (e.g., because the depth difference or reflectivity between the two detected pixels exceeds a certain threshold) may not be able to pass messages between each other (e.g., messages may not be passed between nodes corresponding to pixels in an image representation, such as a matrix, table, graph, etc.). This is because these pixels should not affect their neighboring pixels because they correspond to different physical objects. Therefore, the disparity between these adjacent pixels may not be affected by the belief propagation algorithm.

[0058] In some embodiments, the belief propagator 114 may determine, based at least in part on depth data from the second lidar data, that the target pixel corresponds to a different target object than the adjacent pixels corresponding to the adjacent pixels. Based on this determination, message passing between the target pixel and the adjacent pixels may be stopped or prevented. For example, preventing message passing may be performed by determining (before sending a message) whether the pixels involved correspond to the same physical object. In other embodiments, preventing message passing may be performed by placing a virtual barrier around pixels corresponding to the same object, thereby eliminating the need to perform the determination again in subsequent iterations of the process. In other embodiments, preventing message passing may be performed by adding a label to at least one pixel indicating a physical object (e.g., an object reference number or other indicator). In this embodiment, if the label does not correspond to the same object corresponding to the target pixel, the message may be passed, but the received message may be ignored.

[0059] The system 100 may include a scene flow generator 116 that can, for example, use a coordinate converter 110 to convert the 2.5D optical or depth flow information (e.g., after message passing) back to 3D space to generate a scene flow. For example, by passing and analyzing these messages between pixels of the same depth flow image (e.g., nodes corresponding to pixels in a representation of the depth flow image), the accuracy of various data at each pixel will be improved to an optimized 2D motion vector. The optimized 2D motion vector can indicate the optical and / or depth flow changes of the physical object corresponding to the target pixel. Data for other pixels that are missing or incomplete can also be estimated and improved based at least in part on the optimized 2D motion vector.

[0060] The scene flow generator 116 may calculate one or more motion vectors corresponding to the second set of pixels based at least in part on the one or more data items and the one or more smoothness items, wherein the one or more motion vectors may represent relative pixel positions between the first range image and the second range image. The relative pixel positions may indicate movement of physical objects in the environment near the depth perception sensor 102.

[0061] In an embodiment of the present invention, the scene flow generator 116 can estimate or calculate one or more motion vectors in 2D or 2.5D space, and then convert the one or more motion vectors into 3D space to generate one or more 3D motion vectors. In an embodiment, the processor converts the determined 2D motion vector into 3D space based at least in part on the information in the 2D vector and the corresponding lidar data of the 2D vector. 2D vectors may also be referred to as 2.5D motion vectors because they include depth information and can be converted back to 3D space using a known correspondence between 2D image space and 3D world space positions (e.g., using internal and / or external sensor parameters). Thus, a 3D motion vector for a point in a lidar point cloud can be calculated, where the 3D motion vector represents various information about the displacement of the 3D points in the point cloud, for example, indicating the displacement of a physical object between lidar data frames. Thus, the message passing belief propagation algorithm allows this complex analysis to be simplified and completed before returning to 3D space, and then converted to 3D space to generate 3D scene flow information.

[0062] Thus, the scene flow generator 116 can generate a scene flow based at least in part on one or more 3D motion vectors representing the scene flow between the first range image and the second range image. The scene flow can be used to detect, identify, and / or track various physical objects within the environment of the vehicle 800, such as various physical objects within the field of view or sensory field of the depth perception sensor 102.

[0063] refer to Figures 6A-6B , an example scene stream 602 and a corresponding range image 604 are given. For example, Figure 6A An example scene flow 602 from a first direction is shown, Figure 6B Shown from Figure 6A Example scene stream 602 from a second direction (e.g., zoomed in). Example scene stream 602 displays detected objects 606 and indications of corresponding movement of the detected objects. For example, scene stream 602 includes an unfilled center region 608 that corresponds to the position of depth perception sensor 102 (although not in the Figure 6A or Figure 6B 608 ) and a radius extending therefrom. At the edge of central region 608 are a series of concentric circles 610, which may indicate that information is available for that area, but no obstacles have been detected. If obstacle 606 is detected, there may be an obstacle within concentric circles 610, and there may be an occlusion region 612 (represented by an empty space) behind obstacle 606 (from the perspective of depth perception sensor 102). As the obstacle and / or perception sensor move relative to each other, removing the occlusion, the occluding object may become visible in additional iterations of the process.

[0064] refer to Figure 1A , system 100 may include a vehicle controller 118 that may analyze information in the scene stream to determine obstacles in the physical environment surrounding depth perception sensor 102. Vehicle controller 118 may then direct one or more vehicle actions based on the determined obstacles, such as applying brakes or turning the vehicle. For example, vehicle controller 118 may execute an autonomous driving software stack, which may include a perception layer, a world model management layer, a planning layer, a control layer, a driving layer, an obstacle avoidance layer, and / or one or more other layers.

[0065] Now refer to Figure 1B , Figure 1B 1 shows a time flow diagram illustrating an example embodiment of the system 100 over time intervals. Figure 1B As shown, at time T1 during operation of system 100, lidar data 150 is collected. Lidar data 150 is combined into a 3D lidar point cloud 152, which is projected to generate a range image. Figure 5An example range image is shown in . At time T2 (i.e., after time T1), lidar data 150 is collected again and merged into a second 3D lidar point cloud 152, the lidar point cloud is projected to generate a range image, and one or more range images are used to generate a second 2.5D depth stream image 154. In this 2.5D depth stream image 154, various pixels (e.g., nodes corresponding to pixels) can pass messages between them through a message passing algorithm 156. When passing messages in the first iteration, the value of the depth stream (and possibly other labels and values) can be optimized to produce an optimized value 158 and compared with the range image. After a certain number of iterations, a 2.5D depth stream image 160 and a 3D scene stream 162 can be generated. The lidar data collected at time T3 (not shown) will be further processed and compared with the data from time T2 to further optimize the three-dimensional scene flow over time.

[0066] Now refer to Figure 7 , each block of the method 700 described herein includes a computing process that can be performed using any combination of hardware, firmware, and / or software. For example, various functions can be performed by a processor that executes instructions stored in a memory. The method 700 can also be implemented as computer-usable instructions stored on a computer storage medium. The method 700 can be provided by a standalone application, a service, or a hosted service (standalone or in combination with another hosted service), or a plug-in for another product, etc. In addition, as an example, for Figure 1A The method 700 is described with reference to the scene flow generation system 100 of FIG. However, the method may additionally or alternatively be performed by any one system or any combination of systems, including but not limited to the systems described herein.

[0067] Figure 7 7 is a flow chart of a method 700 for generating a scene flow according to some embodiments of the present invention. At block B702, method 700 includes generating at least a first range image and a second range image. For example, a first range image corresponding to a first set of lidar data generated for a first time and a second range image corresponding to a second set of lidar data generated after the first time may be generated. The range images may encode various information, such as depth, intensity, and time of flight (ToF).

[0068] At block B704, method 700 includes generating a depth flow image using the first range image and the second range image. For example, at least depth values corresponding to the first range image and the second range image can be compared to determine a depth flow value indicating a change in depth over time for a particular (e.g., matching) pixel or point in the range image. To compare the two range images, one of the range images can be converted to the coordinate space of the other range image, wherein the conversion between the first lidar data and the second lidar data used to generate the range image can be performed using the tracked ego-motion of vehicle 800.

[0069] At block B706, method 700 includes passing messages between pixels of the depth stream image. For example, to update values in the depth stream image and / or to fill in missing information for missing or incomplete data, messages can be passed between nodes corresponding to pixels of an image representation (e.g., a matrix, graph, table, etc.) via a belief propagation algorithm.

[0070] At block B708 , method 700 includes computing data items indicating which pixels correspond to which physical objects, which may include a comparison with the first depth stream image.

[0071] At block B710 , method 700 includes computing a smoothness term that adjusts the value of the label based on information received from neighboring pixels.

[0072] At block B712, method 700 includes calculating and / or optimizing a motion vector based at least in part on the data item and the smoothness item. In some embodiments, method 700 may include multiple iterations of the loop of blocks B706 through B712, wherein the smoothness item adjusts the value and the data item identifies the object.

[0073] After a certain threshold, at block B714, method 700 may include generating a scene flow based on the motion vectors.For example, the scene flow may be generated from 3D motion vectors determined by converting 2.5D motion vectors into 3D space.

[0074] At block B716, method 700 includes identifying obstacles based at least in part on the scene flow. For example, obstacles may include physical objects that can directly or indirectly affect the autonomous vehicle 800. Thus, obstacles may be in motion, such as other vehicles or pedestrians, or stationary, such as buildings and trees. Obstacles may have a position relative to the vehicle, a 3D motion vector (which may include acceleration, rotation, or other indications of motion), and a relative size (based on the number of pixels corresponding to the obstacle). In some embodiments, based at least in part on information about the obstacle, system 100 may determine the likelihood that the obstacle will affect the vehicle, and the system may determine remedial actions to be performed to avoid the obstacle. For example, vehicle controller 118 may determine that the vehicle should brake, steer, or accelerate to avoid the obstacle. In the event that the obstacle cannot be avoided, system 100 may determine that one or more remedial actions should be taken, such as minimizing damage or activating other safety features.

[0075] At block 718, method 700 includes controlling the vehicle to avoid the identified obstacle by taking the identified remedial action. Controlling the vehicle may include sending commands to any of a number of vehicle systems, such as regarding Figures 8A-8D described.

[0076] Example autonomous vehicle

[0077] Figure 8A8 is an illustration of an example autonomous vehicle 800 according to some embodiments of the present disclosure. Autonomous vehicle 800 (alternatively referred to herein as "vehicle 800") may include, but is not limited to, a passenger vehicle such as a car, a truck, a bus, an emergency vehicle, a shuttle, an electric or motorized bicycle, a motorcycle, a fire truck, a police vehicle, an ambulance, a boat, a construction vehicle, an underwater vessel, a drone, a vehicle coupled to a trailer, and / or another type of vehicle (e.g., a vehicle that is unmanned and / or accommodates one or more passengers). Autonomous vehicles are generally described according to the 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), "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles" (Standard No. J3016-201806, issued on June 15, 2018, Standard No. J3016-201609, issued on September 30, 2016, and previous and future versions of such standards). The vehicle 800 may be capable of implementing one or more functions consistent with autonomous driving levels 3 to 5. For example, depending on the embodiment, the vehicle 800 may be capable of conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5).

[0078] Vehicle 800 may include components such as a chassis, a body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other vehicle components. Vehicle 800 may include a propulsion system 850, such as an internal combustion engine, a hybrid power plant, an all-electric engine, and / or another type of propulsion system. Propulsion system 850 may be connected to a drivetrain of vehicle 800, which may include a transmission, to achieve propulsion of vehicle 800. Propulsion system 850 may be controlled in response to receiving a signal from throttle / accelerator 852.

[0079] A steering system 854, which may include a steering wheel, may be used to steer the vehicle 800 (e.g., along a desired path or route) when the propulsion system 850 is operating (e.g., when the vehicle is in motion). The steering system 854 may receive signals from a steering actuator 856. For fully automated (Level 5) functionality, a steering wheel may be optional.

[0080] Brake sensor system 846 may be used to operate vehicle brakes in response to receiving signals from brake actuator 848 and / or brake sensors.

[0081] May include one or more system on chip (SoC) 804 ( Figure 8C) and / or one or more GPUs can provide signals (e.g., representing commands) to one or more components and / or systems of the vehicle 800. For example, the one or more controllers can send signals to operate the vehicle brakes via one or more brake actuators 848, to operate the steering system 854 via one or more steering actuators 856, and to operate the propulsion system 850 via one or more throttles / accelerators 852. The one or more controllers 836 can include one or more onboard (e.g., integrated) computing devices (e.g., supercomputers) that process sensor signals and output operational commands (e.g., signals representing commands) to enable autonomous driving and / or assist a human driver in driving the vehicle 800. The one or more controllers 836 can include a first controller 836 for autonomous driving functions, a second controller 836 for functional safety functions, a third controller 836 for artificial intelligence functions (e.g., computer vision), a fourth controller 836 for infotainment functions, a fifth controller 836 for redundancy in emergency situations, and / or other controllers. In some examples, a single controller 836 may handle two or more of the above functions, two or more controllers 836 may handle a single function, and / or any combination thereof.

[0082] The one or more controllers 836 may provide signals for controlling one or more components and / or systems of the vehicle 800 in response to sensor data (e.g., sensor inputs) received from one or more sensors. The sensor data may be received from, for example and without limitation, a global navigation satellite system sensor 858 (e.g., a global positioning system sensor), a RADAR sensor 860, an ultrasonic sensor 862, a LIDAR sensor 864, an inertial measurement unit (IMU) sensor 866 (e.g., an accelerometer, a gyroscope, a magnetic compass, a magnetometer, etc.), a microphone 896, a stereo camera 868, a wide-angle camera 870 (e.g., a fisheye camera), an infrared camera 872, a surround camera 874 (e.g., a 360-degree camera), a long-range and / or mid-range camera 898, a speed sensor 844 (e.g., for measuring the velocity of the vehicle 800), a vibration sensor 842, a steering sensor 840, a brake sensor (e.g., as part of a brake sensor system 846), and / or other sensor types.

[0083] One or more of the controllers 836 may receive input (e.g., represented by input data) from the instrument cluster 832 of the vehicle 800 and provide output (e.g., represented by output data, display data, etc.) via a human machine interface (HMI) display 834, an audible annunciator, a speaker, and / or via other components of the vehicle 800. These outputs may include information such as vehicle speed, velocity, time, map data (e.g., Figure 8CThe HMI display 834 may include information such as the HD map 822 of the vehicle 800, location data (e.g., the location of the vehicle 800 on the map), directions, the locations of other vehicles (e.g., an occupancy grid), information about objects and object states as sensed by the controller 836, etc. For example, the HMI display 834 may display information about the presence of one or more objects (e.g., street signs, warning signs, traffic light changes, etc.) and / or information about driving maneuvers that the vehicle has made, is making, or will make (e.g., changing lanes now, leaving 34B in two miles, etc.).

[0084] The vehicle 800 also includes a network interface 824 that can communicate over one or more networks using one or more wireless antennas 826 and / or a modem. For example, the network interface 824 can be capable of communicating over LTE, WCDMA, UMTS, GSM, CDMA2000, etc. The one or more wireless antennas 826 can also enable communication between objects in the environment (e.g., vehicles, mobile devices, etc.) using one or more local area networks such as Bluetooth, Bluetooth LE, Z-wave, ZigBee, etc. and / or one or more low power wide area networks (LPWANs) such as LoRaWAN, SigFox, etc.

[0085] Figure 8B For use according to some embodiments of the present disclosure Figure 8A An example of camera positions and fields of view for autonomous vehicle 800 is shown. 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 located at different locations on vehicle 800.

[0086] The camera type used for the camera may include, but is not limited to, a digital camera that may be suitable for use with components and / or systems of the vehicle 800. One or more cameras may operate under Automotive Safety Integrity Level (ASIL) B and / or under another ASIL. The camera type may have any image capture rate, such as 60 frames per second (fps), 120fps, 240fps, etc., depending on the embodiment. The camera may be capable of using a rolling shutter, a global shutter, another type of shutter, or a combination thereof. In some examples, the color filter array may include a red-white-white-white (RCCC) color filter array, a red-white-white-blue (RCCB) color filter array, a red-blue-green-white (RBGC) color filter array, a Foveon X3 color filter array, a Bayer sensor (RGGB) color filter array, a monochrome sensor color filter array, and / or another type of color filter array. In some embodiments, a clear pixel camera such as a camera with an RCCC, RCCB, and / or RBGC color filter array may be used in an effort to improve light sensitivity.

[0087] 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 multi-function monocular camera can be installed to provide functions including lane departure warning, traffic sign assistance, and intelligent headlight control. One or more of the cameras (e.g., all of the cameras) can simultaneously record and provide image data (e.g., video).

[0088] One or more of the cameras can be mounted in a mounting assembly, such as a custom-designed (3-D printed) assembly, to cut off stray light and reflections from within the car (e.g., reflections from the dashboard reflected in the windshield mirror) that could interfere with the camera's ability to capture image data. With respect to the wing mirror mounting assembly, the wing mirror assembly can be custom 3-D printed so that the camera mounting plate matches the shape of the wing mirror. In some examples, one or more cameras can be integrated into the wing mirror. For side-view cameras, one or more cameras can also be integrated into the four pillars at each corner of the cab.

[0089] A camera (e.g., a front-facing camera) having a field of view that includes a portion of the environment in front of the vehicle 800 can be used for surround vision to help identify the forward path and obstacles, as well as assist in providing information critical to generating an occupancy grid and / or determining a preferred vehicle path with the help of one or more controllers 836 and / or control SoCs. The front-facing camera can be used to perform many of the same ADAS functions as LIDAR, including emergency braking, pedestrian detection, and collision avoidance. The front-facing camera can also be used for ADAS functions and systems, including lane departure warning (LDW), autonomous cruise control (ACC), and / or other functions such as traffic sign recognition.

[0090] A variety of cameras can be used in the front-facing configuration, including, for example, a monocular camera platform including a CMOS (complementary metal oxide semiconductor) color imager. Another example can be a wide-angle camera 870, which can be used to sense objects entering the field of view from the periphery (e.g., pedestrians, intersection traffic, or bicycles). Although Figure 8B The figure shows only one wide-angle camera, but there can be any number of wide-angle cameras 870 on the vehicle 800. In addition, long-range cameras 898 (e.g., a long-view stereo camera pair) can be used for depth-based object detection, especially for objects for which neural networks have not yet been trained. Long-range cameras 898 can also be used for object detection and classification and basic object tracking.

[0091] One or more stereo cameras 868 may also be included in the front configuration. The stereo camera 868 may include an integrated control unit including an expandable processing unit that may provide a multi-core microprocessor and programmable logic (FPGA) with an integrated CAN or Ethernet interface on a single chip. Such a unit may be used to generate a 3-D map of the vehicle environment, including distance estimates for all points in the image. An alternative stereo camera 868 may include a compact stereo vision sensor that may include two camera lenses (one on the left and one on the right) and an image processing chip that may measure the distance from the vehicle to the target object and use the generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning functions. Other types of stereo cameras 868 may be used in addition to or alternatively to those described herein.

[0092] Cameras with a field of view that includes portions of the environment to the sides of the vehicle 800 (e.g., side-view cameras) can be used for surround viewing, providing information used to create and update occupancy grids and generate side impact collision warnings. For example, surround cameras 874 (e.g., Figure 8B Four surround cameras 874 (shown in FIG) can be placed on the vehicle 800. The surround cameras 874 can include a wide-angle camera 870, a fisheye camera, a 360-degree camera, and / or the like. For example, the four fisheye cameras can be placed on the front, rear, and sides of the vehicle. In an alternative arrangement, the vehicle can use three surround cameras 874 (e.g., left, right, and rear), and can utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround-view camera.

[0093] A camera having a field of view that includes a portion of the environment behind the vehicle 800 (e.g., a rearview camera) can be used to assist with parking, surround view, rear collision warning, and creating and updating occupancy grids. A variety of cameras can be used, including but not limited to cameras that are also suitable as front-facing cameras as described herein (e.g., long-range and / or mid-range cameras 898, stereo cameras 868, infrared cameras 872, etc.).

[0094] Figure 8C For use according to some embodiments of the present disclosure Figure 8A800 . It will be understood that this arrangement and other arrangements described herein are set forth merely as examples. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, functional groupings, etc.) may be used in addition to or in place of those shown, and some elements may be omitted entirely. Further, many of the elements described herein are functional entities that may be implemented as discrete or distributed components or in conjunction with other components, and in any appropriate combination and location. The various functions described herein as being performed by entities may be implemented by hardware, firmware, and / or software. For example, the various functions may be implemented by a processor executing instructions stored in a memory.

[0095] Figure 8C Each of the components, features, and systems of vehicle 800 is illustrated as being connected via bus 802. Bus 802 may include a controller area network (CAN) data interface (alternatively, referred to herein as a "CAN bus"). CAN may be a network internal to vehicle 800 that assists in controlling various features and functions of vehicle 800, such as actuation of brakes, acceleration, braking, steering, windshield wipers, and the like. The CAN bus may be configured to have tens or even hundreds of nodes, each with its own unique identifier (e.g., CAN ID). The CAN bus may be read to find steering wheel angle, ground speed, engine revolutions per minute (RPM), button positions, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0096] Although bus 802 is described here as a CAN bus, this is not intended to be limiting. For example, FlexRay and / or Ethernet may be used in addition to or in lieu of a CAN bus. Furthermore, although bus 802 is represented by a single line, this is not intended to be limiting. For example, there may be any number of buses 802, 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 using different protocols. In some examples, two or more buses 802 may be used to perform different functions and / or may be used for redundancy. For example, a first bus 802 may be used for collision avoidance functionality, and a second bus 802 may be used for drive control. In any example, each bus 802 may communicate with any component of vehicle 800, and two or more buses 802 may communicate with the same component. In some examples, each SoC 804, each controller 836, and / or each computer within the vehicle may have access to the same input data (e.g., input from sensors within vehicle 800) and may be connected to a common bus such as a CAN bus.

[0097] The vehicle 800 may include one or more controllers 836, such as those described herein. Figure 8A Controller 836 may be used for a variety of functions. Controller 836 may be coupled to any of the other various components and systems of vehicle 800 and may be used for control of vehicle 800, artificial intelligence of vehicle 800, infotainment for vehicle 800, and / or the like.

[0098] The vehicle 800 may include one or more system-on-chips (SoCs) 804. The SoC 804 may include one or more CPUs 806, one or more GPUs 808, one or more processors 810, one or more caches 812, one or more accelerators 814, one or more data stores 816, and / or other components and features not shown. The SoC 804 may be used to control the vehicle 800 in a variety of platforms and systems. For example, the one or more SoCs 804 may be combined with an HD map 822 in a system (e.g., a system of the vehicle 800), which may be downloaded from one or more servers (e.g., a server) via a network interface 824. Figure 8D one or more servers 878) to obtain map refreshes and / or updates.

[0099] The CPU 806 may include a CPU cluster or CPU complex (alternatively, referred to herein as a "CCPLEX"). The CPU 806 may include multiple cores and / or L2 caches. For example, in some embodiments, the CPU 806 may include eight cores in a coherent multiprocessor configuration. In some embodiments, the CPU 806 may include four dual-core clusters, each with a dedicated L2 cache (e.g., a 2MB L2 cache). The CPU 806 (e.g., CCPLEX) may be configured to support simultaneous cluster operations such that any combination of CPU 806 clusters can be active at any given time.

[0100] The CPU 806 may implement power management capabilities including one or more of the following features: each hardware block may be automatically clock gated when idle to conserve dynamic power; each core clock may be gated when the core is not actively executing instructions due to the execution of WFI / WFE instructions; each core may be independently power gated; each core cluster may be independently clock gated when all cores are clock gated or power gated; and / or each core cluster may be independently power gated when all cores are power gated. The CPU 806 may further implement an enhanced algorithm for managing power states, in which allowed power states and expected wakeup times are specified, and hardware / microcode determines the optimal power state to enter for the core, cluster, and CCPLEX. The processing core may support a simplified power state entry sequence in software, with this work being offloaded to the microcode.

[0101] The GPU 808 may include an integrated GPU (alternatively referred to herein as an "iGPU"). The GPU 808 may be programmable and efficient for parallel workloads. In some examples, the GPU 808 may use an enhanced tensor instruction set. The GPU 808 may include one or more streaming microprocessors, each of which may include an L1 cache (e.g., an L1 cache with at least 96KB of storage capacity), and two or more of these streaming microprocessors may share an L2 cache (e.g., an L2 cache with 512KB of storage capacity). In some embodiments, the GPU 808 may include at least eight streaming microprocessors. The GPU 808 may use a computing application programming interface (API). In addition, the GPU 808 may use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0102] In the case of automotive and embedded use, GPU 808 can be power optimized to achieve optimal performance. For example, GPU 808 can be manufactured on fin field effect transistors (FinFETs). However, this is not intended to be limiting, and GPU 808 can be manufactured using other semiconductor manufacturing processes. Each streaming microprocessor can merge several mixed precision processing cores divided into multiple blocks. For example and without limitation, 64 PF32 cores and 32 PF64 cores can be divided into four processing blocks. In such an example, each processing block can be allocated 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed precision NVIDIA tensor cores for deep learning matrix arithmetic, L0 instruction cache, warp scheduler, dispatch unit and / or 64KB register file. In addition, the streaming microprocessor may include independent parallel integer and floating point data paths to provide efficient execution of workloads using a mix of computation and addressing calculations. The streaming microprocessor may include independent thread scheduling capabilities to allow for finer-grained synchronization and collaboration between parallel threads. A streaming microprocessor may include a combined L1 data cache and shared memory unit to increase performance while simplifying programming.

[0103] The GPU 808 can include high bandwidth memory (HBM) and / or a 16GB HBM2 memory subsystem that provides a peak memory bandwidth of approximately 900 GB / s in some examples. In some examples, synchronous graphics random access memory (SGRAM), such as fifth generation graphics double data rate synchronous random access memory (GDDR5), can be used in addition to or in lieu of HBM memory.

[0104] The GPU 808 may include unified memory technology that includes access counters to allow memory pages to be more accurately migrated to the processor that accesses them most frequently, thereby improving the efficiency of memory ranges shared between processors. In some examples, address translation services (ATS) support may be used to allow the GPU 808 to directly access the CPU 806 page tables. In such an example, when the GPU 808 memory management unit (MMU) experiences a miss, an address translation request may be transmitted to the CPU 806. In response, the CPU 806 may look up the virtual-to-physical mapping for the address in its page table and transmit the translation back to the GPU 808. In this way, unified memory technology may allow a single unified virtual address space to be used for memory of both the CPU 806 and the GPU 808, thereby simplifying GPU 808 programming and porting applications to the GPU 808.

[0105] In addition, GPU 808 can include access counters that can track how often GPU 808 accesses the memory of other processors. The access counters can help ensure that memory pages are moved to the physical memory of the processor that accesses them most frequently.

[0106] SoC 804 may include any number of caches 812, including those described herein. For example, cache 812 may include an L3 cache usable by one or more CPUs 806 and one or more GPUs 808 (e.g., connected to CPUs 806 and GPUs 808). Cache 812 may include a write-back cache that can track the state of lines, for example, using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). Depending on the embodiment, the L3 cache may include 4MB or more, although smaller cache sizes may also be used.

[0107] The SoC 804 may include an arithmetic logic unit (ALU) that may be utilized in performing any of a variety of tasks or operations associated with the vehicle 800, such as processing a DNN. Furthermore, the SoC 804 may include a floating point unit (FPU) (or other math coprocessor or digital coprocessor type) for performing mathematical operations within the system. For example, the SoC 104 may include one or more FPUs integrated as execution units within the CPU 806 and / or GPU 808.

[0108] SoC 804 may include one or more accelerators 814 (e.g., hardware accelerators, software accelerators, or a combination thereof). For example, SoC 804 may include a hardware accelerator 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 accelerator cluster to accelerate neural networks and other calculations. The hardware accelerator cluster may be used to supplement the GPU 808 and offload some tasks of the GPU 808 (e.g., freeing up more cycles of the GPU 808 for performing other tasks). As an example, the accelerator 814 may be used for targeted workloads (e.g., perception, convolutional neural networks (CNNs), etc.) that are sufficiently stable to be easily controlled for acceleration. When used herein, the term "CNN" may include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and fast RCNNs (e.g., for object detection).

[0109] The accelerator 814 (e.g., a hardware accelerator cluster) may include a deep learning accelerator (DLA). The DLA may include one or more tensor processing units (TPUs) that can be configured to provide an additional 10 trillion operations per second for deep learning applications and reasoning. The TPU may be an accelerator configured to perform image processing functions (e.g., for CNN, RCNN, etc.) and optimized for performing image processing functions. The DLA may be further optimized for a specific set of neural network types and floating-point operations and reasoning. The design of the DLA may provide higher performance per millimeter than a general-purpose GPU and far exceed the performance of the CPU. The TPU may perform several functions, including a single-instance convolution function, support for INT8, INT16, and FP16 data types for both features and weights, and post-processor functions.

[0110] DLA can quickly and efficiently execute neural networks, particularly CNNs, on processed or unprocessed data for any of a wide variety of functions, such as, but not limited to: CNNs for object recognition and detection using data from camera sensors; CNNs for distance estimation using data from camera sensors; CNNs for emergency vehicle detection and recognition and detection using data from microphones; CNNs for facial recognition and vehicle owner identification using data from camera sensors; and / or CNNs for safety and / or security-related events.

[0111] The DLA can perform any function of the GPU 808, and by using an inference accelerator, for example, the designer can target any function to either the DLA or the GPU 808. For example, the designer can focus the processing of CNNs and floating-point operations on the DLA and leave other functions to the GPU 808 and / or other accelerators 814.

[0112] The accelerator 814 (e.g., a hardware accelerator cluster) may include one or more programmable vision accelerators (PVAs), which may be referred to herein alternatively as computer vision accelerators. 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. One or more PVAs may provide a balance between performance and flexibility. For example, each PVA may include, for example and without limitation, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors.

[0113] The RISC core can interact with an image sensor (e.g., an image sensor of any camera described herein), one or more image signal processors, and / or the like. Each of these RISC cores can include any amount of memory. Depending on the embodiment, the RISC core can use any of a number of protocols. In some examples, the RISC core can execute a real-time operating system (RTOS). The RISC core can be implemented using one or more integrated circuit devices, application specific integrated circuits (ASICs), and / or memory devices. For example, the RISC core can include an instruction cache and / or tightly coupled RAM.

[0114] The DMA can enable components of the PVA to access system memory independently of the CPU 806. The DMA can support any number of features used to provide optimizations for the PVA, including but not limited to support for multi-dimensional addressing and / or circular addressing. In some examples, the DMA can support addressing in up to six or more dimensions, which can include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0115] A vector processor can be a programmable processor that can be designed to efficiently and flexibly execute programming for computer vision algorithms and provide signal processing capabilities. In some examples, the PVA can include a PVA core and two vector processing subsystem partitions. The PVA core can include a processor subsystem, one or more DMA engines (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem can operate as the main processing engine of the PVA and can include a vector processing unit (VPU), an instruction cache, and / or a vector memory (e.g., VMEM). The VPU core can include a digital signal processor, such as, for example, a single instruction multiple data (SIMD), a very long instruction word (VLIW) digital signal processor. The combination of SIMD and VLIW can enhance throughput and speed.

[0116] Each of the vector processors can include an instruction cache and can be coupled to dedicated memory. As a result, in some examples, each of the vector processors can be configured to execute independently of the other vector processors. In other examples, the vector processors included in a particular PVA can be configured to employ data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA can execute the same computer vision algorithm, but on different regions of an image. In other examples, the vector processors included in a particular PVA can execute different computer vision algorithms simultaneously on the same image, or even execute different algorithms on sequential images or portions of images. Among other things, any number of PVAs can be included in a hardware accelerator cluster, and any number of vector processors can be included in each of these PVAs. In addition, the PVAs can include additional error correction code (ECC) memory to enhance overall system security.

[0117] One or more accelerators 814 (e.g., a hardware accelerator cluster) may include an on-chip computer vision network and SRAM to provide high bandwidth, low latency SRAM for one or more accelerators 814. In some examples, the on-chip memory may include at least 4MB of SRAM consisting of, for example and not limited to, eight field-configurable memory blocks that can be accessed 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 DLA may access the memory via a backbone that provides high-speed memory access to the PVA and DLA. The backbone may include an on-chip computer vision network that interconnects the PVA and DLA to the memory (e.g., using APB).

[0118] The on-chip computer vision network can include an interface that ensures that both the PVA and DLA provide ready and valid signals before transmitting any control signals / addresses / data. Such an interface can provide separate phases and separate channels for transmitting control signals / addresses / data, as well as burst-based communication for continuous data transmission. This type of interface can comply with ISO 26262 or IEC 61508 standards, but other standards and protocols can also be used.

[0119] In some examples, the SoC 804 may include a real-time ray tracing hardware accelerator such as that described in U.S. patent application Ser. No. 16 / 101,232 filed on Aug. 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the position and extent of objects (e.g., within a world model) to generate real-time visualization simulations for use in RADAR signal interpretation, for sound propagation synthesis and / or analysis, for SONAR system simulations, for general wave propagation simulations, for comparison with LIDAR data for positioning and / or other functions, and / or for other uses. In some embodiments, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing related operations.

[0120] The accelerator 814 (e.g., a hardware accelerator cluster) has a wide range of uses in autonomous driving. The PVA can be a programmable vision accelerator that can be used in key processing stages in ADAS and autonomous vehicles. The capabilities of the PVA are a good match for algorithmic domains that require predictable processing, low power, and low latency. In other words, the PVA performs well on semi-intensive or intensive rule computations, and even on small data sets that require predictable runtimes with low latency and low power. Therefore, in the context of a platform for autonomous vehicles, the PVA is designed to run classic computer vision algorithms because they are efficient at object detection and integer math operations.

[0121] For example, according to one embodiment of the technology, PVA is used to perform computer stereo vision. In some examples, a semi-global matching-based algorithm can be used, but this is not intended to be limiting. Many applications for Level 3-5 autonomous driving require on-the-fly motion estimation / stereo matching (e.g., structure from motion, pedestrian recognition, lane detection, etc.). PVA can perform computer stereo vision functions on input from two monocular cameras.

[0122] In some examples, PVA can be used to perform dense optical flow, by processing raw RADAR data (e.g., using a 4D Fast Fourier Transform) to provide processed RADAR. In other examples, PVA is used for time-of-flight depth processing, by processing raw time-of-flight data to provide processed time-of-flight data.

[0123] The DLA can be used to run any type of network to enhance control and driving safety, including, for example, a neural network that outputs a confidence metric for each object detection. Such confidence values can be interpreted as probabilities, or as providing a relative "weight" of each detection compared to other detections. This confidence value enables the system to make further decisions about which detections should be considered true positives versus false positives. For example, the system can set a threshold for confidence and only consider detections that exceed the threshold as true positives. In an automatic emergency braking (AEB) system, a false positive detection could cause the vehicle to automatically apply emergency braking, which is clearly undesirable. Therefore, only the most confident detections should be considered triggers for AEB. The DLA can run a neural network to regress the confidence value. This neural network can take as its input at least some subset of parameters, such as bounding box dimensions, a ground plane estimate obtained (e.g., from another subsystem), inertial measurement unit (IMU) sensor 866 output related to the vehicle 800's orientation and distance, and 3D position estimates of objects obtained from the neural network and / or other sensors (e.g., LIDAR sensor 864 or RADAR sensor 860).

[0124] The SoC 804 may include one or more data stores 816 (e.g., memory). The data store 816 may be on-chip memory of the SoC 804 that may store neural networks to be executed on the GPU and / or DLA. In some examples, the data store 816 may be large enough to store multiple instances of the neural network for redundancy and safety. The data store 812 may include an L2 or L3 cache 812. References to the data store 816 may include references to memory associated with the PVA, DLA, and / or other accelerators 814 as described herein.

[0125] SoC 804 may include one or more processors 810 (e.g., embedded processors). Processor 810 may include a boot and power management processor, which may be a dedicated processor and subsystem for handling boot power and management functions and related safety implementations. The boot and power management processor may be part of the SoC 804 boot sequence and may provide runtime power management services. The boot power and management processor may provide clock and voltage programming, auxiliary system low power state transitions, SoC 804 thermal and temperature sensor management, and / or SoC 804 power state management. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the SoC 804 may use the ring oscillator to detect the temperature of the CPU 806, GPU 808, and / or accelerator 814. If it is determined that the temperature exceeds a threshold, the boot and power management processor may enter a temperature fault routine and place the SoC 804 in a lower power state and / or place the vehicle 800 in a driver safety parking mode (e.g., to safely park the vehicle 800).

[0126] The processor 810 may also include a set of embedded processors that can be used as an audio processing engine. The audio processing engine can be an audio subsystem that allows for full hardware support for multi-channel audio through multiple interfaces and a wide range of flexible audio I / O interfaces. In some examples, the audio processing engine is a dedicated processor core having a digital signal processor with dedicated RAM.

[0127] The processor 810 may also include an always-on processor engine that can provide the necessary hardware features 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 (such as timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0128] The processor 810 may also include a safety cluster engine, which includes a dedicated processor subsystem that handles safety management of automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (such as timers, interrupt controllers, etc.), and / or routing logic. In safety mode, the two or more cores can operate in lockstep mode and act as a single core with comparison logic to detect any differences between their operations.

[0129] Processor 810 may also include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0130] Processor 810 may also 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.

[0131] The processor 810 may include a video image compositer, which may be a processing block (e.g., implemented on a microprocessor) that implements the video post-processing functions required by the video playback application to produce the final image for the player window. The video image compositer may perform lens distortion correction for the wide-angle camera 870, the surround camera 874, and / or for the in-cab monitoring camera sensor. The in-cab monitoring camera sensor is preferably monitored by a neural network running on another instance of the advanced SoC, configured to recognize in-cab events and respond accordingly. The in-cab system may perform lip reading to activate mobile phone service and place calls, dictate emails, change vehicle destinations, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain functions are only available to the driver when the vehicle is operating in autonomous mode and are disabled in other circumstances.

[0132] The video image compositer can include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, in the presence of motion in the video, the noise reduction appropriately weights spatial information and downweights information provided by neighboring frames. In the case where an image or portion of an image does not include motion, the temporal noise reduction performed by the video image compositer can use information from previous images to reduce noise in the current image.

[0133] The video image compositor can also be configured to perform stereo rectification on the input stereo footage frames. The video image compositor can further be used for user interface composition when the operating system desktop is in use and the GPU 808 does not need to continuously render new surfaces. Even when the GPU 808 is powered on and active for 3D rendering, the video image compositor can be used to offload the GPU 808 to improve performance and responsiveness.

[0134] The SoC 804 may also include a Mobile Industry Processor Interface (MIPI) camera serial interface, a high-speed interface for receiving video and input from a camera, and / or a video input block that may be used for camera and related pixel input functions. The SoC 804 may also include an input / output controller that may be controlled by software and may be used to receive I / O signals that are not assigned to a specific role.

[0135] The SoC 804 may also include a wide range of peripheral interfaces to enable communication with peripherals, audio codecs, power management, and / or other devices. The SoC 804 may be used to process data from cameras (connected via Gigabit multimedia serial links and Ethernet), sensors (e.g., LIDAR sensor 864, RADAR sensor 860, etc., which may be connected via Ethernet), data from the bus 802 (e.g., vehicle 800 speed, steering wheel position, etc.), and data from the GNSS sensor 858 (connected via Ethernet or a CAN bus). The SoC 804 may also include dedicated high-performance mass storage controllers, which may include their own DMA engines and which may be used to free the CPU 806 from routine data management tasks.

[0136] SoC 804 can be an end-to-end platform with a flexible architecture that spans levels 3-5 of automation, providing a comprehensive functional safety architecture that leverages and efficiently uses computer vision and ADAS technologies for diversity and redundancy, along with deep learning tools to provide a platform for a flexible and reliable driving software stack. SoC 804 can be faster, more reliable, and even more energy- and space-efficient than conventional systems. For example, when combined with CPU 806, GPU 808, and data storage 816, accelerator 814 can provide a fast and efficient platform for level 3-5 autonomous vehicles.

[0137] This technology therefore provides capabilities and functionality that cannot be achieved with conventional systems. For example, computer vision algorithms can be executed on CPUs, which can be configured using high-level programming languages such as the C programming language to execute a wide variety of processing algorithms across a wide variety of visual data. However, CPUs often fail to meet the performance requirements of many computer vision applications, such as those related to execution time and power consumption. In particular, many CPUs are unable to execute complex object detection algorithms in real time, a requirement for in-vehicle ADAS applications and practical Level 3-5 autonomous vehicles.

[0138] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a cluster of hardware accelerators, the technology described herein allows multiple neural networks to be executed simultaneously and / or sequentially, and the results to be combined to achieve Level 3-5 autonomous driving capabilities. For example, a CNN executed on a DLA or dGPU (e.g., GPU 820) can include text and word recognition, allowing the supercomputer to read and understand traffic signs, including signs for which a neural network has not been specifically trained. The DLA can also include a neural network that can recognize, interpret, and provide semantic understanding of the signs, and pass that semantic understanding to a path planning module running on the CPU complex.

[0139] As another example, as required for Level 3, 4, or 5 driving, multiple neural networks can be run simultaneously. For example, a warning sign consisting of "Caution: Flashing lights indicate icing conditions" along with a light can be interpreted by several neural networks, either independently or collectively. The sign itself can be identified as a traffic sign by a first neural network deployed (e.g., a trained neural network), and the text "Flashing lights indicate icing conditions" can be interpreted by a second neural network deployed, which informs the vehicle's path planning software (preferably executing on a CPU complex) that icing conditions exist when the flashing lights are detected. The flashing lights can be identified by operating a third neural network deployed over multiple frames, which informs the vehicle's path planning software of the presence (or absence) of the flashing lights. All three neural networks can run simultaneously, for example, within the DLA and / or on GPU 808.

[0140] In some examples, a CNN for facial recognition and owner recognition can use data from a camera sensor to identify the presence of an authorized driver and / or owner of the vehicle 800. The always-on sensor processing engine can be used to unlock the vehicle and turn on the lights when the owner approaches the driver's door, and in security mode, disable the vehicle when the owner leaves the vehicle. In this way, the SoC 804 provides security against theft and / or carjacking.

[0141] In another example, a CNN for emergency vehicle detection and identification can use data from microphone 896 to detect and identify emergency vehicle sirens. In contrast to conventional systems that use general classifiers to detect sirens and manually extract features, SoC 804 uses CNN to classify environmental and urban sounds and to classify visual data. In a preferred embodiment, the CNN running on the DLA is trained to identify the relative closing speed of emergency vehicles (for example, by using the Doppler effect). The CNN can also be trained to identify emergency vehicles specific to the local area in which the vehicle is operating, as identified by the GNSS sensor 858. Thus, for example, when operating in Europe, the CNN will seek to detect European sirens, and when in the United States, the CNN will seek to identify only North American sirens. Once an emergency vehicle is detected, with the assistance of the ultrasonic sensor 862, the control program can be used to execute the emergency vehicle safety routine to slow the vehicle, pull over, stop the vehicle, and / or idle the vehicle until the emergency vehicle passes.

[0142] The vehicle may include a CPU 818 (e.g., a discrete CPU or dCPU) that may be coupled to the SoC 804 via a high-speed interconnect (e.g., PCIe). The CPU 818 may include, for example, an X86 processor. The CPU 818 may be used to perform any of a variety of functions, including, for example, arbitrating potentially inconsistent results between ADAS sensors and the SoC 804, and / or monitoring the status and health of the controller 836 and / or the infotainment SoC 830.

[0143] The vehicle 800 may include a GPU 820 (e.g., a discrete GPU or dGPU) that may be coupled to the SoC 804 via a high-speed interconnect (e.g., NVIDIA's NVLINK). The GPU 820 may provide additional artificial intelligence functionality, for example, by executing redundant and / or different neural networks, and may be used to train and / or update the neural network based at least in part on input from sensors of the vehicle 800 (e.g., sensor data).

[0144] The vehicle 800 may also include a network interface 824, which may include one or more wireless antennas 826 (e.g., one or more wireless antennas for different communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). The network interface 824 can be used to enable wireless connections to the cloud (e.g., to a server 878 and / or other network devices), to other vehicles, and / or to computing devices (e.g., a passenger's client device) via the Internet. In order to communicate with other vehicles, a direct link can be established between the two vehicles, and / or an indirect link can be established (e.g., across a network and through the Internet). The direct link can be provided using a vehicle-to-vehicle communication link. The vehicle-to-vehicle communication link can provide the vehicle 800 with information about vehicles approaching the vehicle 800 (e.g., vehicles in front of, to the side of, and / or behind the vehicle 800). This functionality can be part of the cooperative adaptive cruise control functionality of the vehicle 800.

[0145] The network interface 824 may include a SoC that provides modulation and demodulation functions and enables the controller 836 to communicate over a wireless network. The network interface 824 may include an RF front-end for up-conversion from baseband to radio frequency and down-conversion from radio frequency to baseband. The frequency conversion may be performed by a well-known process and / or may be performed using a super-heterodyne process. In some examples, the RF front-end function may be provided by a separate chip. The network interface may include wireless functions for communicating over LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0146] The vehicle 800 may also include data storage 828, which may include off-chip storage (e.g., outside the SoC 804). The data storage 828 may include one or more storage elements, including RAM, SRAM, DRAM, VRAM, flash memory, a hard disk, and / or other components and / or devices that can store at least one bit of data.

[0147] The vehicle 800 may also include a GNSS sensor 858. The GNSS sensor 858 (e.g., GPS, assisted GPS sensor, differential GPS (DGPS) sensor, etc.) is used to assist with mapping, perception, occupancy grid generation, and / or path planning functions. Any number of GNSS sensors 858 may be used, including, for example and without limitation, GPS using a USB connector with an Ethernet to serial (RS-232) bridge.

[0148] The vehicle 800 may also include one or more RADAR sensors 860. The RADAR sensor 860 can be used by the vehicle 800 for remote vehicle detection even in darkness and / or in adverse weather conditions. The RADAR functional safety level can be ASIL B. The RADAR sensor 860 can use CAN and / or bus 802 (e.g., to transmit data generated by the RADAR sensor 860) for control and access to object tracking data, and in some examples access Ethernet to access raw data. A variety of RADAR sensor types can be used. For example and without limitation, the RADAR sensor 860 can be suitable for front, rear, and side RADAR use. In some examples, a pulsed Doppler RADAR sensor is used.

[0149] The RADAR sensor 860 can include different configurations, such as long-range with a narrow field of view, short-range with a wide field of view, short-range side coverage, and so on. In some examples, the long-range RADAR can be used for adaptive cruise control functions. The long-range RADAR system can provide a wide field of view (e.g., within a range of 250m) achieved by two or more independent scans. The RADAR sensor 860 can help distinguish between static objects and moving objects and can be used by the ADAS system for emergency braking assistance and forward collision warnings. The long-range RADAR sensor may include a single-station multimode RADAR with multiple (e.g., six or more) fixed RADAR antennas and high-speed CAN and FlexRay interfaces. In the example with six antennas, the central four antennas can create a focused beam pattern that is designed to record the surroundings of the vehicle 800 at a higher rate with minimal traffic interference from adjacent lanes. The other two antennas can expand the field of view, making it possible to quickly detect vehicles entering or leaving the lane of the vehicle 800.

[0150] As an example, a medium-range RADAR system may include a range of up to 860m (front) or 80m (rear) and a field of view of up to 42 degrees (front) or 850 degrees (rear). A short-range RADAR system may include, but is not limited to, a RADAR sensor designed to be mounted on both ends of the rear bumper. When mounted on both ends of the rear bumper, such a RADAR sensor system can create two beams that continuously monitor the blind spots behind and beside the vehicle.

[0151] Short-range RADAR systems can be used in ADAS systems for blind spot detection and / or lane change assistance.

[0152] Vehicle 800 may also include ultrasonic sensors 862. Ultrasonic sensors 862, which may be located on the front, rear, and / or sides of vehicle 800, may be used for parking assistance and / or for creating and updating an occupancy grid. A variety of ultrasonic sensors 862 may be used, and different ultrasonic sensors 862 may have different detection ranges (e.g., 2.5 m, 4 m). Ultrasonic sensors 862 may operate at functional safety level ASIL B.

[0153] Vehicle 800 may include a LIDAR sensor 864. LIDAR sensor 864 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. LIDAR sensor 864 may be ASIL B functional safety level. In some examples, vehicle 800 may include multiple LIDAR sensors 864 (e.g., two, four, six, etc.) that may use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

[0154] In some examples, the LIDAR sensor 864 may be capable of providing a list of objects and their distances for a 360-degree field of view. Commercially available LIDAR sensors 864 may have, for example, an advertised range of approximately 800 meters, an accuracy of 2-3 cm, and support for 800 Mbps Ethernet connections. In some examples, one or more non-obtrusive LIDAR sensors 864 may be used. In such examples, the LIDAR sensor 864 may be implemented as a small device that can be embedded in the front, back, sides, and / or corners of the vehicle 800. In such examples, the LIDAR sensor 864 may provide a field of view of up to 120 degrees horizontally and 35 degrees vertically, with a range of 200 meters, even for low-reflectivity objects. The front-mounted LIDAR sensor 864 may be configured for a horizontal field of view between 45 and 135 degrees.

[0155] In some examples, LIDAR technologies such as 3D flash LIDAR may also be used. 3D flash LIDAR uses flashes of laser as an emission source to illuminate the vehicle's surroundings up to about 200 m. The flash LIDAR unit includes a receiver that records the laser pulse transmission time and reflected light on each pixel, which in turn corresponds to the range from the vehicle to the object. Flash LIDAR can allow a highly accurate and distortion-free image of the surrounding environment to be generated with each laser flash. In some examples, four flash LIDAR sensors can be deployed, one on each side of the vehicle 800. Available 3D flash LIDAR systems include solid-state 3D staring array LIDAR cameras (e.g., non-scanning LIDAR devices) with no moving parts other than a fan. The flash LIDAR device can use 5 nanosecond Class I (eye-safe) laser pulses per frame and can capture the reflected laser light in the form of a 3D range point cloud and co-registered intensity data. By using flash LIDAR, and because flash LIDAR is a solid-state device with no moving parts, the LIDAR sensor 864 may be less susceptible to motion blur, vibration, and / or shock.

[0156] The vehicle may also include an IMU sensor 866. In some examples, the IMU sensor 866 may be located at the center of the rear axle of the vehicle 800. The IMU sensor 866 may include, for example and without limitation, an accelerometer, a magnetometer, a gyroscope, a magnetic compass, and / or other sensor types. In some examples, such as in a six-axis application, the IMU sensor 866 may include an accelerometer and a gyroscope, while in a nine-axis application, the IMU sensor 866 may include an accelerometer, a gyroscope, and a magnetometer.

[0157] In some embodiments, the IMU sensor 866 can be implemented as a miniature high-performance GPS-assisted inertial navigation system (GPS / INS) that combines micro-electromechanical systems (MEMS) inertial sensors, a high-sensitivity GPS receiver, and advanced Kalman filtering algorithms to provide estimates of position, velocity, and attitude. Thus, in some examples, the IMU sensor 866 can enable the vehicle 800 to estimate heading without the need for input from a magnetic sensor by directly observing and correlating velocity changes from the GPS to the IMU sensor 866. In some examples, the IMU sensor 866 and the GNSS sensor 858 can be combined into a single integrated unit.

[0158] The vehicle may include microphones 896 positioned in and / or around the vehicle 800. The microphones 896 may be used for, among other things, emergency vehicle detection and identification.

[0159] The vehicle may also include any number of camera types, including stereo cameras 868, wide angle cameras 870, infrared cameras 872, surround cameras 874, long and / or medium range cameras 898, and / or other camera types. These cameras may be used to capture image data around the entire periphery of the vehicle 800. The type of camera used depends on the embodiment and the requirements of the vehicle 800, and any combination of camera types may be used to provide the necessary coverage around the vehicle 800. Additionally, 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 another number of cameras. As an example and not limitation, the cameras may support Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Each of the cameras described herein may include a GMSL and / or Gigabit Ethernet network. Figure 8A and Figure 8B Described in more detail.

[0160] Vehicle 800 may also include a vibration sensor 842. Vibration sensor 842 can measure vibrations of vehicle components, such as axles. For example, changes in vibration can indicate changes in the road surface. In another example, when two or more vibration sensors 842 are used, the difference between the vibrations can be used to determine friction or slippage of the road surface (e.g., when there is a vibration difference between a powered drive shaft and a freely rotating shaft).

[0161] The vehicle 800 may include an ADAS system 838. In some examples, the ADAS system 838 may include a SoC. The ADAS system 838 may include autonomous / adaptive / automatic cruise control (ACC), cooperative adaptive cruise control (CACC), forward collision warning (FCW), automatic emergency braking (AEB), lane departure warning (LDW), lane keeping assist (LKA), blind spot warning (BSW), rear cross traffic warning (RCTW), collision warning system (CWS), lane centering (LC), and / or other features and functions.

[0162] The ACC system can utilize RADAR sensor 860, LIDAR sensor 864, and / or cameras. The ACC system can include longitudinal ACC and / or lateral ACC. Longitudinal ACC monitors and controls the distance to the vehicle immediately in front of vehicle 800, automatically adjusting the vehicle speed to maintain a safe distance from the vehicle in front. Lateral ACC maintains distance and, when necessary, recommends that vehicle 800 change lanes. Lateral ACC is related to other ADAS applications such as LCA and CWS.

[0163] CACC uses information from other vehicles, which can be received from other vehicles indirectly via a wireless link or through a network connection (e.g., through the Internet) via the network interface 824 and / or the wireless antenna 826. A direct link can be provided by a vehicle-to-vehicle (V2V) communication link, while an indirect link can be an infrastructure-to-vehicle (I2V) communication link. Typically, the V2V communication concept provides information about the vehicle immediately ahead (e.g., the vehicle immediately ahead of the vehicle 800 and in the same lane as it), while the I2V communication concept provides information about traffic further ahead. The CACC system can include either or both of the I2V and V2V information sources. Given information about the vehicle ahead of the vehicle 800, CACC can be more reliable, and it has the potential to improve the smoothness of traffic flow and reduce road congestion.

[0164] The FCW system is designed to alert the driver to hazards so that the driver can take corrective action. The FCW system uses a front-facing camera and / or RADAR sensor 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC, which is electrically coupled to driver feedback such as a display, speaker, and / or vibrating component. The FCW system can provide warnings in the form of, for example, audible, visual warnings, vibrations, and / or rapid brake pulses.

[0165] The AEB system detects an impending forward collision with another vehicle or other object and can automatically apply the brakes if the driver does not take corrective action within specified time or distance parameters. The AEB system can use a front-facing camera and / or RADAR sensor 860 coupled to 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, and if the driver does not take corrective action, the AEB system can automatically apply the brakes in an effort to prevent or at least mitigate the effects of the predicted collision. The AEB system can include technologies such as dynamic brake support and / or collision approach braking.

[0166] The LDW system provides visual, audible, and / or tactile warnings, such as steering wheel or seat vibrations, to alert the driver when the vehicle 800 crosses a lane marking. When the driver indicates an intention to leave the lane by activating a turn signal, the LDW system is deactivated. The LDW system may utilize a front-facing camera coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0167] The LKA system is a variation of the LDW system. If the vehicle 800 begins to leave its lane, the LKA system provides steering input or braking to correct the vehicle 800.

[0168] The BSW system detects and warns the driver of vehicles in the car's blind spot. The BSW system can provide visual, audible, and / or tactile alerts to indicate that merging or changing lanes is unsafe. The system can provide additional warnings when the driver uses a turn signal. The BSW system can use a rear-facing camera and / or RADAR sensor 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0169] The RCTW system can provide visual, audible, and / or tactile notifications when an object is detected outside the range of the rear-mounted camera while the vehicle 800 is in reverse. Some RCTW systems include automatic emergency braking (AEB) to ensure that the vehicle brakes are applied to avoid a collision. The RCTW system can use one or more rear-mounted RADAR sensors 860 coupled to a dedicated processor, DSP, FPGA, and / or ASIC that is electrically coupled to driver feedback such as a display, speaker, and / or vibration component.

[0170] Conventional ADAS systems may be prone to false positive results, which may be annoying and distracting to the driver, but are typically not catastrophic because the ADAS system alerts the driver and allows the driver to decide whether a safety condition actually exists and take action accordingly. However, in the autonomous vehicle 800, in the event of conflicting results, the vehicle 800 itself must decide whether to pay attention to the results from the main computer or the auxiliary computer (e.g., the first controller 836 or the second controller 836). For example, in some embodiments, the ADAS system 838 can be a backup and / or auxiliary computer for providing perception information to the backup computer rationality module. The backup computer rationality monitor can run redundant and diverse software on hardware components to detect failures in perception and dynamic driving tasks. The output from the ADAS system 838 can be provided to the supervisory MCU. If the outputs from the main computer and the auxiliary computer conflict, the supervisory MCU must determine how to reconcile the conflict to ensure safe operation.

[0171] In some examples, the primary computer can be configured to provide a confidence score to the supervisory MCU, indicating the primary computer's confidence in the selected result. If the confidence score exceeds a threshold, the supervisory MCU can follow the direction of the primary computer regardless of whether the secondary computer provides conflicting or inconsistent results. In the event that the confidence score does not meet the threshold and the primary and secondary computers indicate different results (e.g., a conflict), the supervisory MCU can arbitrate between these computers to determine the appropriate result.

[0172] The supervisory MCU can be configured to run a neural network that is trained and configured to determine conditions under which the secondary computer provides a false alarm based, at least in part, on outputs from the primary and secondary computers. Thus, the neural network in the supervisory MCU can learn when the output of the secondary computer can be trusted and when it cannot. For example, when the secondary computer is a RADAR-based FCW system, the neural network in the supervisory MCU can learn when the FCW system is identifying a metal object that is not actually a danger, such as a drain grate or manhole cover, which triggers an alarm. Similarly, when the secondary computer is a camera-based LDW system, the neural network in the supervisory MCU can learn to disregard the LDW when a cyclist or pedestrian is present and lane departure is actually the safest strategy. In embodiments that include a neural network running on the supervisory MCU, the supervisory MCU can include at least one of a DLA or a GPU suitable for running the neural network with associated memory. In preferred embodiments, the supervisory MCU can include and / or be included as a component of the SoC 804.

[0173] In other examples, the ADAS system 838 may include an auxiliary computer that uses traditional computer vision rules to perform ADAS functions. In this way, the auxiliary computer can use classic computer vision rules (if-then), and the presence of a neural network in the supervisory MCU can improve reliability, safety, and performance. For example, diverse implementations and intentional non-identity make the entire system more fault-tolerant, especially with respect to failures caused by software (or software-hardware interface) functions. For example, if there is a software vulnerability or bug in the software running on the main computer and the non-identical software code running on the auxiliary computer provides the same overall result, the supervisory MCU can be more confident that the overall result is correct and that the vulnerability in the software or hardware on the main computer did not cause a substantial error.

[0174] In some examples, the output of the ADAS system 838 can be fed into the primary computer's perception block and / or the primary computer's dynamic driving task block. For example, if the ADAS system 838 indicates a forward collision warning due to an object immediately ahead, the perception block can use this information when identifying the object. In other examples, the secondary computer can have its own neural network that is trained and thus reduces the risk of false positives as described herein.

[0175] The vehicle 800 may also include an infotainment SoC 830 (e.g., an in-vehicle infotainment system (IVI)). Although illustrated and described as an SoC, the infotainment system may not be an SoC and may include two or more separate components. The infotainment SoC 830 may include a combination of hardware and software that can be used to provide audio (e.g., music, personal digital assistant, navigation instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), phone (e.g., hands-free calling), network connectivity (e.g., LTE, WiFi, etc.), and / or information services (e.g., a navigation system, rear parking assistance, radio data system, vehicle-related information such as fuel level, total distance covered, brake fuel level, oil level, door open / closed, air filter information, etc.) to the vehicle 800. For example, the infotainment SoC 830 may include a radio, a disc player, a navigation system, a video player, USB and Bluetooth connectivity, an onboard computer, in-vehicle entertainment, WiFi, steering wheel audio controls, hands-free voice controls, a head-up display (HUD), an HMI display 834, a telematics device, a control panel (e.g., for controlling and / or interacting with various components, features, and / or systems), and / or other components. The infotainment SoC 830 may further be used to provide information (e.g., visual and / or auditory) to a user of the vehicle, such as information from an ADAS system 838, autonomous driving information such as planned vehicle maneuvers, trajectories, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0176] The infotainment SoC 830 may include GPU functionality. The infotainment SoC 830 may communicate with other devices, systems, and / or components of the vehicle 800 via a bus 802 (e.g., a CAN bus, Ethernet, etc.). In some examples, the infotainment SoC 830 may be coupled to a supervisory MCU so that in the event of a failure of a primary controller 836 (e.g., a primary and / or backup computer of the vehicle 800), the infotainment system's GPU may perform some self-driving functions. In such an example, the infotainment SoC 830 may place the vehicle 800 in a driver-safe parking mode as described herein.

[0177] The vehicle 800 may also include an instrument cluster 832 (e.g., a digital instrument panel, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 832 may include a controller and / or a supercomputer (e.g., a separate controller or a supercomputer). The instrument cluster 832 may include a set of instruments, such as a speedometer, fuel level, oil pressure, a tachometer, an odometer, a turn indicator, a shift position indicator, a seat belt warning light, a parking brake warning light, an engine check light, airbag (SRS) system information, lighting controls, safety system controls, navigation information, etc. In some examples, information can be displayed and / or shared between the infotainment SoC 830 and the instrument cluster 832. In other words, the instrument cluster 832 may be included as part of the infotainment SoC 830, or vice versa.

[0178] Figure 8D For cloud-based servers and Figure 8A 880). Depending on the embodiment, each of the servers 878 can include any number of GPUs 884, CPUs 880, and / or PCIe switches. For example, each of the servers 878 can include eight, sixteen, thirty-two, and / or more GPUs 884.

[0179] Server 878 can receive image data from a vehicle via network 890 that represents images showing unexpected or changed road conditions, such as recently begun road construction. Server 878 can transmit neural network 892, updated neural network 892, and / or map information 894, including information about traffic and road conditions, via network 890 and to the vehicle. Updates to map information 894 can include updates to HD map 822, such as information about construction sites, potholes, curves, flooding, or other obstacles. In some examples, neural network 892, updated neural network 892, and / or map information 894 can have been generated from new training and / or data received from any number of vehicles in the environment and / or based on experience with training performed at a data center (e.g., using server 878 and / or other servers).

[0180] Server 878 can be used to train a machine learning model (e.g., a neural network) based on training data. The training data can be generated by the vehicle and / or can be generated in simulation (e.g., using a game engine). In some examples, the training data is labeled (e.g., in cases where the neural network benefits from supervised learning) and / or undergoes other preprocessing, while in other examples, the training data is not labeled and / or preprocessed (e.g., in cases where the neural network does not require supervised learning). Training can be performed according to any one or more categories of machine learning techniques, including but not limited to the following categories: supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, joint learning, transfer learning, feature learning (including principal component and cluster analysis), multilinear subspace learning, manifold learning, representation learning (including alternative dictionary learning), rule-based machine learning, anomaly detection, and any variants or combinations thereof. Once the machine learning model is trained, the machine learning model can be used by the vehicle (e.g., transmitted to the vehicle via network 890), and / or the machine learning model can be used by server 878 to remotely monitor the vehicle.

[0181] In some examples, server 878 can receive data from the vehicle and apply that data to a state-of-the-art real-time neural network for real-time intelligent reasoning. Server 878 can include a deep learning supercomputer and / or a dedicated AI computer powered by GPU 884, such as the DGX and DGX Station machines developed by NVIDIA. However, in some examples, server 878 can include the deep learning infrastructure of a data center using only CPU power.

[0182] The deep learning infrastructure of server 878 may be capable of rapid real-time inference and may use this capability to assess and verify the health of the processors, software, and / or associated hardware in vehicle 800. For example, the deep learning infrastructure may receive periodic updates from vehicle 800, such as an image sequence and / or objects that vehicle 800 has located in the image sequence (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 objects and compare them to the objects identified by vehicle 800. If the results do not match and the infrastructure concludes that the AI in vehicle 800 has malfunctioned, server 878 may transmit a signal to vehicle 800 instructing the vehicle's 800 fail-safe computer to take control, notify passengers, and complete a safe parking maneuver.

[0183] For inference, the server 878 may include a GPU 884 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT). The combination of GPU-powered servers and inference acceleration can enable real-time responses. In other examples, such as where performance is less important, CPU, FPGA, and other processor-powered servers can be used for inference.

[0184] Example computing device

[0185] Figure 9 9 is a block diagram of an example computing device 900 suitable for implementing some embodiments of the present disclosure. Computing device 900 may include an interconnect system 902 that directly or indirectly couples the following devices: memory 904, one or more central processing units (CPUs) 906, one or more graphics processing units (GPUs) 908, a communication interface 910, input / output (I / O) ports 912, input / output components 914, a power supply 916, one or more presentation components 918 (e.g., display(s)), and one or more logic units 920. In at least one embodiment, computing device(s) 900 may include one or more virtual machines (VMs), and / or any of its components may include virtual components (e.g., virtual hardware components). For non-limiting examples, one or more of GPUs 908 may include one or more vGPUs, one or more of CPUs 906 may include one or more vCPUs, and / or one or more of logic units 920 may include one or more virtual logic units. As such, computing device(s) 900 may include discrete components (e.g., a full GPU dedicated to computing device 900), virtual components (e.g., a portion of a GPU dedicated to computing device 900), or a combination thereof.

[0186] although Figure 9 The various blocks of are shown as being connected via interconnect system 902 using wires, but this is not intended to be limiting and is provided for clarity only. For example, in some embodiments, presentation component 918 (such as a display device) may be considered to be I / O component 914 (e.g., if the display is a touch screen). As another example, CPU 906 and / or GPU 908 may include memory (e.g., memory 904 may represent a storage device in addition to the memory of GPU 908, CPU 906, and / or other components). In other words, Figure 9 The computing devices are illustrative only. No distinction is made between such categories 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, as all are considered Figure 9 within the range of computing devices.

[0187] Interconnect system 902 can represent one or more links or buses, such as address bus, data bus, control bus or its combination.Interconnect system 902 can include one or more bus or link types, such as industry standard architecture (ISA) bus, extended industry standard architecture (EISA) bus, video electronics standard association (VESA) bus, peripheral component interconnect (PCI) bus, peripheral component interconnect express (PCIe) bus and / or another type of bus or link. In some embodiments, there is a direct connection between components. As an example, CPU 906 can be directly connected to memory 904. Further, CPU 906 can be directly connected to GPU 908. In the case of direct or point-to-point connection between components, interconnect system 902 can include PCIe link to perform connection. In these examples, PCI bus does not need to be included in computing device 900.

[0188] Memory 904 may include any of a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by computing device 900. Computer-readable media can include volatile and non-volatile media, as well as removable and non-removable media. By way of example and not limitation, computer-readable media can include computer storage media and communication media.

[0189] Computer storage media may include volatile and nonvolatile media and / or removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, and / or other data types. For example, memory 904 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 memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage devices 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 900. As used herein, computer storage media does not include signals themselves.

[0190] Computer storage media can 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 includes any information delivery media. The term "modulated data signal" may refer to a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, computer storage media can include wired media (such as a wired network or direct-wired connection) and wireless media (such as acoustic, RF, infrared, and other wireless media). Combinations of any of the above should also be included within the scope of computer-readable media.

[0191] The CPU 906 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. The CPUs 906 may each include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) capable of handling numerous software threads simultaneously. The CPU 906 may include any type of processor and may include different types of processors depending on the type of computing device 900 being implemented (e.g., a processor with fewer cores for a mobile device and a processor with more cores for a server). For example, depending on the type of computing device 900, the processor may be an Advanced RISC Machine (ARM) processor implemented using Reduced Instruction Set Computing (RISC) or an x86 processor implemented using Complex Instruction Set Computing (CISC). The computing device 900 may also include one or more CPUs 906 in addition to one or more microprocessors or supplemental coprocessors (such as a math coprocessor).

[0192] In addition to or in lieu of CPU(s) 906, GPU(s) 908 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. One or more of the GPUs 908 may be integrated GPUs (e.g., with one or more of the CPUs 906) and / or one or more of the GPUs 908 may be discrete GPUs. In embodiments, one or more of the GPUs 908 may be coprocessors for one or more of the CPUs 906. The GPU 908 may be used by the computing device 900 to render graphics (e.g., 3D graphics) or perform general-purpose computations. For example, the GPU 908 may be used for general-purpose computing on a GPU (GPGPU). The GPU 908 may include hundreds or thousands of cores capable of handling hundreds or thousands of software threads simultaneously. The GPU 908 may generate pixel data for an output image in response to rendering commands (e.g., rendering commands received from the CPU 906 via a host interface). The GPU 908 may include graphics memory (e.g., display memory) for storing pixel data or any other suitable data (e.g., GPGPU data). The display memory may be included as part of the memory 904. The GPU 908 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 may connect the GPUs through a switch (e.g., using NVSwitch). When combined, each GPU 908 may generate pixel data or GPGPU data for different portions or for different outputs (e.g., a first GPU for a first image and a second GPU for a second image). Each GPU may include its own memory or may share memory with other GPUs.

[0193] In addition to or in lieu of the CPU 906 and / or GPU 908, the logic unit 920 may be configured to execute at least some of the computer-readable instructions to control one or more components of the computing device 900 to perform one or more of the methods and / or processes described herein. In embodiments, the CPU(s) 906, the GPU(s) 908, and / or the logic unit(s) 920 may perform any combination of methods, processes, and / or portions thereof, either discretely or jointly. One or more of the logic units 920 may be part of and / or integrated into one or more of the CPU 906 and / or GPU 908, and / or one or more of the logic units 920 may be discrete components or otherwise external to the CPU 906 and / or GPU 908. In embodiments, one or more of logic units 920 may be a co-processor of one or more of CPUs 906 and / or one or more of GPUs 908 .

[0194] Examples of logic unit 920 include one or more processing cores and / or components thereof, such as a tensor core (TC), a tensor processing unit (TPU), a pixel vision core (PVC), a vision processing unit (VPU), a graphics processing cluster (GPC), a texture processing cluster (TPC), a streaming multiprocessor (SM), a tree transverse unit (TTU), an artificial intelligence accelerator (AIA), a deep learning accelerator (DLA), an arithmetic logic unit (ALU), an application-specific integrated circuit (ASIC), a floating point unit (FPU), an input / output (I / O) element, a peripheral component interconnect (PCI) or a peripheral component interconnect express (PCIe) element, etc.

[0195] The communication interface 910 may include one or more receivers, transmitters, and / or transceivers that enable the computing device 900 to communicate with other computing devices via an electronic communication network (including wired and / or wireless communications). The communication interface 910 may include components and functionality that enable communication over any of a number of different networks, such as a wireless network (e.g., Wi-Fi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), a wired network (e.g., via Ethernet or Wi-Fi), a low-power wide-area network (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.

[0196] The I / O ports 912 can enable the computing device 900 to be logically coupled to other devices including I / O components 914, (one or more) presentation components 918, and / or other components, some of which can be built into (e.g., integrated into) the computing device 900. Illustrative I / O components 914 include a microphone, a mouse, a keyboard, a joystick, a game pad, a game controller, a satellite dish, a scanner, a printer, a wireless device, and the like. The I / O components 914 can provide a natural user interface (NUI) that processes air gestures, voice, or other physiological input generated by the user. In some cases, the input can be transmitted to an appropriate network element for further processing. The NUI can implement any combination of voice recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition on and near the screen, air gestures, head and eye tracking, and touch recognition (as described in more detail below) associated with the display of the computing device 900. The computing device 900 may include a depth camera for gesture detection and recognition, such as a stereo camera system, an infrared camera system, an RGB camera system, touch screen technology, and combinations thereof. Additionally, computing device 900 may include an accelerometer or gyroscope that enables detection of motion (e.g., as part of an inertial measurement unit (IMU)). In some examples, computing device 900 may use the output of the accelerometer or gyroscope to render immersive augmented or virtual reality.

[0197] The power supply 916 may include a hardwired power supply, a battery power supply, or a combination thereof. The power supply 916 may provide power to the computing device 900 to enable the components of the computing device 900 to operate.

[0198] The presentation component 918 may include a display (e.g., a monitor, a touch screen, a television screen, a head-up display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. The presentation component 918 may receive data from other components (e.g., GPU 908, CPU 906, etc.) and output the data (e.g., as images, video, sound, etc.).

[0199] Sample Data Center

[0200] Figure 10 An example data center 1000 that can be used in at least one embodiment of the present disclosure is shown. The data center 1000 can include a data center infrastructure layer 1010, a framework layer 1020, a software layer 1030, and / or an application layer 1040.

[0201] like Figure 10As shown, the data center infrastructure layer 1010 may include a resource coordinator 1012, grouped computing resources 1014, and node computing resources ("node CRs") 1016(1)-1016(N), where "N" represents any complete positive integer. In at least one embodiment, the node CRs 1016(1)-1016(N) may include, but is not limited to, any number of central processing units ("CPUs") or other processors (including accelerators, field programmable gate arrays (FPGAs), graphics processors or graphics processing units (GPUs), etc.), memory devices (e.g., dynamic read-only memories), 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 of the node CRs 1016(1)-1016(N) may correspond to a server having one or more of the above-mentioned computing resources. Furthermore, in some embodiments, node CRs 1016(1)-10161(N) may include one or more virtual components, such as vGPUs, vCPUs, etc., and / or one or more of node CRs 1016(1)-1016(N) may correspond to a virtual machine (VM).

[0202] In at least one embodiment, the grouped computing resources 1014 may include individual groups of node CRs 1016 housed in one or more racks (not shown), or multiple racks housed in data centers at different geographical locations (also not shown). Individual groups of node CRs 1016 within the grouped computing resources 1014 may include grouped computing, network, memory, or storage resources that can be configured or allocated to support one or more workloads. In at least one embodiment, several node CRs 1016 including CPUs, GPUs, and / or other processors may be grouped in one or more racks to provide computing resources to support one or more workloads. One or more racks may also include any number of power modules, cooling modules, and / or network switches in any combination.

[0203] Resource coordinator 1022 may configure or otherwise control one or more node CRs 1016(1)-1016(N) and / or grouped computing resources 1014. In at least one embodiment, resource coordinator 1022 may comprise a software design infrastructure ("SDI") management entity for data center 1000. Resource coordinator 1022 may comprise hardware, software, or some combination thereof.

[0204] In at least one embodiment, Figure 10As shown, the framework layer 1020 may include a job scheduler 1032, a configuration manager 1034, a resource manager 1036, and / or a distributed file system 1038. The framework layer 1020 may include a framework that supports the software 1032 of the software layer 1030 and / or one or more applications 1042 of the application layer 1040. The software 1032 or the applications 1042 may respectively include web-based service software or applications, such as those provided by Amazon Web Services, Google Cloud, and Microsoft Azure. The framework layer 1020 may be, but is not limited to, a free and open source software web application framework (such as Apache Spark™ (hereinafter referred to as “Spark”)) that can utilize the distributed file system 1038 for large-scale data processing (e.g., “big data”). In at least one embodiment, the job scheduler 1032 may include a Spark driver to facilitate scheduling workloads supported by the different layers of the data center 1000. Configuration manager 1034 may be capable of configuring different layers, such as software layer 1030 and framework layer 1020 (which includes Spark and distributed file system 1038 for supporting large-scale data processing). Resource manager 1036 may be capable of managing clustered or grouped computing resources that are mapped to distributed file system 1038 and job scheduler 1032 or allocated to support distributed file system 1038 and job scheduler 1032. In at least one embodiment, clustered or grouped computing resources may include grouped computing resources 1014 at data center infrastructure layer 1010. Resource manager 1036 may coordinate with resource coordinator 1012 to manage these mapped or allocated computing resources.

[0205] In at least one embodiment, the software 1032 included in the software layer 1030 may include software used by at least a portion of the node CRs 1016(1)-1016(N), the grouped computing resources 1014, and / or the distributed file system 1038 of the framework layer 1020. The one or more types of software may include, but are not limited to, Internet web search software, email virus scanning software, database software, and streaming video content software.

[0206] In at least one embodiment, the applications 1042 included in the application layer 1040 may include one or more types of applications used by at least a portion of the node CRs 1016(1)-1016(N), the grouped computing resources 1014, and / or the distributed file system 1038 of the framework layer 1020. The one or more types of applications may include, but are not limited to, any number of genomic applications, cognitive computing, 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 conjunction with one or more embodiments.

[0207] In at least one embodiment, any of the configuration manager 1034, the resource manager 1036, and the resource coordinator 1012 can implement any number and type of self-modification actions based on any amount and type of data acquired in any technically feasible manner. The self-modification actions can save the data center operator of the data center 1000 from making potentially poor configuration decisions and potentially avoiding underutilized and / or poorly performing portions of the data center.

[0208] According to one or more embodiments described herein, data center 1000 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. For example, the machine learning model(s) may be trained by computing weight parameters according to a neural network architecture using the software and / or computing resources described above with respect to data center 1000. In at least one embodiment, a trained or deployed machine learning model corresponding to one or more neural networks may be used to infer or predict information using the resources described above with respect to data center 1000 using weight parameters computed using one or more training techniques, such as, but not limited to, those described herein.

[0209] In at least one embodiment, data center 1000 may use a CPU, an application-specific integrated circuit (ASIC), a GPU, an FPGA, and / or other hardware (or virtual computing resources corresponding thereto) to perform training and / or inference using the aforementioned resources. In addition, one or more of the software and / or hardware resources described above may be configured to allow a user to train or perform inference services on information, such as image recognition, speech recognition, or other artificial intelligence services.

[0210] Sample network environment

[0211] A network environment suitable for implementing embodiments of the present 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 configured to: Figure 9 The data center 1000 may be implemented on one or more instances of the computing device(s) 900 - for example, each device may include similar components, features, and / or functionality of the computing device(s) 900. In addition, in the case of implementing a backend device (e.g., a server, NAS, etc.), the backend device may be included as part of the data center 1000, an example of which is described herein with respect to FIG. Figure 10 Describe in more detail.

[0212] The components of the network environment can communicate with each other via a network, which can be wired, wireless, or both. The network can include multiple networks or one of multiple networks. For example, the network can include 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 the Public Switched Telephone Network (PSTN)), and / or one or more private networks. In the case where the network includes a wireless telecommunications network, components such as base stations, communication towers, or even access points (and other components) can provide wireless connectivity.

[0213] Compatible network environments may include one or more peer-to-peer network environments (in which case the server may not be included in the network environment) and one or more client-server network environments (in which case one or more servers may be included in the network environment). In a peer-to-peer network environment, the functionality described herein for the server may be implemented on any number of client devices.

[0214] In at least one embodiment, the network environment may include one or more cloud-based network environments, distributed computing environments, combinations thereof, and the like. The 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. The framework layer may include software supporting the software layer and / or a framework for one or more applications at the application layer. The software or application may include network-based service software or applications, respectively. In an embodiment, one or more client devices may use network-based service software or applications (e.g., by accessing the service software and / or application via one or more application programming interfaces (APIs)). The framework layer may be, but is not limited to, a free and open source software network application framework that can use a distributed file system for large-scale data processing (e.g., "big data").

[0215] A cloud-based network environment can 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). Any of these different functions can be distributed across multiple locations from a central or core server (e.g., one or more data centers that can be distributed across a state, region, country, global, etc.). If the connection to the user (e.g., client device) is relatively close to an edge server, the core server can assign at least a portion of the functionality to the edge server. A cloud-based network environment can 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).

[0216] The client device(s) may include the Figure 9 At least some of the components, features, and functionality of the described example computing device(s) 900. By way of example and not limitation, the client device may be implemented as a personal computer (PC), a laptop computer, a mobile device, a smartphone, a tablet computer, a smartwatch, a wearable computer, a personal digital assistant (PDA), an MP3 player, a virtual reality headset, a global positioning system (GPS) or device, a video player, a camera, a surveillance device or system, a vehicle, a boat, a spacecraft, a virtual machine, a drone, a robot, a handheld communication device, a hospital device, a gaming device or system, an entertainment system, a vehicle computer system, an embedded system controller, a remote control, an appliance, a consumer electronic device, a workstation, an edge device, any combination of the depicted devices, or any other suitable device.

[0217] The present disclosure can be described in the general context of machine-usable instructions or computer code executed by a computer or other machine such as a personal digital assistant or other handheld device, including computer-executable instructions such as program modules. Generally, program modules including routines, programs, objects, components, data structures, etc. refer to code that performs a specific task or implements a specific abstract data type. The present disclosure can be practiced in a variety of system configurations, including handheld devices, consumer electronics, general-purpose computers, more specialized computing devices, etc. The present disclosure can also be practiced in a distributed computing environment where tasks are performed by remote processing devices linked through a communication network.

[0218] As used herein, the phrase "and / or" with respect to two or more elements should be interpreted as referring to only one element or 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. In addition, "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.

[0219] The subject matter of the present disclosure is described in detail herein to meet statutory requirements. However, the description itself is not intended to limit the scope of the present disclosure. On the contrary, the present inventors have contemplated that the claimed subject matter may also be embodied in other ways to include steps that are different from the steps described herein in conjunction with other current or future technologies, or combinations of similar steps. Moreover, although the terms "step" and / or "block" may be used herein to imply different elements of the method employed, these terms should not be interpreted as implying any particular order among or between the various steps disclosed herein, unless the order of the steps is explicitly described.

Claims

1. A processor, comprising: One or more circuits for: generating a depth flow image representing one or more changes in depth values between corresponding pixels of at least two consecutive range images; transferring data corresponding to one or more neighboring pixels of the depth stream image using a belief propagation algorithm to update one or more values associated with the one or more neighboring pixels of the depth stream image; calculating one or more 2.5D motion vectors in image space corresponding to movement of the one or more adjacent pixels of the depth stream image between the at least two consecutive range images based at least in part on the one or more updated values; as well as A scene flow representation is generated by converting the one or more 2.5D motion vectors into corresponding one or more 3D motion vectors in a 3D world space.

2. The processor according to claim 1, wherein the processor is included in at least one of the following: control systems for autonomous or semi-autonomous machines; Perception systems for autonomous or semi-autonomous machines; a system for performing simulation operations; Systems for performing deep learning operations; Systems implemented using edge devices; Systems implemented using robots; A system for merging one or more virtual machines VM; A system implemented at least in part in a data center; or A system implemented at least in part using cloud computing resources.

3. The processor of claim 1 , further comprising a processing circuit for: determining one or more positions of one or more objects based at least in part on the scene flow representation; and One or more operations are performed based at least in part on the one or more positions of the one or more objects. 4 . The processor of claim 1 , further comprising processing circuitry for performing one or more operations based at least in part on the scene stream.

5. The processor of claim 1 , wherein the at least two continuous range images are generated using corresponding lidar point clouds. The processor according to claim 1 , wherein: When it is determined that a pair of the other one or more adjacent pixels do not correspond to the same object, data corresponding to the pair of the other one or more adjacent pixels is blocked from being transferred.

7. A method comprising: generating a depth flow image based at least in part on a first lidar range image generated using one or more lidar sensors at a first moment in time and a second lidar range image generated using the one or more lidar sensors at a second moment in time after the first moment in time, the depth flow image comprising one or more pixels labeled with one or more depth flow values; transferring data corresponding to at least one of the one or more pixels of the depth stream image to generate an updated depth stream image, the transferred data indicating, at least in part, respective depth stream values of the one or more pixels and respective costs of the one or more pixels; as well as Based at least in part on the updated depth stream image, one or more 2.5D motion vectors representing relative pixel positions between the first lidar range image and the second lidar range image are calculated.

8. The method of claim 7, wherein the first lidar range image and the second lidar range image are generated using data representing one or more lidar point clouds.

9. The method of claim 7, wherein the first lidar range image is transformed to the same coordinate system as the second lidar range image based at least in part on the calculated ego-motion.

10. The method of claim 7, wherein the first lidar range image and the second lidar range image each represent at least one of reflectivity information, intensity information, depth information, time of flight (ToF) information, return information, or classification information. 11 . The method of claim 7 , wherein the method further comprises converting the one or more 2.5D motion vectors into a 3D space to generate one or more 3D motion vectors.

12. The method of claim 11, wherein the one or more 3D motion vectors represent a scene flow representation between the first lidar range image and the second lidar range image.

13. The method of claim 7, wherein the data corresponding to the at least one pixel of the one or more pixels is transferred using a belief propagation algorithm.

14. The method of claim 13, wherein the depth stream image comprises a first pixel adjacent to a second pixel, and the method further comprises: determining, based at least in part on depth data from the second lidar data, that the first pixel corresponds to a second object different from the first object, the first object corresponding to the second pixel; as well as The communicating of data corresponding to the first pixel is prevented based at least in part on the determining.

15. A system comprising: one or more processing units; as well as One or more memory units storing instructions that, when executed by the one or more processing units, cause the one or more processing units to perform the following operations, including: generating a depth flow image representing one or more changes in one or more depth values between corresponding pixels of at least two consecutive range images; transferring data corresponding to one or more neighboring pixels of the depth stream image using a belief propagation algorithm to update one or more values associated with a pixel of the depth stream image; calculating one or more image-space 2.5D motion vectors between the at least two consecutive range images corresponding to movement of the one or more adjacent pixels of the depth stream image based at least in part on the one or more updated values; and The image space 2.5D motion vector is converted into a 3D motion vector in a 3D world space to generate a scene flow representation.

16. The system of claim 15, wherein the system is included in at least one of the following: control systems for autonomous or semi-autonomous machines; Perception systems for autonomous or semi-autonomous machines; a system for performing simulation operations; Systems for performing deep learning operations; Systems implemented using edge devices; Systems implemented using robots; A system for merging one or more virtual machines VM; A system implemented at least in part in a data center; or A system implemented at least in part using cloud computing resources.

17. The system of claim 15, wherein the operations further comprise: determining one or more positions of one or more objects based at least in part on the scene flow; as well as One or more operations are performed based at least in part on the one or more positions of the one or more objects.

18. The system of claim 15, wherein the operations further comprise performing one or more operations based at least in part on the scene flow representation.

19. The system of claim 15, wherein the at least two continuous range images are generated using corresponding lidar point clouds.

20. The system of claim 15, wherein at least one message is blocked from being communicated between a pair of one or more other adjacent pixels when it is determined that the pair of one or more other adjacent pixels do not correspond to the same object.

Images