Multi-view Deep Neural Networks for LiDAR Perception

The multi-view deep neural network for LiDAR perception enhances object detection and classification in autonomous vehicles by processing LiDAR data through multiple stages, addressing visibility and computational inefficiencies in traditional systems.

JP7586689B2Active Publication Date: 2024-11-19NVIDIA CORP
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2020189118
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2020-06-29
Filing Date
2020-11-13
Publication Date
2024-11-19
Estimated Expiration
2040-11-13

AI Technical Summary

Technical Problem

Conventional LiDAR perception systems for autonomous vehicles face challenges in accurately detecting and classifying objects due to limitations in panoramic segmentation DNNs for perspective views and BEV object detection, which struggle with visibility and computational inefficiencies in 3D convolution.

Method used

A multi-view deep neural network approach that processes LiDAR data through multiple stages, including a first stage for class division in a perspective view and a second stage for class division in a top-down view, regressing instance geometry to enhance object detection and classification accuracy.

Benefits of technology

The multi-view approach improves the detection and classification of objects, particularly vulnerable road users, by mitigating the drawbacks of traditional methods, ensuring high recall and providing a simple and effective method for object detection and regression of dimensions and orientation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007586689000001
    Figure 0007586689000001
  • Figure 0007586689000002
    Figure 0007586689000002
  • Figure 0007586689000003
    Figure 0007586689000003
Patent Text Reader

Abstract

To provide a multi-view deep neural network (DNN) for LiDAR perception.SOLUTION: A deep neural network (DNN) is used to detect objects from sensor data of a three-dimensional (3D) environment. For example, a multi-view perception DNN includes multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example DNN includes a first stage that performs class segmentation in a first view (perspective view) and a second stage that performs class segmentation and / or regresses instance geometry in a second view. The DNN outputs are processed to generate 2D / 3D bounding boxes and class labels for detected objects in the 3D environment. The techniques described herein are used to detect and classify animate objects and / or parts of an environment, and the detection and classification are provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.SELECTED DRAWING: Figure 4
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 62 / 936,080, filed November 15, 2019, and U.S. Provisional Patent Application No. 62 / 938,852, filed November 21, 2019, which are incorporated by reference in their entireties. [Background technology]

[0002] Designing a system that safely drives a vehicle autonomously without supervision is extremely challenging. An autonomous vehicle should at least have the ability to perform as the functional equivalent of a cautious driver, utilizing a perception and action system that has an incredible ability to identify and react to moving and stationary obstacles in a complex environment to avoid collisions with other objects or structures along the vehicle's path. Thus, the ability to detect instances of animated objects (e.g., cars, pedestrians, etc.) and other parts of the environment is often critical for an autonomous driving perception system. Traditional perception methods often rely on camera or LiDAR sensors to detect objects in the environment, and various approaches have been developed to perform LiDAR and camera perception using deep neural networks (DNNs). Such classes of DNNs include DNNs that perform panoramic segmentation of camera images in a perspective view and DNNs that perform top-down or "Bird's Eye View" (BEV) object detection from the LiDAR point cloud. However, these traditional approaches have several drawbacks.

[0003] For example, conventional panoramic segmentation DNNs generally perform class and instance segmentation of images in a perspective view (e.g., RBG imagery from a forward-facing camera or LiDAR range scan). FIG. 1 is a diagram of an exemplary LiDAR range scan with perspective views and segmented classes from a panoramic segmentation of a LiDAR range scan. In FIG. 1, each LiDAR input (range scan) is shown with a corresponding classification mask showing an exemplary segmentation output (segmented classes). For simplicity, the segmented instances are omitted in FIG. 1. FIG. 2 is a diagram of an exemplary panoramic segmentation of a camera image with a perspective view. In FIG. 2, the top image is the input image to be segmented, the middle image is a classification mask showing the segmented classes superimposed on the input image, and the bottom image is an instance mask showing the segmented instances superimposed on the input image.

[0004] Panorama segmentation in perspective views often works well for certain classes, such as pedestrians and cyclists, due to the feature geometry (e.g., consistent structure) of objects within a certain class. However, panorama segmentation often has challenges when evaluating features that are not visible from the view of the view being analyzed. For example, panorama segmentation may be able to detect pedestrians in a forward-looking image, but panorama segmentation DNNs often struggle to accurately predict the 3D bounding box or BEV two-dimensional (2D) bounding box of the detected object. Similarly, panorama segmentation DNNs using perspective views often struggle to accurately detect objects with salient features that are not visible from the view of the view being analyzed. As such, conventional panorama segmentation DNNs have limited accuracy in predicting object classification, object instances, dimensions, and orientations.

[0005] Conventional DNNs performing object detection from BEV (top-down) projection images of LiDAR point clouds often detect a single class, including cars, trucks, buses, pedestrians, and / or cyclists, in predicting BEV 2D bounding boxes. FIG. 3 is a diagram of an exemplary object detection performed on a top-down projection image of a LiDAR point cloud. Because a top-down view of a pedestrian or cyclist often looks similar to a top-down view of other objects such as poles, tree trunks, or shrubs, DNNs performing BEV object detection often struggle to accurately detect these objects. As such, conventional DNNs performing BEV object detection have limited accuracy in predicting object classification, dimension, and orientation. A potential solution to this problem is to use 3D convolutions on 3D voxelized volumes. However, 3D convolutions are computationally very expensive and require processing a significant amount of empty voxel space in the search volume, which can lead to significant inefficiencies. Summary of the Invention

[0006] Embodiments of the present disclosure relate to LiDAR perception for autonomous machines using deep neural networks (DNNs). For example, the systems and methods described herein use object detection techniques to identify or detect instances of obstacles (e.g., cars, trucks, pedestrians, cyclists, etc.) and other objects such as environmental parts for use by autonomous vehicles, semi-autonomous vehicles, robots, and / or other object types. In contrast to conventional systems such as those described above, the systems of the present disclosure may include multiple constituent DNNs or stages chained together that sequentially process different views of a 3D environment. An exemplary multi-view perception DNN may include a first stage that performs class classification in a first view (e.g., a perspective view) and a second stage that performs class classification and / or regresses instance geometry in a second (e.g., a top-down) view.

[0007] For example, the first stage can extract classification data (e.g., confidence maps, segmentation masks, etc.) from a LiDAR range image or an RGB image. The extracted classification data can be transformed to a second view of the environment, for example, by labeling corresponding 3D locations (e.g., identified by corresponding pixels in the LiDAR range image) with the extracted classification data and projecting the labeled 3D locations to the second view. In some embodiments, geometry data (e.g., height data) of objects in 3D space can be obtained from sensor data (e.g., by projecting the LiDAR point cloud onto one or more height maps in a top-down view) and / or images of the 3D space (e.g., by unprojecting the images into world space and projecting to a top-down view). The extracted classification data and / or geometry data can be stacked and fed to a second stage of a DNN, which can extract classification data (e.g., class confidence maps) and / or regress various types of information about the detected objects, such as their position, geometry, and / or orientation in the second view. The DNN output can be processed to generate 2D and / or 3D bounding boxes and class labels for objects detected in the 3D environment.

[0008] As such, the techniques described herein can be used to detect and classify animated objects and / or parts of the environment, and these detections and classifications can be provided to an autonomous vehicle driving stack to enable safe planning and control of the autonomous vehicle. Unlike conventional approaches, by processing multiple views of sensor data sequentially with a multi-view perceptual DNN, the present techniques can preserve the benefits of processing each view individually while mitigating potential drawbacks. Using the techniques described herein, motorbikes, cyclists, pedestrians, and other vulnerable road users (VRUs) or objects can be detected with high recall. Furthermore, embodiments of the present disclosure can provide a simple and effective method for the detection and classification of objects and the regression of their dimensions and orientations, where conventional methods struggle.

[0009] The present systems and methods for multi-view LiDAR perception are described in detail below with reference to the accompanying drawings. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram of an example LiDAR range scan with segmented classes from a perspective view and a panoramic segmentation of the LiDAR range scan. [Diagram 2] 1 is a diagram of an example panoramic section of a camera image with a perspective view. [Diagram 3] FIG. 13 illustrates an example object detection performed on a top-down projection image of a LiDAR point cloud. [Figure 4] 1 is a data flow diagram illustrating an example process of an object detection system according to some embodiments of the present disclosure. [Diagram 5] 1 is a data flow diagram illustrating an example process for preprocessing sensor data for a machine learning model in an object detection system, in accordance with some embodiments of the present disclosure. [Figure 6] FIG. 2 is a diagram of an example multi-view perception machine learning model, in accordance with some embodiments of the present disclosure. [Figure 7] 1 is a data flow diagram illustrating an example post-processing process for generating object detections in an object detection system, according to some embodiments of the present disclosure. [Figure 8] 1 is a diagram of an example data flow through an example multi-view perceptual machine learning model, in accordance with some embodiments of the present disclosure. [Figure 9] 1 is a flow diagram illustrating a method for multi-view object detection using sensor data, according to some embodiments of the present disclosure. [Figure 10] 1 is a flow diagram illustrating a method for perspective and top-down view object detection using LiDAR data, according to some embodiments of the present disclosure. [Figure 11] 1 is a flow diagram illustrating a method for multi-view object detection including projections of labeled sensor data, according to some embodiments of the present disclosure. [Figure 12] FIG. 1 is a diagram of an example technique for annotating sensor data from different sensors, according to some embodiments of the present disclosure. [Figure 13] FIG. 13 is an illustration of an example annotation of car and truck classes in camera space, according to some embodiments of the present disclosure. [Figure 14] 1 is a diagram of an example annotation of a pedestrian class in camera space, according to some embodiments of the present disclosure. [Figure 15] 1 is a diagram of an example annotation of a top-down bounding box in a LiDAR space, in accordance with some embodiments of the present disclosure. [Figure 16A] 1 is an illustration of an example autonomous vehicle, according to some embodiments of the present disclosure. [Figure 16B] 16B is an illustration of camera positions and fields of view for the example autonomous vehicle of FIG. 16A, according to some embodiments of the present disclosure. [Figure 16C] FIG. 16B is a block diagram of an example system architecture of the example autonomous vehicle of FIG. 16A in accordance with some embodiments of the present disclosure. [Figure 16D]FIG. 16B is a system diagram of communication between a cloud-based server and the example autonomous vehicle of FIG. 16A, according to some embodiments of the present disclosure. [Figure 17] FIG. 1 is a block diagram of an example computing device suitable for use in implementing some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Systems and methods are disclosed for multi-view LiDAR perception for autonomous machines using deep neural networks (DNNs). For example, the systems and methods described herein use object detection techniques to identify or detect instances of obstacles (e.g., cars, trucks, pedestrians, cyclists, etc.) and other objects, e.g., parts of the environment, for use by autonomous vehicles, semi-autonomous vehicles, robots, and / or other object types.

[0012] The present disclosure is described with respect to an exemplary autonomous vehicle 1600 (an example of which is described herein with respect to FIGS. 16A-16D and alternatively referred to herein as “vehicle 1600” or “ego vehicle 1600”), but this is not intended to be limiting. For example, the systems and methods described herein may be used by non-autonomous vehicles, semi-autonomous vehicles (e.g., in one or more advanced driver assistance systems (ADAS)), robots, warehouse vehicles, off-road vehicles, flying vessels, boats, and / or other vehicle types. Additionally, the present disclosure may be described with respect to autonomous driving, but this is not intended to be limiting. For example, the systems and methods described herein may be used in other technology fields, such as robotics (e.g., path planning for robots), aviation systems (e.g., path planning for drones or other aircraft), marine systems (e.g., path planning for boats or other water vessels), and / or localization, path planning, and / or other processes.

[0013] At a high level, DNNs can be used to detect objects from LiDAR data and / or other sensor data capturing a three-dimensional (3D) environment. In some embodiments, a DNN can include multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example multi-view perceptual DNN can include a first stage that performs class classification in a first view (e.g., a perspective view) and a second stage that performs class classification and / or regresses instance geometry in a second (e.g., top-down) view.

[0014] In some examples, the input to the DNN can be formed from LiDAR data (e.g., LiDAR range images, projections of LiDAR point clouds, etc.) and / or data from other sensors (e.g., images from any number of cameras), and the first stage can extract classification data (e.g., class confidence data, such as confidence maps for any number of classes) from the input. The confidence map and / or synthetic partition mask can partition (and thus represent) a first view (e.g., a perspective view) of the 3D space. The confidence map and / or synthetic partition can be projected onto a second (e.g., top-down) view to generate transformed classification data for processing by the next DNN stage. For example, the extracted classification data can be used to label corresponding 3D locations (e.g., identified by LiDAR range images), and the labeled 3D locations (e.g., labeled LiDAR range images) can be reprojected onto a second view of the environment.

[0015] In some embodiments, geometry data (e.g., height data) of objects in 3D space may be obtained from LiDAR data (e.g., by projecting the LiDAR point cloud onto one or more height maps in a top-down view) and / or images of the 3D space (e.g., by unprojecting the images onto world space and projecting onto a top-down view). The transformed classification data and geometry data may be stacked and fed into a second stage of a DNN that may extract classification data (e.g., class confidence data, such as confidence maps for any number of classes) and / or regress various types of information, e.g., position, geometry, and / or orientation, about the detected objects. The DNN output may be processed to generate 2D and / or 3D bounding boxes and class labels of the detected objects in the 3D environment.

[0016] In general, a multi-view perception DNN can accept as input some representation of LiDAR data and / or other sensor data of a 3D environment. In some embodiments, to form the input to the DNN, raw LiDAR detections from the environment around an ego object (e.g., a moving vehicle) can be pre-processed into a format that the DNN understands. Specifically, LiDAR data (e.g., raw LiDAR detections from an ordered or unordered LiDAR point cloud) can be accumulated, transformed into a single coordinate system (e.g., centered on the ego actor), ego-motion corrected (e.g., to the last known location of the ego actor), and / or projected to form a LiDAR range image.

[0017] In some circumstances, forming a range scan image, such as a LiDAR range image, may result in some sensor data being lost. For example, when accumulating detections over time while the ego body is moving, when accumulating detections from different sensors mounted at different locations on the ego body (i.e., capturing sensor data from different views of the scene), and / or when collapsing sensor data into a range image with insufficient resolution to represent adjacent sensor data, it may be possible for reflections from multiple objects in a scene to be binned together into one range scan pixel. In some embodiments, when reflections are binned together in a pixel of a range image, the reflection with the closest range may be represented in the range image and the other reflections may be dropped. Additionally or alternatively, the resolution of the range image may be selected in such a way as to reduce sensor data loss and / or limit degradation of accuracy. For example, the altitude (or vertical resolution) of the range image may be set to correspond with the number of horizontal scan lines of the sensor capturing sensor data (e.g., one row of pixels in the range image for each corresponding LiDAR sensor scan line). The width (or horizontal resolution) of the range image may be set based on the horizontal resolution of the sensor capturing the sensor data. In general, the horizontal resolution may be a design choice: a lower resolution may have fewer collisions but may be easier to process (and vice versa).

[0018] In some embodiments, the LiDAR range image may be fed to a multi-view perception DNN (e.g., the first stage of the DNN). Additionally or alternatively, the LiDAR range image and / or other sensor data may be stacked into corresponding channels of an input tensor and fed to the multi-view perception DNN. In any case, the DNN may include multiple stages chained together that sequentially process data from multiple views to predict classification data and / or object instance data for detected objects in a 3D environment. These outputs may be processed into 2D and / or 3D bounding boxes and class labels for the detected objects. In an example application of an autonomous vehicle, the DNN may be used to predict one or more bounding boxes (e.g., 2D bounding boxes, 3D bounding boxes in a top-down view) for each detected object on a road or sidewalk, a class label for each detected object, and a 2D mask (e.g., in a top-down view) that bounds drivable space, sidewalks, buildings, trees, poles, and other stationary environmental parts. In some embodiments, the 2D bounding box in the top-down view may be fitted to a 3D bounding box by deriving the box height from the predicted object instance data.

[0019] In embodiments where a multi-view perceptual DNN includes a series of multiple stages, the different stages may be trained together or separately. In some embodiments, the stages may be trained together by performing a transformation from the output of a first stage (first view) to the input to a second stage (second view) using a differentiable operation (e.g., differentiable reprojection). The training data may be obtained by annotating data from multiple sensors in the sensor setup. Since data may be obtained from different sensors at different frequencies, in some embodiments, a particular sensor (e.g., a LiDAR sensor) may be used as a reference sensor. For each frame of sensor data from the reference sensor (e.g., for each frame of LiDAR data), a set of sensor data may be curated by identifying a frame of sensor data from each of the other sensors in the sensor setup that is closest in time to the frame of sensor data from the reference sensor. This set of sensor data (e.g., the image obtained most recently in time to T from each of the multiple cameras in the sensor setup plus the frame of LiDAR data at timestamp T) may be referred to as a set of curated sensor data at timestamp T. For each set of curated sensor data, data from each sensor may be labeled independently from data from other sensors. In some embodiments, object detection and tracking may be applied to track the movement of annotated objects from frame to frame over time. As such, annotation tracking may be used to track objects from frame to frame (e.g., using persistent identifiers for the annotated objects). In some embodiments, object trajectories and / or detections from sensor data from a particular sensor may be linked to corresponding object trajectories and / or detections of the same object from sensor data from a different sensor. Annotations and / or links between different types of sensor data of the same object may be generated manually and / or automatically and used to generate training data for a multi-view perception DNN.

[0020] As such, the techniques described herein can be used to detect and classify animated objects and / or parts of an environment, and these detections and classifications can be provided to an autonomous vehicle driving stack to enable safe planning and control of the autonomous vehicle. Unlike conventional approaches, by processing multiple views of sensor data sequentially with a multi-view perceptual DNN, the present techniques can preserve the benefits of processing each view individually while mitigating potential drawbacks. Using the techniques described herein, motorcycles, bicycles, pedestrians, and other vulnerable road user (VRU) objects can be detected with high recall. Additionally, embodiments of the present disclosure can provide a simple and effective method for detecting and classifying objects, as well as regressing their dimensions and orientations, where conventional methods struggle.

[0021] Exemplary Object Detection System Referring to FIG. 4, FIG. 4 is a data flow diagram illustrating an example process of an object detection system according to some embodiments of the present disclosure. It should be understood that this and other arrangements described herein are described merely as examples. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, groupings of functions, etc.) may be used in addition to or instead of those shown, and some elements may be omitted entirely. Furthermore, many of the elements described herein are functional entities that may be implemented as separate or distributed components or in conjunction with other components, and in any suitable combination and location. Various functions described herein as being performed by an entity may be implemented by hardware, firmware, and / or software. For example, various functions may be implemented by a processor executing instructions stored in a memory.

[0022] At a high level, the process 400 may include a machine learning model 408 configured to detect objects, e.g., animated objects and / or instances of portions of the environment, based on sensor data 402 of a three-dimensional (3D) environment. The sensor data 402 may be preprocessed 404 into input data 406 having a format understood by the machine learning model 408, which may be fed to the machine learning model 408 to detect objects 416 in the 3D environment. In some embodiments, the machine learning model 408 may include multiple constituent machine learning models or stages chained together to sequentially process different views of the 3D environment. The machine learning model 408 may predict a representation of class confidences (e.g., class confidence data 410) of the detected objects and / or a representation of object instance data (e.g., instance regression data 412) of the detected objects, which may be post-processed 414 into object detections 416 including bounding boxes, closed polylines, or other bounding shapes that identify the location, size, and / or orientation of the detected objects. The object detection 416 may correspond to obstacles, stationary parts of the environment, and / or other objects around the autonomous vehicle and may be used by the control components of the autonomous vehicle (e.g., the controller 1636, the ADAS system 1638, the SOC 1604, the software stack 422, and / or other components of the autonomous vehicle 1600 of FIGS. 16A-16D) to assist the autonomous vehicle in performing one or more operations within the environment (e.g., obstacle avoidance, route planning, mapping, etc.).

[0023] In general, object detection may be performed using sensor data 402 from any number and type of sensors, such as, but not limited to, LiDAR sensors, RADAR sensors, cameras, and / or other sensor types, such as those described below with respect to the autonomous vehicle 1600 of Figures 16A-16D. For example, the sensors 401 may include one or more sensors 401 of an ego object or ego actor (e.g., the LiDAR sensor 1664 of the autonomous vehicle 1600 of Figures 16A-16D), which may be used to generate sensor data 402 representing objects in a 3D environment around the ego object.

[0024] Taking LiDAR data as an example, object detection may be performed using LiDAR data (e.g., sensor data 402) from one or more LiDAR sensors (e.g., sensor 401). In general, a LiDAR system may include a transmitter that emits pulses of laser light. The emitted light waves reflect off certain objects and materials, and one of the LiDAR sensors may detect these reflections and their reflection characteristics, such as direction, azimuth, height, range (e.g., beam time of flight), intensity, reflectivity, signal-to-noise ratio (SNR), and / or the like. The reflections and reflection characteristics may depend on objects in the environment, speed, materials, sensor mounting location and orientation, etc. Firmware associated with the LiDAR sensor may be used to control the LiDAR sensor to capture and / or process sensor data 402, such as reflection data from the sensor's field of view.

[0025] In general, the sensor data 402 may include raw sensor data, LiDAR point cloud data, and / or reflectance data that has been processed into some other format. For example, the reflectance data may be combined with location and orientation data (e.g., from GNSS and IMU sensors) to form a point cloud that represents detected reflectances from the environment. Each detection in the point cloud may include metadata about the detection, such as one or more of the detection's three-dimensional location and reflectance characteristics. Some non-limiting examples of LiDAR sensors include Velodyne HDL / VLS series and Ouster OS1 / OS2 series LiDAR sensors, and non-limiting example operating (e.g., scanning) frequencies may be >=5 Hz. Although these examples describe sensor data 402 as LiDAR data, sensor data 402 may additionally or alternatively include sensor data from other sensors, such as RADAR data (e.g., RADAR point clouds), image data (e.g., RBG imagery from one or more cameras mounted around the ego actor), and / or other types.

[0026] The sensor data 402 may be pre-processed 404 into a format understood by the machine learning model 408. For example, in embodiments where the sensor data 402 includes LiDAR data (and / or other data, such as RADAR data), the LiDAR data (and / or other data) may be accumulated, transformed into a single coordinate system (e.g., centered on the ego actor / vehicle), ego-motion corrected (e.g., to the last known location of the ego actor / vehicle), and / or projected to form a projected image of a desired size (e.g., spatial dimensions). For example, the (accumulated, ego-motion corrected) LiDAR point cloud may be projected to form a LiDAR range image having a perspective view. Any suitable perspective projection may be used (e.g., spherical, cylindrical, pinhole, etc.). In some cases, the type of projection may depend on the type of sensor. By way of non-limiting example, for a rotating sensor, a spherical or cylindrical projection may be used. In some embodiments, for a time-of-flight (ToF) camera (e.g., flash LiDAR), pinhole projection may be used. In one example, the (accumulated, ego-motion corrected) RADAR point cloud may be orthogonally projected to form an overhead image with a desired ground sampling distance. In either case, the projected image (e.g., LiDAR range image) and / or other reflectance data may be stored and / or encoded into an appropriate representation (e.g., input data 406) that may serve as input to machine learning model 408.

[0027] 5 is a data flow diagram illustrating an example process for pre-processing 404 sensor data 402 for a machine learning model 408 in an object detection system according to some embodiments of the present disclosure. The sensor data 402 may be accumulated 510 (which may include transforming to a single coordinate system), ego-motion corrected 520, for example, in multiple channels that store different reflectance characteristics, and / or encoded 530 into a suitable representation such as a projection image (e.g., LiDAR range image) and / or a tensor.

[0028] More specifically, sensor data 402, such as LiDAR data, may be accumulated 510 from multiple sensors, such as some or all of the multiple peripheral LiDAR sensors 1664, from different locations of the autonomous vehicle 1600 and may be transformed to a single vehicle coordinate system (e.g., vehicle-centered). Additionally or alternatively, the sensor data 402 may be accumulated 510 over time to increase the density of the accumulated sensor data. Sensor detections may be accumulated over any desired window of time (e.g., 0.5 seconds, 1 second, 2 seconds, etc.). The size of the window may be selected based on the sensor and / or application (e.g., a smaller window may be selected for a noisy application such as a highway scenario). As such, each input to the machine learning model 408 may be generated from accumulated detections from each window of time from a rotating window (e.g., from a duration ranging from t window size to the present). Each window to evaluate may be incremented by any suitable step size, which may not necessarily correspond to a window size. Thus, each successive input to the machine learning model 408 may be based on successive windows, which may not need to overlap.

[0029] In some embodiments, ego-motion correction 520 may be applied to the sensor data 402. For example, accumulated detections may be ego-motion corrected to the latest known vehicle location. More specifically, the location of an older detection may be communicated to the latest known location of a moving vehicle using the known motion of the vehicle to estimate where the older detection would be located (e.g., relative to the vehicle's current location) at a desired time (e.g., the current time). The result may be a set of accumulated, ego-motion corrected sensor data 402 (e.g., LiDAR point cloud) for a particular time slice.

[0030] In some embodiments, the (accumulated, ego-motion corrected) sensor data 402 may be encoded 530 into a suitable representation, such as a projected image, which may include multiple channels storing different features, such as reflectance properties. More specifically, the accumulated, ego-motion corrected detections may be projected to form a projected image of a desired size (e.g., spatial dimension). Any desired view of the environment, e.g., top-down, frontal, perspective, and / or other, may be selected for the projected image. In one example, a LiDAR point cloud may be projected to form a LiDAR range image having a perspective view of the environment (e.g., spherical, cylindrical, pinhole), and the LiDAR range image may be used as input data 406 to a machine learning model 408. In some embodiments, images having the same or different views may be generated, with each image being input into a separate channel of the machine learning model 408. As a non-limiting example, different sensors 401 (whether of the same type or different types) may be used to generate image data (e.g., LiDAR range images, camera images, etc.) having the same (e.g., perspective) view of the environment in a common image space, and image data from different sensors 401 or sensor modalities may be stored in separate channels of a tensor. These are intended as examples only and other variations may be implemented within the scope of this disclosure.

[0031] Since image data may be evaluated as input to the machine learning model 408, there may be a tradeoff between prediction accuracy and computational needs. As such, a desired spatial dimension of the projected image may be selected as a design choice. Additionally or alternatively, to reduce data loss due to reduced image resolution, the dimension of the projected image may be based on characteristics of the corresponding sensor 401 that captured the sensor data 402. As a non-limiting example, the elevation (or vertical resolution) of the LiDAR range image may be set to correspond to the number of horizontal scan lines (e.g., one row of pixels in the range image per scan line of the corresponding LiDAR sensor) of the sensor capturing the sensor data 402, and the width (or horizontal resolution) of the LiDAR range image may be set based on the horizontal resolution of the sensor 401 that captures the sensor data 402.

[0032] In some embodiments, a projection image may include multiple layers, with pixel values ​​in different layers storing different reflectance characteristics. In some embodiments, for each pixel that bins sensor data representing multiple reflections, a set of features may be calculated, determined, or otherwise selected from the reflectance characteristics of the reflections (e.g., direction, azimuth, height, range, intensity, reflectivity, SNR, etc.). In some cases, when sensor data representing multiple reflections are binned together at a pixel of a projection image (e.g., a range image), the sensor data representing one of the reflections (e.g., the reflection with the closest range) may be represented in the projection image, and the sensor data representing the other reflections may be dropped. For example, in a range image having a pixel that bins multiple reflections together, the pixel may store a range value corresponding to the reflection with the closest range. Additionally or alternatively, when there are multiple reflections binned together within a pixel, thereby forming a tower of points, a particular feature of that pixel may be calculated by aggregating (e.g., using a standard deviation, average, etc.) the corresponding reflectance characteristics of the multiple overlapping reflections. In general, any given pixel may have multiple associated feature values ​​that may be stored in corresponding channels of a tensor. In any event, the sensor data 402 may be encoded 530 into various types of input data 406 (e.g., a projection image such as a LiDAR range image, a tensor encoding the projection image, and corresponding reflectance properties), which may serve as input to a machine learning model 408.

[0033] At a high level, the machine learning model 408 may detect objects, such as instances of obstacles, stationary parts of the environment, and / or other objects represented in the input data 406 (e.g., LiDAR range images, camera images, and / or other sensor data stacked in corresponding channels of the input tensor). For example, the machine learning model 408 may extract object instance data, such as position, geometry, and / or orientation data (e.g., instance regression data 412), and / or classification data (e.g., class confidence data 410), representing the detected objects in the 3D environment. The classification data and object instance data may be post-processed 414 to generate class labels, as well as bounding boxes, closed polylines, or other bounding shapes that identify the positions, geometries, and / or orientations of the 2D and / or 3D detected object instances.

[0034] In some embodiments, the machine learning model 408 may be implemented using a DNN, such as a convolutional neural network (CNN). Although certain embodiments are described with the machine learning model 408 being implemented using a neural network, and more specifically, a CNN, this is not intended to be limiting. For example, and without limitation, the machine learning model 408 may include any type of machine learning model, such as linear regression, logistic regression, decision trees, support vector machines (SVMs), naive Bayes, k-nearest neighbors (Knn), K-means clustering, random forests, dimensionality reduction algorithms, gradient boosting algorithms, machine learning models using neural networks (e.g., autoencoders, convolutional, recurrent, perceptrons, long / short term memory (LSTM), Hopfield, Boltzmann, deep belief, deconvolution, generative adversarial, liquid state machines, etc.), and / or other types of machine learning models.

[0035] In some examples, the machine learning model 408 may include a common trunk (or stream of layers) with several heads (or at least partially discrete streams of layers) for predicting different outputs based on the input data 406. For example, the machine learning model 408 may include, but is not limited to, a feature extractor (e.g., DNN, encoder / decoder) including convolutional layers, pooling layers, and / or other layer types, where the output of the feature extractor is provided as input to a first head for predicting classification data and a second head for predicting the position, geometry, and / or orientation of a detected object. In some instances, the first head and the second head may receive parallel inputs, thereby producing different outputs for similar input data.

[0036] In general, the machine learning model 408 may include feature extractors configured to evaluate images having different views of a 3D environment. For example, the machine learning model 408 may include separate feature extractors in multiple stages chained together to sequentially process data from multiple views of a 3D environment. For example, the machine learning model 408 may include a first stage having a first feature extractor configured to extract classification data from an image having a first view of the environment (e.g., a perspective view), and the output of the first feature extractor may be transformed to a second view of the environment (e.g., a top-down view) and fed to a second feature extractor that may include several heads and a common trunk that extracts different outputs such as classification data and object instance data of detected objects. Additionally or alternatively, multiple images may be generated with different views, each image may be fed to a separate side-by-size feature extractor, and the latent space tensor output by the separate feature extractors may be combined to form classification data and / or object instance data. These architectures are intended as examples, and other architectures are contemplated within the scope of this disclosure.

[0037] Turning now to FIG. 6, FIG. 6 is a diagram of an example multi-view perceptual machine learning model (e.g., an example implementation of the machine learning model 408 of FIG. 4) according to some embodiments of the present disclosure. In the example of FIG. 6, the machine learning model 408 is shown as an example architecture including multiple stages chained together for sequential processing. In a first stage, the encoder / decoder 605 can extract class confidence data 610 (e.g., one or more confidence maps in a first view, such as a perspective view) from the input data 406, which can be transformed 615 to a second view to form transformed class confidence data 630 (e.g., one or more confidence maps in a second view, such as a top-down view). This data can be supplemented with geometry data 640 of the object represented in the sensor data 402 (e.g., a representative dimension of the object geometry in a direction orthogonal to the two dimensions of the second view, e.g., height). The transformed class confidence data 630 (e.g., one or more confidence maps in a top-down view) and / or geometry data 640 (e.g., one or more height maps), respectively, may be encoded 645 (e.g., stacked into corresponding channels of a tensor) and fed to a second stage including an encoder / decoder trunk 650 connected to a class confidence head 655 and an instance regression head 660, which may extract class confidence data 610 (e.g., one or more confidence maps in a top-down view) and instance regression data 412 (e.g., object instance data such as position, geometry, and / or orientation).

[0038] The Encoder / Decoder 605 may be implemented using encoder and decoder components with skip connections (e.g., similar to a feature pyramid network, U-Net, etc.). For example, the Encoder / Decoder 605 may accept input data 406, such as LiDAR range images and / or RBG images, and apply various convolution, pooling, and / or other types of operations to extract class confidence data 610 for any number of supported classes. In an example implementation, the Encoder / Decoder 605 may include an encoding (reduction) path and a decoding (expansion) path. Along the reduction path, each decomposition may include any number of layers (e.g., convolution, dilated convolution, inception block, etc.) and a downsampling operation (e.g., max pooling). Along the expansion path, each decomposition may include any number of layers (e.g., deconvolution, upsampling followed by convolution, and / or other types of operations). In the augmentation path, each decomposition of a feature map may be upsampled and concatenated (e.g., in the depth dimension) with a feature map of the same decomposition from the reduction path. In this example, corresponding decompositions of the reduction and augmentation paths may be connected with a skip connection, which may be used to append or concatenate feature maps from the corresponding decompositions.

[0039] The output of the Encoder / Decoder 605 may be class confidence data 610 for any number of supported classes (e.g., one channel per class). Examples of supported classes may include vehicles (e.g., cars, buses, trucks, etc.), vulnerable road users (e.g., motorcycles, bicycles, pedestrians, etc.), environmental parts (e.g., drivable spaces, sidewalks, buildings, trees, poles, etc.), subclasses thereof (e.g., pedestrians walking), any combination thereof, and / or others. For example, the class confidence data 610 may include a representation of one or more confidence maps (e.g., one per class). As a non-limiting example, the Encoder / Decoder 605 may output a tensor with N channels corresponding to the N classes (e.g., one confidence map per channel). Thus, each pixel in the tensor may store a depthwise pixel value that represents the likelihood, score, or logit that the pixel is part of the corresponding class for each channel. In some implementations, the sum of the depthwise pixel values ​​may be normalized to some value (e.g., 1). In some embodiments, the predicted values ​​may be used to generate a different class classification mask for each class (channel) and / or may be collapsed into a single composite classification mask in which each pixel contains a class label (e.g., represented by a different integer). In the embodiment shown in FIG. 6, the encoder / decoder 605 outputs class confidence data 610 (e.g., one or more confidence maps), but in other embodiments, the encoder / decoder 605 may additionally or alternatively output other types of classification data (e.g., an N-class classification mask that stores a binary value for each pixel, a composite classification mask that stores the most likely class label for each pixel).

[0040] In general, the output of the encoder / decoder 605 (e.g., class confidence data 610 or other classification data) may be transformed 615 from a first view of an environment to a second view of the environment. For example, the classification data extracted by the encoder / decoder 605 may be used to label 620 corresponding 3D locations in the environment, and the labeled 3D locations may be projected 625 onto the second view.

[0041] As an example of a confidence map for a particular class, the confidence map may have a spatial dimension that corresponds to the input to the encoder / decoder 605 (e.g., a LiDAR range image), and the confidence map may include a classification value (e.g., a likelihood, score, or logit) for each pixel. In some cases, the classification value may be mapped to a known 3D location identified by the corresponding sensor data 402 and / or input data 406. For example, the corresponding input LiDAR range image may have a known correspondence between the range scan pixel and a corresponding point in a LiDAR point cloud (LiDAR detection), which may have a known 3D location. Thus, the classification value from the predicted confidence map may be associated with the 3D location of the LiDAR detection represented by the corresponding range scan pixel in the input LiDAR range image.

[0042] In another example, assume that the input to the Encoder / Decoder 605 includes a representation of an RGB image generated by a camera, and the Encoder / Decoder 605 classifies each pixel of the RGB image by generating one or more classification values ​​for each pixel. The classification values ​​may be associated with 3D positions identified from some other sensor data, e.g., LiDAR or RADAR detections, or 3D positions from a 3D representation of the environment, such as a 3D map of the environment. For example, LiDAR or RADAR data captured at the same time slice as the input RGB image may be projected to form a range image having the same view as the input RGB image. In this case, the classification values ​​from the predicted confidence map may be associated with 3D positions in a manner similar to the previous example, by identifying the 3D positions of the sensor detections (e.g., points in a point cloud) represented by the corresponding range scan pixels in the range image. In another example, image data generated by a sensor (e.g., an RGB image generated by a camera) with a known orientation and position within a 3D representation of the environment (e.g., a 3D map or some other world space) may not be projected into world space to identify the 3D position of an object in world space that corresponds to each pixel. These are just a few examples and other variations may be implemented within the scope of this disclosure.

[0043] As such, 3D positions from the sensor data (e.g., sensor data 402) for a corresponding time slice or from a corresponding portion of the 3D representation of the environment may be labeled 620 with classification data (e.g., classification values, labels) extracted by the encoder / decoder 605. The labeled 3D positions may be projected 625 into a second view of the environment, for example, by orthogonally projecting the labeled 3D positions to form a projected image (e.g., an overhead image having a top-down view) with a desired spatial dimension and ground sampling distance. In one example implementation, the semantically labeled range image may be converted into a top-down representation.

[0044] In some cases, projecting 625 the labeled 3D locations may simply involve folding the labeled 3D locations into a plane or otherwise projecting (e.g., orthogonally) the labeled 3D locations into the projection image (e.g., by folding the point cloud into a plane by discarding the z-values ​​of the labeled points). In some cases, projecting 625 the labeled 3D locations may involve multiple points being binned together at the pixels of the projection image. As such, when multiple points are binned together at the pixels of the projection image (e.g., a top-down image), any technique may be used to select or otherwise represent one or more of the points (e.g., select one of the points such as the highest or lowest point). In some cases, filtering may be applied (e.g., by leaving out points that are more than 3 or 4 meters). Each pixel of the resulting projection image with which there is a projected 3D location may store the extracted classification data (e.g., classification value or label) with which the 3D location was labeled.

[0045] Thus, the result may be transformed classification data (e.g., transformed confidence maps or partition masks) that represent the classification data extracted in the second view of the environment. In general, any number of confidence maps (e.g., N confidence maps that store a classification value for each pixel in a corresponding class), individual class partition masks (e.g., N class partition masks that store a binary value indicating each pixel in a corresponding class), and / or a composite partition mask (e.g., having a single channel that stores a label of the most likely class for each pixel) may be transformed 615. If the classification data being transformed has N channels, the transformed classification data may have N corresponding channels. In the example shown in FIG. 6, the class confidence data 610 (e.g., one or more confidence maps in a perspective view) may be transformed to form transformed class confidence data 630 (e.g., one or more confidence maps in a top-down view). However, in other embodiments, other types of classification data (e.g., N class classification masks that store a binary value for each pixel, a composite classification mask that stores the most likely class label for each pixel) may additionally or alternatively be converted 615.

[0046] In general, the transformed classification data 630 may represent a second view of the environment, and thus the transformed classification data may reveal object characteristics (e.g., position, geometry, orientation) in the two dimensions represented by the second view. In some embodiments, geometry data 640 representing object characteristics (e.g., orthogonal to the second view) may be generated 635 from the sensor data (e.g., sensor data 402) of a corresponding time slice. For example, sensor data representing 3D positions of detected objects in the environment may be sampled or otherwise processed to represent characteristics of the detected objects in a particular dimension (e.g., an orthogonal dimension), for example, by taking one or more slices of the sensor data in a particular dimension. Taking as an example a LiDAR or RADAR point cloud, where the second view is a top-down view, one or more slices of the point cloud may be taken in the height dimension to generate geometry data 640 having a planar representation of the detected objects, such as a minimum and maximum height map. In this example, each pixel of such a map may represent a column in the top-down view. When sensor data representing multiple objects in a 3D environment are binned together at a single pixel (e.g., multiple points from a point cloud are in a column represented by a pixel), any type of sampling or statistical metric may be used to represent the collection of data (e.g., the points in the column). For example, each pixel may store the minimum height of all points in the column (e.g., a minimum height map), the maximum height of all points in the column (e.g., a maximum height map), the median height of all points in the column, the average height of all points in the column, the variance of the heights of all points in the column, and / or the like. As such, one or more slices of geometry data 640 (e.g., height data) having a planar representation of detected objects in the environment may be generated 635.

[0047] In another example, geometry data 640 of objects in an environment may be generated 635 from image data (e.g., RGB images) generated by a sensor (e.g., a camera). For example, the known orientation and position of the sensor that captured the image data may be used to project the image data onto a 3D representation of the environment (e.g., a 3D map or some other world space) and identify 3D positions of objects in world space that correspond to each pixel. In this case, one or more slices of the identified 3D positions may be taken to generate 635 the geometry data 640 (one or more height maps).

[0048] In general, the transformed classification data (e.g., transformed class confidence data 630) and / or geometry data (e.g., geometry data 640) may be encoded 645 or otherwise packaged into some suitable representation in the encoder / decoder trunk 650. For example, in an embodiment where the transformed class confidence data 630 includes N transformed confidence maps and the geometry data 640 includes M height maps, the transformed class confidence data 630 and the geometry data 640 may be encoded 645 into a tensor having N channels for the N confidence maps and M channels for the M height maps. This is merely one example and any suitable representation of the transformed classification data and / or geometry data may be implemented.

[0049] 6 includes an encoder / decoder trunk 650, a class confidence head 655, and an instance regression head 660. The second stage can extract features from the transformed classification data and / or representations of the geometry data (e.g., tensors with M+N channels) and can perform class distinction and / or regress instance geometries in a second view.

[0050] The encoder / decoder trunk 650 may be implemented using encoder and decoder components with skip connections (e.g., similar to a feature pyramid network, U-Net, etc.). For example, the encoder / decoder trunk 650 may accept representations of transformed class confidence data 630 and / or geometry data 640 and may apply various convolution, pooling, and / or other types of operations to extract features into some latent space. In FIG. 6, the encoder / decoder trunk 650 is shown in an exemplary implementation including an encoding (reduction) path and an exemplary decoding (expansion) path. Along the reduction path, each decomposition may include any number of layers (e.g., convolution, dilated convolution, inception block, etc.) and downsampling operations (e.g., max pooling). Along the expansion path, each decomposition may include any number of layers (e.g., deconvolution, upsampling followed by convolution, and / or other types of operations). In the expanding path, each decomposition of the feature map may be upsampled and concatenated (e.g., in the depth dimension) with the feature map of the same decomposition from the contracting path. In this example, corresponding decompositions of the contracting and expanding paths may be connected with a skip connection that can be used to append or concatenate feature maps from corresponding decompositions. As such, the encoder / decoder trunk 650 can extract features into several latent space tensors that may be input to the class confidence head 655 and the instance regression head 660.

[0051] The class confidence head 655 may include any number of layers 655A, 655B, 655C (e.g., convolution, pooling, softmax, and / or other types of classifiers, etc.) that predict classification data from the output of the encoder / decoder trunk 650. For example, the class confidence head 655 may include a channel (e.g., a stream of classifiers in addition to layers) for each class of object to be detected (e.g., vehicle, car, truck, vulnerable road user, pedestrian, cyclist, motorcycle, drivable space, sidewalk, building, tree, pole, subclasses thereof, any combination thereof, etc.) such that the class confidence head 655 extracts classification data (e.g., class confidence data 410) in any suitable format. For example, the class confidence head 655 may predict a confidence map that represents an estimated confidence level that a particular object is present (regardless of class), a separate confidence map for each class, and / or the like. In some embodiments, the class confidence data 410 predicted by the class confidence head 655 can take the form of a multi-channel tensor, where each channel can be thought of as a heat map that stores the classification value (e.g., likelihood, score, or logit) that each pixel belongs to the class corresponding to the channel.

[0052] The instance regression head 660 may include any number of layers 660A, 660B, 660C (e.g., convolution, pooling, softmax, and / or other types of classifiers, and / or other types of operations) that predict object instance data (e.g., location, geometry, and / or orientation of detected objects) from the output of the encoder / decoder trunk 650. The instance regression head 660 may include N channels (e.g., streams of classifiers in addition to layers), where each channel regresses a particular type of information about a detected object instance, such as where the object is located (e.g., a dx / dy vector pointing to the center of the object), the height of the object, the width of the object, the orientation of the object (e.g., rotation angles such as sine and / or cosine), some statistical measure of it (e.g., min, max, mean, median, variance, etc.), and / or the like. As a non-limiting example, the instance regression head 660 may include separate dimensions identifying the x dimension of the center of the detected object, the y dimension of the center of the detected object, the width of the detected object, the height of the detected object (e.g., displacement from the ground), the sine of the orientation of the detected object (e.g., rotation angle in 2D image space), the cosine of the orientation of the detected object, and / or other types of information. These types of object instance data are intended merely as examples, and other types of object information may be regressed additionally or alternatively. The instance regression head 660 may include a separate regression channel for each class, or one set of channels for all classes. In some implementations, the instance regression data 412 predicted by the instance regression head 660 may take the form of a multi-channel tensor, where each channel may include floating point numbers that regress a particular type of object information, such as a particular object dimension.

[0053] As such, the machine learning model 408 may predict multi-channel classification data (e.g., class confidence data 410) and / or multi-channel object instance data (e.g., instance regression data 412) from a particular input (e.g., input data 406). Some possible training techniques are described in more detail below. In operation, the output of the machine learning model 408 may be post-processed (e.g., decoded) to generate bounding boxes, closed polylines, or other bounding shapes that identify the location, geometry, and / or orientation of the detected object instances. For example, when the machine learning model 408 predicts the class confidence data 410 and / or instance regression data 412 for a particular view of the environment (e.g., a top-down view), a bounding box, closed polyline, or other bounding shape may be identified for that view (e.g., in the same image space as the input to the second stage of the machine learning model 408). In some embodiments, because object instance data may be noisy and / or may yield multiple candidates, boundary shapes may be generated using non-maximum suppression, density-based spatial clustering of application with noise (DBSCAN), and / or other features.

[0054] FIG. 7 is a data flow diagram illustrating an example post-processing process 414 for generating object detections 416 in an object detection system according to some embodiments of the present disclosure. In this example, the post-processing process 414 includes instance decoding 710 and filtering and / or clustering 720. In general, the instance decoding 710 can identify 2D and / or 3D candidate bounding boxes (or other bounding shapes) (e.g., for each object class) based on the object instance data (e.g., position, geometry, and / or orientation data) from a corresponding channel of the instance regression data 412 and / or a confidence map or mask from a corresponding channel of the classification data for that class (e.g., class confidence data 410). More specifically, the predicted confidence map and the predicted object instance data can specify information about the detected object instance, such as where the object is located, object height, object width, object orientation, and / or the like. This information can be used to identify candidate object detections (e.g., candidates with unique center points, object height, object width, object orientation, and / or the like). The result may be a set of 2D and / or 3D candidate bounding boxes (or other bounding shapes) for each object class.

[0055] Various types of filtering and / or clustering 720 may be applied to remove duplicates and / or noise from the candidate bounding boxes (or other bounding shapes) for each object class. For example, in some embodiments, duplicates may be removed using non-maximum suppression. Non-maximum suppression may be used when two or more candidate bounding boxes have associated confidence values ​​that indicate the candidate bounding boxes may correspond to the same object instance. In such an instance, the highest confidence value of an object instance may be used to determine which candidate bounding box should be used for that object instance, and non-maximum suppression may be used to remove or suppress other candidates.

[0056] For example, each candidate bounding box (or other bounding shape) may be associated with a corresponding confidence / likelihood value associated with one or more corresponding pixels from a corresponding channel of the class confidence data 410 for the class being evaluated (e.g., using the confidence / likelihood value of a representative pixel such as a center pixel, using an average or some other composite value calculated over the candidate region, etc.). Thus, candidate bounding shapes with confidence / likelihood of being a member of the object class below some threshold (e.g., 50%) may be filtered out. Additionally or alternatively, the candidate bounding box (or other shape) with the highest confidence / likelihood score for a particular class may be assigned an instance ID, a metric such as intersection of unions (IoU) may be calculated for each of the other candidates in the class, and candidates with IoU above some threshold may be filtered out to remove duplicates. The process may then be repeated, assigning an instance ID to the candidate with the highest confidence / likelihood score, removing duplicates, and repeating until no candidates remain. The process may be repeated for each of the other classes to remove duplicate candidates.

[0057] In some embodiments, a clustering technique such as Density-Based Spatial Clustering with Noise (DBSCAN) may be used to remove duplicate candidate boundary shapes. For example, the candidate boundary shapes may be clustered (e.g., the centers of the candidate boundary shapes may be clustered), candidates within each cluster may be determined to correspond to the same object instance, and duplicate candidates from each cluster may be removed.

[0058] As such, the extracted classification data and / or object instance data may be decoded (e.g., via instance decoding 710), filtered, and / or clustered (e.g., via filtering and / or clustering 720) to identify bounding boxes, closed polylines, or other bounding shapes of detected objects within each particular class (e.g., based on data from corresponding channels of class confidence data 410 and instance regression data 412). A class label may be applied to each identified bounding shape based on the particular class being evaluated (e.g., based on a known mapping between channels and class labels). In some cases, 2D bounding shapes may be determined first, and a third dimension (e.g., height) of each of the 2D bounding shapes may be inferred from the extracted object instance data (e.g., based on a particular regression channel of object height, based on separate regression channels of minimum and maximum values ​​in the height dimension, etc.). As such, 2D and / or 3D bounding shapes and class labels may be identified for detected objects in the environment.

[0059] In summary, the machine learning model 408 can accept a representation of sensor data having a first view, such as a LiDAR range image, perform segmentation on the representation of the sensor data to extract classification data, convert the classification data to a representation having a second view, and perform segmentation and instance regression on the second view to extract classification data and / or object instance data. The extracted classification data and / or object instance data can be post-processed to generate class labels and 2D and / or 3D bounding boxes, closed polylines, or other bounding shapes that identify the position, size, and / or orientation of the detected object instances in the projected image.

[0060] 8 is a diagram of an example data flow through an example multi-view perception machine learning model, according to some embodiments of the present disclosure. In FIG. 8, a LiDAR range image 810 is input to a first stage of a neural network (e.g., encoder / decoder 605 of FIG. 6), which segments the LiDAR range image to generate a segmented LiDAR range image 820. The segmented LiDAR range image 820 is transformed to a top-down view 830, stacked with height data, and fed through a second stage of the neural network (e.g., encoder / decoder trunk 650, class confidence head 655, and instance regression head 660 of FIG. 6). Note that classified regions of the segmented LiDAR range image 820 (e.g., drivable space 825) have been transformed to corresponding regions in the top-down view 830 (e.g., transformed drivable space 835). The second stage of the neural network extracts classification data and object instance data, which are post-processed to generate bounding boxes for the detected objects.

[0061] After the position, geometry, orientation, and / or class label of an object instance are determined, the 2D pixel coordinates defining the object instance may be converted to 3D world coordinates for use with the corresponding class label by the autonomous vehicle in performing one or more actions (e.g., obstacle avoidance, lane keeping, lane changing, route planning, mapping, etc.). In some embodiments, a low-level LiDAR perception stack that does not use DNNs may process sensor data to detect objects in parallel to the machine learning model 408 (e.g., for redundancy). As such, returning to FIG. 4, the object detection 416 (e.g., bounding boxes, closed polylines, or other bounding shapes) may be used by the control components of the autonomous vehicle 1600 shown in FIGS. 16A-16D, such as the autonomous driving software stack 422 executing on one or more components of the vehicle 1600 (e.g., SoC 1604, CPU 1618, GPU 1620, etc.). For example, the vehicle 1600 may use this information (e.g., obstacle instances) to navigate, plan, or otherwise perform one or more actions within the environment (e.g., obstacle avoidance, lane keeping, lane changing, merging, splitting, etc.).

[0062] In some examples, object detection 416 may be used by one or more layers of an autonomous driving software stack 422 (alternatively referred to herein as “driving stack 422”). Driving stack 422 may include a sensor manager (not shown), a perception component (e.g., corresponding to a perception layer of driving stack 422), a world model manager 426, a planning component 428 (e.g., corresponding to a planning layer of driving stack 422), a control component 430 (e.g., corresponding to a control layer of driving stack 422), an obstacle avoidance component 432 (e.g., corresponding to an obstacle or collision avoidance layer of driving stack 422), an actuation component 434 (e.g., corresponding to an actuation layer of driving stack 422), and / or other components corresponding to additional and / or alternative layers of driving stack 422. In some instances, process 400 may be performed by a perception component, which may provide layers of driving stack 422 to the world model manager, as described in more detail herein.

[0063] The sensor manager can manage and / or abstract sensor data 402 from sensors of the vehicle 1600. For example, with reference to FIG. 16C, the sensor data 402 can be generated by a RADAR sensor 1660 (e.g., permanently, at intervals, based on certain conditions). The sensor manager can receive sensor data 402 from sensors in different formats (e.g., sensors of the same type can output sensor data in different formats) and can be configured to convert the different formats to a unified format (e.g., for each sensor of the same type). As a result, other components, features, and / or functionality of the autonomous vehicle 1600 can use the unified format, thereby simplifying processing of the sensor data 402. In some instances, the sensor manager can use the unified format to apply control back to the sensors of the vehicle 1600, for example, to set frame rate or to perform gain control. The sensor manager may also update sensor packets or communications corresponding to the sensor data with a timestamp to help inform the processing of the sensor data by the various components, features, and functionality of the autonomous vehicle control system.

[0064] The world model manager 426 may be used to generate, update, and / or define a world model. The world model manager 426 may use information generated by and received from the perception components (e.g., location of detected obstacles) of the driving stack 422. The perception components may include obstacle perception devices, path perception devices, wait perception devices, map perception devices, and / or other perception components. For example, the world model may be defined, at least in part, based on the affordances of obstacles, paths, and wait states that may be perceived in real-time or near real-time by the obstacle perception devices, path perception devices, wait perception devices, and / or map perception devices. The world model manager 426 may continually update the world model based on newly generated and / or received input (e.g., data) from the obstacle perception devices, path perception devices, wait perception devices, map perception devices, and / or other components of the autonomous vehicle control system.

[0065] The world model can be used to help inform the driving stack 422 to the planning component 428, the control component 430, the obstacle avoidance component 432, and / or the actuation component 434. The obstacle perception device can perform obstacle perception, which can be based on where the vehicle 1600 is allowed or able to drive (e.g., based on the location of a drivable path defined by avoiding detected obstacles) and how fast the vehicle 1600 can drive without colliding with an obstacle (e.g., an object such as a structure, entity, vehicle, etc.) detected by sensors of the vehicle 1600 and / or the machine learning model 408.

[0066] The path perception device can perform path perception, such as by perceiving the nominal paths available in a particular situation. In some instances, the path perception device can further consider lane changes for path perception. The lane graph can represent one or more paths available to the vehicle 1600 and can be as simple as a single path on a highway on-ramp. In some instances, the lane graph can include paths to a desired lane and / or indicate available changes onto a highway (or other road type), or can include nearby lanes, lane changes, forks, turns, cloverleaf interchanges, merges, and / or other information.

[0067] The wait perception device can be responsible for determining constraints of the vehicle 1600 as a result of rules, conventions, and / or practical considerations. For example, the rules, conventions, and / or practical considerations may relate to traffic signals, multi-way stops, yield ahead, merging, toll booths, gates, police or other emergency personnel, road workers, stopped buses or other vehicles, one-way bridge arbitration, ferry entrances, etc. Thus, the wait perception device can be utilized to identify potential obstacles and to implement one or more controls (e.g., slowing down, stopping, etc.) that may not be possible relying solely on the obstacle perception device.

[0068] The map perception device may include mechanisms by which movements are discriminated and, in some instances, for determining the particular instance of which conventions apply to a particular locale. For example, the map perception device may determine from data representing a previous drive or journey that a particular intersection does not allow U-turns at a particular time of day, that the electronic signs indicating lane directionality change depending on the time of day, that two traffic signals in close proximity (e.g., slightly offset from one another) are associated with different roads, that in Rhode Island, a first passenger vehicle waiting to turn left at a traffic light will violate the law by turning in front of oncoming traffic when the light turns green, and / or other information. The map perception device may inform the vehicle 1600 of static or stationary infrastructure objects and obstacles. The map perception device may also generate information for the wait perception device and / or the path perception device, such as to determine which lights at an intersection must be green for the vehicle 1600 to take a particular path.

[0069] In some instances, information from the map perception device can be sent, transmitted, and / or provided to a server (e.g., to a map manager of server 1678 of FIG. 16D ), and information from the server can be sent, transmitted, and / or provided to the map perception device and / or localization manager of vehicle 1600. The map manager can include a cloud mapping application located remotely from vehicle 1600 and accessible by vehicle 1600 via one or more networks. For example, the map perception device and / or localization manager of vehicle 1600 can communicate with the map manager of the server and / or one or more other components or functions to inform the map perception device and / or localization manager of the past and current driving or travel of vehicle 1600, as well as the past and current driving or travel of other vehicles. The map manager can provide mapping output (e.g., map data) that can be localized by the localization manager based on the particular location of the vehicle 1600, and the localized mapping output can be used by the world model manager 426 to generate and / or update a world model.

[0070] The planning component 428 may include a route planner, a lane planner, a motion planner, and a motion selector, among other components, features, and / or functionalities. The route planner may use information from the map perception device, the map manager, and / or the localization manager, among other information, to generate a planned path, which may consist of GNSS waypoints (e.g., GPS waypoints), 3D world coordinates (e.g., Cartesian, polar, etc.) indicating coordinates relative to an origin of the vehicle 1600, and the like. The waypoints may represent a particular distance into the future of the vehicle 1600, e.g., a number of city blocks, a number of kilometers, a number of feet, a number of inches, a number of miles, etc., that may be used as a target for the lane planner.

[0071] The lane planner can use as input a target point and direction in a lane graph (e.g., a lane graph from a path perception device), object poses in the lane graph (e.g., from a localization manager), and / or a distance into the future from a route planner. The target point and direction can be mapped to a best matching drivable point and direction in the lane graph (e.g., based on GNSS and / or compass orientation). A graph search algorithm can be run on the lane graph from the current edge in the lane graph to find the shortest path to the target point.

[0072] The motion planner can determine the feasibility of basic motions of the vehicle 1600, such as staying in the lane or changing lanes to the left or right, so that feasible motions can be matched with the most desirable motions output from the lane planner. For example, if a desired motion is determined to be unsafe and / or available, a default motion can be selected instead (e.g., when a desired motion or lane change is unsafe, the default motion can be to stay in the lane).

[0073] The control component 430 can follow the trajectory or path (lateral and longitudinal) received from the motion selector of the planning component 428 (e.g., based on the object detection 416) as closely as possible and within the capabilities of the vehicle 1600. The control component 430 can use hard feedback to address unplanned events or movements that were not modeled and / or anything that causes a deviation from the ideal (e.g., unexpected delays). In some instances, the control component 430 can use a forward prediction model that receives the control as an input variable and produces a prediction that can be compared to the desired state (e.g., compared to the desired lateral and longitudinal paths requested by the planning component 428). Controls that minimize the deviation can be determined.

[0074] Although the planning component 428 and the control component 430 are shown separately, this is not intended to be limiting. For example, in some embodiments, the delineation between the planning component 428 and the control component 430 may not be precisely defined. As such, at least some of the components, features, and / or functionality attributed to the planning component 428 may be associated with the control component 430, and vice versa. This may also be true for any of the separately shown components of the driving stack 422.

[0075] The obstacle avoidance component 432 can assist the autonomous vehicle 1600 in avoiding collisions with objects (e.g., moving and stationary objects). The obstacle avoidance component 432 can include computational mechanisms at the “root level” of obstacle avoidance and can act as the “survival brain” or “reptilian brain” of the vehicle 1600. In some instances, the obstacle avoidance component 432 can be used independent of the components, features, and / or functionality of the vehicle 1600 that are required to obey traffic rules and drive courteously. In such instances, the obstacle avoidance component can ignore traffic laws, rules of the road, and courteous driving norms to ensure that no collisions occur between the vehicle 1600 and objects. As such, the obstacle avoidance layer can be a separate layer from the road layer rules, which can ensure that the vehicle 1600 is only performing safe actions from an obstacle avoidance perspective. On the other hand, the road layer rules can ensure that the vehicle obeys road traffic laws and customs and adheres to legal conventional right-of-way (as described above).

[0076] In some instances, the drivable path and / or object detection 416 may be used by the obstacle avoidance component 432 in determining control or action to take. For example, the drivable path may provide instructions to the obstacle avoidance component 432 of where the vehicle 1600 may maneuver without colliding with objects, structures, and / or the like, or at least where static structures may not be present.

[0077] In non-limiting examples, the obstacle avoidance component 432 may be implemented as a separate, discrete feature of the vehicle 1600. For example, the obstacle avoidance component 432 may operate separately from (e.g., in parallel with, before, and / or after) the planning layer, the control layer, the actuation layer, and / or other layers of the driving stack 422.

[0078] As such, the vehicle 1600 can use this information (e.g., as a road edge, or a barrier) to navigate, plan, or otherwise perform one or more actions within the environment (e.g., lane keeping, lane changing, merging, splitting, etc.).

[0079] 9-11, each block of the methods 900, 1000, and 1100 described herein includes computational processes that may be performed using any combination of hardware, firmware, and / or software. For example, various functions may be performed by a processor executing instructions stored in a memory. The methods may also be implemented as computer usable instructions stored on a computer storage medium. The methods may be provided by a standalone application, a service, or a hosted service (standalone or in combination with another hosted service), or a plug-in to another product, to name a few. Additionally, the methods 900, 1000, and 1100 are described with respect to the object detection system described herein, as an example. However, the methods may additionally or alternatively be performed by any one system or any combination of systems, including but not limited to those described herein.

[0080] FIG. 9 is a flow diagram illustrating a method 900 for multi-view object detection using sensor data, according to some embodiments of the disclosure. The method 900 includes, at block B 902, determining first data representing a first view of an environment from sensor data from at least one sensor in the environment. For example, the sensor 401 of FIG. 4 (e.g., the LiDAR sensor 1664 of the autonomous vehicle 1600 of FIGS. 16A-16D) may be used to generate sensor data 402 representing objects in a 3D environment around the sensor 401. The sensor data 402 may be pre-processed 404 to form a projection image of a desired size (e.g., spatial dimension). For example, the (accumulated, ego-motion corrected) LiDAR point cloud may be projected to form a LiDAR range image having a perspective view. The projection image (e.g., LiDAR range image) and / or other reflection data may be stored and / or encoded into an appropriate representation (e.g., input data 406) of the machine learning model 408.

[0081] The method 900 includes, at block B904, extracting classification data representing one or more classifications in the first view based at least on the first data using one or more neural networks (NNs). For example, the input data 406 of FIG. 4 may serve as input to a machine learning model 408, which may include multiple constituent machine learning models or stages chained together to sequentially process different views of the 3D environment. In an exemplary first stage, the encoder / decoder 605 of FIG. 6 may extract classification data (e.g., one or more confidence maps, one or more segmentation masks) in the first view, such as a perspective view, from the input data 406. In the example shown in FIG. 6, the class confidence data 610 may represent one or more confidence maps that store pixel values ​​representing the likelihood, score, or logit that each pixel is part of the corresponding class of each map.

[0082] The method 900 includes generating transformed classification data representing one or more classifications in a second view of the environment based at least on projecting the one or more classifications from the first view onto the second view at block B906. For example, in the example shown in Figure 6, the classification data may correspond to class confidence data 610 (which may represent one or more confidence maps in the first view, such as a perspective view), and the class confidence data 610 may be transformed 615 to the second view to form transformed class confidence data 630 (e.g., one or more confidence maps in the second view, such as a top-down view).

[0083] The method 900 includes, at block B 908, generating, using the one or more NNs, second data representing one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data. For example, in the embodiment shown in FIG. 6, the transformed classification data may correspond to the transformed class confidence data 630. The transformed class confidence data 630 (e.g., one or more confidence maps in a top-down view) and / or the geometry data 640 (e.g., one or more height maps) may be encoded 645 (e.g., stacked into corresponding channels of a tensor) and fed to a second stage of the machine learning model 408, which may comprise an encoder / decoder trunk 650 connected to a class confidence head 655 and an instance regression head 660. The class confidence head 655 and instance regression head 660 can extract class confidence data 610 (e.g., one or more confidence maps in a top-down view) and instance regression data 412 (e.g., object instance data such as position, geometry, and / or orientation), which can be post-processed (e.g., decoded) to generate bounding boxes, closed polylines, or other bounding shapes that identify the position, size, and / or orientation of the detected object instances.

[0084] 10 is a flow diagram illustrating a method 1000 for perspective and top-down view object detection using LiDAR data, according to some embodiments of the present disclosure. The method 1000 includes, at block B1002, receiving LiDAR data from one or more LiDAR sensors in an environment. For example, firmware associated with one or more LiDAR sensors (e.g., sensor 401 of FIG. 4) may be used to control the LiDAR sensors to capture and / or process LiDAR data, such as one or more LiDAR point clouds.

[0085] The method 1000 includes generating, at block B1004, first data representing a perspective view of the environment from the LiDAR data. For example, in an embodiment in which the sensor data 402 of FIG. 4 includes LiDAR data, the LiDAR data can be accumulated, transformed to a single coordinate system (e.g., centered on an ego actor / vehicle associated with the LiDAR sensor), ego-motion corrected (e.g., to a last known location of the ego actor / vehicle), and / or projected to form a projected image of a desired size (e.g., spatial dimension). For example, the (accumulated, ego-motion corrected) LiDAR point cloud can be projected to form a LiDAR range image having a perspective view. The projected image (e.g., LiDAR range image) and / or other reflectance data can be stored and / or encoded in an appropriate representation (e.g., input data 406) of the machine learning model 408.

[0086] The method 1000 includes, at block B1006, generating classification data from the first data using one or more neural networks (NNs), the classification data representing one or more classifications in the perspective view. For example, the input data 406 of FIG. 4 may serve as input to a machine learning model 408, which may include multiple constituent machine learning models or stages chained together to sequentially process different views of the 3D environment. In an exemplary first stage, the encoder / decoder 605 of FIG. 6 may extract classification data (e.g., one or more confidence maps, one or more segmentation masks) in the first view, such as the perspective view, from the input data 406.

[0087] The method 1000 includes generating transformed classification data representing one or more classifications in a top-down view of the environment at block B1008 by projecting one or more classifications in the perspective view to the top-down view using the LiDAR data. For example, in the example shown in FIG. 6, the classification data may correspond to class confidence data 610 (which may represent one or more confidence maps in a first view, such as a perspective view). In some cases, the classification values ​​represented by the class confidence data 610 may be associated with corresponding 3D positions of the LiDAR detections represented by corresponding range scan pixels in the LiDAR range image to generate labeled 3D positions. The labeled 3D positions may be projected 625 to a second view of the environment, for example, by orthogonally projecting the labeled 3D positions to form a projected image (e.g., an overhead image having a top-down view) having a desired spatial dimension and ground sampling distance. Each pixel in the resulting projection image onto which there is a projected 3D location may store the extracted classification data (eg, classification value or label) with which the 3D location is labeled.

[0088] The method 1000 includes, at block B1010, generating, using one or more NNs, second data representing one or more bounding shapes of one or more objects detected in the environment based at least on the transformed classification data in the top-down view. For example, the transformed classification data (e.g., transformed class confidence data 630) and / or geometry data (e.g., geometry data 640) can be encoded 645 and provided to a second stage of the machine learning model 408 of FIG. 6. The output of the second stage (e.g., class confidence data 410 and instance regression data 412) can be post-processed (e.g., decoded) to generate bounding boxes, closed polylines, or other bounding shapes that identify the positions, geometries, and / or orientations of the detected object instances.

[0089] FIG. 11 is a flow diagram illustrating a method 1100 for multi-view object detection including projection of labeled sensor data, according to some embodiments of the present disclosure. The method 1100 includes, at block B1102, generating classification data representing one or more classifications from image data representing an image of a first view of an environment, using one or more neural networks (NNs). For example, the input data 406 (e.g., a representation of a LiDAR range image) of FIG. 4 may serve as an input to a machine learning model 408, which may include multiple constituent machine learning models or stages chained together to sequentially process different views of a 3D environment. In an exemplary first stage, the encoder / decoder 605 of FIG. 6 may extract classification data (e.g., one or more confidence maps, one or more segmentation masks) in a first view, such as a perspective view, from the input data 406.

[0090] The method 1100 includes, at block B1104, associating the classification data with corresponding three-dimensional (3D) locations identified from the corresponding sensor data to generate labeled sensor data. For example, the 3D locations from the sensor data (e.g., sensor data 402) of the corresponding time slice may be labeled 620 with the classification data (e.g., classification values, labels) extracted by the encoder / decoder 605. In some embodiments, the corresponding input LiDAR range image may have a known correspondence between range scan pixels and corresponding points in the LiDAR point cloud (LiDAR detections) that may have known 3D locations. Thus, for example, a classification value from the predicted confidence map may be associated with the 3D location of the LiDAR detection represented by the corresponding range scan pixel in the input LiDAR range image.

[0091] The method 1100 includes, at block B1106, projecting the labeled sensor data onto a second view of the environment to generate transformed classification data representative of one or more classifications in the second view. For example, the labeled 3D locations may be projected 625 into the second view of the environment, e.g., by orthogonally projecting the labeled 3D locations to form a projected image having a desired spatial dimension and ground sampling distance (e.g., an overhead image having a top-down view). Each pixel of the resulting projected image with a projected 3D location may store extracted classification data (e.g., a classification value or label) with which the 3D location was labeled.

[0092] The method 1100 includes generating, at block B1108, using one or more neural networks (NNs) second data representing one or more bounding shapes of one or more objects detected in the environment based at least on the transformed classification data. For example, the transformed classification data (e.g., transformed class confidence data 630) and / or geometry data (e.g., geometry data 640) may be encoded 645 and provided to a second stage of the machine learning model 408 of FIG. 6. The output of the second stage (e.g., class confidence data 410 and instance regression data 412) may be post-processed (e.g., decoded) to generate bounding boxes, closed polylines, or other bounding shapes that identify the positions, geometries, and / or orientations of the detected object instances.

[0093] Training a Machine Learning Model for an Object Detection System To train a machine learning model of the object detection system (e.g., machine learning model 408 of FIG. 4), input training data may be generated from the sensor data using techniques described herein for operating machine learning model 408. Ground truth training data may be obtained by annotating data from multiple sensors in a sensor setup.

[0094] For example, a scene may be observed with a LiDAR sensor and one or more cameras (e.g., LiDAR sensor 1664, stereo camera 1668, wide view camera 1670 (e.g., fisheye camera), infrared camera 1672, surround camera 1674 (e.g., 360 degree camera), long range and / or mid-range camera 1698, etc., of the autonomous vehicle 1600 of FIGS. 16A-16D ) to collect LiDAR data and image data. Because data may be acquired from different sensors at different frequencies, in some embodiments, a particular sensor (e.g., a LiDAR sensor) may be used as a reference sensor. For each frame of sensor data from the reference sensor (e.g., for each frame of LiDAR data), a set of sensor data may be curated by identifying the frame of sensor data from each of the other sensors in the sensor setup that is closest in time to the frame of sensor data from the reference sensor. This set of sensor data (e.g., a frame of LiDAR data at timestamp T plus images acquired proximate in time to T from each of the multiple cameras in the sensor setup) can be referred to as a set of curated sensor data at timestamp T.

[0095] In general, each set of curated sensor data may be used to generate input training data (e.g., input data 406 and / or geometry data 640 of FIG. 6) for the machine learning model 408, as described in more detail above. To generate ground truth training data for each set of curated sensor data, data from each sensor may be annotated and / or labeled independently from data from the other sensors. For example, frames of LiDAR data may be projected to form projection images, and the projection images may be annotated with 2D or 3D boundary shapes and supported class labels. Additionally or alternatively, each image (or a composite image formed from constituent images from a set of curated sensor data) may be annotated with 2D or 3D boundary shapes and supported class labels. The annotations may be generated manually and / or automatically and used to derive corresponding ground truth outputs (e.g., classification data such as confidence maps or segmentation masks, object instance data, etc.) of the machine learning model 408.

[0096] FIG. 12 is a diagram of an example technique for annotating sensor data from different sensors, according to some embodiments of the disclosure. In the example shown in FIG. 12, sensor data may be captured from a LiDAR sensor and one or more cameras. In general, the sensor data may be captured and curated to produce training data having any desired frequency. The timeline at the top of FIG. 12 indicates that frames of LiDAR data in this example are generated more frequently than camera images. In this case, the LiDAR sensor may be used as a reference sensor, and camera images captured closest in time to each LiDAR frame may be identified to form a set of curated sensor data (e.g., set of curated sensor data 1210). In the example shown in FIG. 12, curated sensor data having any desired frequency may be produced, but curated sensor data having a frequency of one frame per second may be produced.

[0097] In FIG. 12, a curated sensor data set 1210 includes a composite RBG image 1220 and a frame of LiDAR data 1230. A triangle in the frame of LiDAR data 1230 shows correspondence with three images generated by three different cameras, with the edges of the triangle numbered to show the field of view of the three corresponding camera frustums. In this example, the composite RBG image 1220 is formed by stitching together three different RGB images. The composite RBG image 1220 and the frame of LiDAR data 1230 can be independently annotated with 2D or 3D boundary shapes and supported class labels. In FIG. 12, a labeled image 1240 and a labeled LiDAR frame 1250 show example annotations (class labels omitted for brevity).

[0098] In general, sensor data (e.g., frames of LiDAR data, RBG images) may be annotated (e.g., manually, automatically, etc.) with labels or other markers that identify the location, geometry, orientation, and / or class of relevant object instances in the sensor data. In some instances, the labels may be generated within a 2D or 3D drawing program (e.g., annotation program), a computer-aided design (CAD) program, a labeling program, another type of suitable program, and / or may be hand-drawn. In any instance, the labels may be synthetically produced (e.g., generated from a computer model or rendering), realistically produced (e.g., designed and produced from real-world data), machine-automated (e.g., using feature analysis and learning to extract features from data and then generate labels), human-annotated (e.g., a labeler or annotation expert defines the location of the labels), and / or a combination thereof (e.g., a human identifies vertices of a polyline, a machine generates polygons using a polygon rasterizer). In general, labels may include 2D and / or 3D bounding boxes, closed polylines, or other bounding shapes that are drawn, annotated, overlaid, and / or otherwise associated with the sensor data.

[0099] FIG. 13 is an illustration of an example annotation of car and truck classes in camera space, according to some embodiments of the present disclosure. For example, a boundary 1310 is identified around the truck, and the boundary 1310 is labeled with a corresponding class label 1320 (e.g., truck) and one or more subclass labels 1330 (e.g., occlusion: front view, truncated: front view). FIG. 14 is an illustration of an example annotation of a pedestrian class in camera space, according to some embodiments of the present disclosure. For example, a boundary 1410 is identified around the pedestrian, and the boundary 1410 is labeled with a corresponding class label 1420 (e.g., person) and one or more subclass labels 1430 (e.g., occlusion: front view, truncated: bottom truncated). FIG. 15 is an illustration of an example annotation of a top-down bounding box in LiDAR space, according to some embodiments of the present disclosure. For example, a boundary (e.g., the 2D bounding box shown in FIG. 15) may be identified around the car and the truck. Each boundary can be labeled with a corresponding class label, which is omitted from Figure 15 for brevity.

[0100] In some embodiments, object detection and tracking may be applied to track the movement of annotated objects from frame to frame over time. As such, annotation tracking may be used to track objects from frame to frame (e.g., using persistent identifiers of the annotated objects), and the tracked annotations may be updated to fine-tune the annotations in individual frames.

[0101] In some embodiments, object trajectories and / or detections from sensor data from a particular sensor may be linked to corresponding object trajectories and / or detections of the same object from sensor data from a different sensor. Returning to FIG. 12 , the third column illustrates one example in which the labeled image 1240 and the labeled LiDAR frame 1250 may be linked. For example, annotations from different types of sensor data may be projected into a common space (e.g., annotations from the labeled image 1240 may be projected into the image space of the labeled LiDAR frame 1250, or vice versa, annotations from both the labeled image 1240 and the labeled LiDAR frame 1250 may be projected into some other space, etc.). In the example illustrated in FIG. 12 , annotations from both the labeled image 1240 and the labeled LiDAR frame 1250 may be projected into the 3D space illustrated by the labeled frame 1260. In some embodiments, annotations from different types of sensor data and / or different class labels may be presented in a manner that indicates the different classes or sensor data (e.g., presented with different colors, labels, etc.). Corresponding annotations from different types of sensor data (e.g., overlapping annotations with a threshold IoU) may be linked (e.g., manually and / or automatically). In some embodiments, linked annotations or object trajectories from different types of sensor data may be used as input to an object detection and tracking process to track frame-to-frame movement of annotated objects over time with improved accuracy.

[0102] Additionally or alternatively, the linked sensor data can facilitate identification and / or editing of annotations. For example, the annotation software can present any combination of sensor data, annotations, and / or label views and can accept new annotations and / or edits to existing annotations. As a non-limiting example, the labeled LiDAR frame 1270 shows a 2D top-down bounding box from the labeled LiDAR frame 1250 that has been extended to 3D using 3D software based on correspondence with annotations from the labeled image 1240 shown by the labeled frame 1260. While the above has focused on LiDAR-to-camera linking, any type of sensor data can be linked to any other type of sensor data (including linking between sensor data from two different types of the same sensor, such as a camera-to-camera linking). These are intended as merely illustrative examples and other variations can be implemented within the scope of this disclosure.

[0103] In general, each set of curated sensor data may be used to generate input training data for the machine learning model 408 (e.g., input data 406 and / or geometry data 640 of FIG. 6), and the annotations (e.g., borders, enclosed regions, class labels) may be used to generate ground truth data for the machine learning model 408 (e.g., class confidence data 610, class confidence data 410, and instance regression data 412 of FIG. 6). To generate ground truth data from the annotations, the position, geometry, orientation, and / or class of each annotation may be used to generate a confidence map and / or partition mask that matches the view, size, and dimensionality of the output of the machine learning model 408 (e.g., class confidence data 610 and class confidence data 410). As a non-limiting example, for a given class and corresponding dimension of a tensor storing a different confidence map in each channel, the pixel values ​​of pixels within each labeled bounding shape of that class may be set to a value indicating a positive classification (e.g., 1). The process may be repeated and stored in the corresponding channel of the ground truth class confidence tensor. Additionally or alternatively, the position, geometry, orientation, and / or class of each annotation may be used to generate object instance data that matches the view, size, and dimensionality of the instance regression data 412. For example, for each pixel contained with an annotation, the annotation may be used to calculate corresponding position, geometry, and / or orientation information (e.g., where the object is located with respect to each pixel (e.g., object center), object height, object width, object orientation (e.g., rotation angle relative to the orientation of the projected image), and / or the like). The calculated object instance data may be stored in the corresponding channel of the ground truth instance regression tensor. Thus, the annotations may be used to generate ground truth class classification and / or instance regression data, which may be used with corresponding input training data (e.g., input images, input tensors) as part of a training dataset for training the machine learning model 408.

[0104] In embodiments where the machine learning model 408 includes a series of multiple stages, the different stages may be trained together or separately. In some embodiments, the stages may be trained together by performing a transformation (e.g., transformation 615 of FIG. 6 ) from the output of a first stage (first view) to the input to a second stage (second view) using a differentiable operation (e.g., differentiable reprojection). For example, training input data may be input to a first stage of the machine learning model 408, and the output of the first stage may be transformed 615 using a differentiable transformation, encoded (e.g., with ground truth geometry data 640), and fed to a second stage of the machine learning model 408.

[0105] In any case, the ground truth class partition and / or instance regression data may be used to train the machine learning model 408. For example, one or more loss functions (e.g., a single loss function, a loss function for each output type, etc.) may be used to compare the accuracy of the output of the machine learning model 408 to the ground truth, and the parameters of the machine learning model 408 may be updated (e.g., using backward passes, back propagation, forward passes, etc.) until the accuracy reaches an optimal or acceptable level. In some embodiments where the machine learning model 408 includes multiple heads, the multiple heads may be co-trained with a common trunk or may be trained individually. In embodiments including multi-view scenarios with separate DNN feature extractors, the DNN feature extractors may be co-trained separately or together. These and other variations are contemplated within this disclosure.

[0106] In some implementations, a classification loss function such as a multi-class cross-entropy loss may be used for the classification task. The multi-class cross-entropy loss may include a pixel-wise cross-entropy loss summed across all classes. For example, the classification loss (e.g., of the first and / or second stages of the machine learning model 408) may be expressed as H(p,y)=-Σ i y i log(p i ), where y is the ground truth data (e.g., 1 if the pixel is in a class, 0 otherwise), p is the predicted depthwise pixel vector with each dimension corresponding to a particular class, and p is the predicted classification data per pixel (e.g., the likelihood, score, or logit that the pixel is in a given class i), and a summation may be performed pixel-by-pixel across all classes (e.g., across all depth channels). The classification loss for the entire frame may be calculated by computing the classification loss for each pixel and taking the average.

[0107] In some embodiments, a regression loss function such as L1 or L2 loss may be used for the regression task. For example, the regression loss (e.g., of the second stage of the machine learning model 408) is given by R(v,t)=|vt|, where v is the predicted depthwise pixel vector to be regressed and t is the target / ground truth depthwise pixel vector, and the regression loss per pixel may be summed over all depth channels. In some embodiments, the regression loss may be used only for those pixels that belong to the corresponding classification mask (e.g., predicted by or derived from the class confidence head 655 of FIG. 6). Thus, for each pixel (within the mask), the regression loss may be summed over each depth channel and the average may be calculated over all object pixels, for example, by normalizing (e.g., dividing) by the number of pixels that fall in the object mask (e.g., of all classes).

[0108] In some embodiments, the total loss may be calculated as the sum of the classification loss (e.g., from the first and / or second stage of the machine learning model 408) and the regression loss (e.g., from the second stage of the machine learning model 408). In some embodiments, the contributions to the loss from different tasks may be weighted with fixed and / or automatic weights. Additionally or alternatively, the classification loss may be weighted to counteract class imbalances present in the training dataset. These and other variations may be implemented within the scope of the present disclosure.

[0109] Exemplary Autonomous Vehicle 16A is a diagram of an example autonomous vehicle 1600 according to some embodiments of the disclosure. Autonomous vehicle 1600 (alternatively referred to herein as “vehicle 1600”) may include, but is not limited to, passenger vehicles, such as cars, trucks, buses, first responder vehicles, shuttles, electric or mopeds, motorcycles, fire engines, police vehicles, ambulances, boats, construction vehicles, submarines, drones, and / or other types of vehicles (e.g., unmanned and / or carrying one or more passengers). Autonomous vehicles are generally described in terms of automation levels as 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 published June 15, 2018, Standard No. J3016-201609 published September 30, 2016, and previous and future versions of this standard). The mobile vehicle 1600 may be capable of functionality according to one or more of the autonomous driving levels Levels 3 through 5. For example, the mobile vehicle 1600 may be capable of conditional automation (Level 3), high automation (Level 4), and / or full automation (Level 5), depending on the embodiment.

[0110] The mobile vehicle 1600 may include components such as a chassis, body, wheels (e.g., 2, 4, 6, 8, 18, etc.), tires, axles, and other components of the mobile vehicle. The mobile vehicle 1600 may include a propulsion system 1650, such as an internal combustion engine, a hybrid power plant, a fully electric engine, and / or another propulsion system type. The propulsion system 1650 may be connected to a drive train of the mobile vehicle 1600, which may include a transmission, to enable propulsion of the mobile vehicle 1600. The propulsion system 1650 may be controlled in response to receiving a signal from a throttle / accelerator 1652.

[0111] A steering system 1654, which may include a steering wheel, may be used to steer the vehicle 1600 (e.g., along a desired course or route) when the propulsion system 1650 is operating (e.g., when the vehicle is moving). The steering system 1654 may receive signals from a steering actuator 1656. The steering wheel may be an option for fully automated (Level 5) functionality.

[0112] The brake sensor system 1646 may be used to operate the vehicle brakes in response to receiving a signal from a brake actuator 1648 and / or a brake sensor.

[0113] A controller 1636, which may include one or more system on chip (SoC) 1604 (FIG. 16C) and / or a GPU, can provide signals (e.g., representations of commands) to one or more components and / or systems of the vehicle 1600. For example, the controller can send signals to operate vehicle brakes via one or more brake actuators 1648, operate a steering system 1654 via one or more steering actuators 1656, and operate a propulsion system 1650 via one or more throttle / accelerators 1652. The controller 1636 may include one or more on-board (e.g., integrated) computing devices (e.g., supercomputers) to process sensor signals and output operational commands (e.g., signals representing commands) to enable autonomous driving and / or to assist a driver in driving the vehicle 1600. The controllers 1636 may include a first controller 1636 for autonomous driving functions, a second controller 1636 for functional safety functions, a third controller 1636 for artificial intelligence functions (e.g., computer vision), a fourth controller 1636 for infotainment functions, a fifth controller 1636 for redundancy in emergency situations, and / or other controllers. In some instances, a single controller 1636 may handle two or more of the aforementioned functions, and two or more controllers 1636 may handle a single function, and / or any combination thereof.

[0114] Controller 1636 may provide signals to control one or more components and / or systems of vehicle 1600 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, global navigation satellite system sensors 1658 (e.g., global positioning system sensors), RADAR sensors 1660, ultrasonic sensors 1662, LiDAR sensors 1664, inertial measurement unit (IMU) sensors 1666 (e.g., accelerometers, gyroscopes, magnetic compasses, magnetometers, etc.), microphones 1696, stereo cameras 1668, wide-view cameras 1670 (e.g., fisheye cameras), infrared cameras 1672, surround cameras 1674 (e.g., 360 degree cameras), long-range and / or medium-range cameras 1698, speed sensors 1644 (e.g., for measuring the speed of the moving vehicle 1600), vibration sensors 1642, steering sensors 1640, brake sensors (e.g., as part of a brake sensor system 1646), and / or other sensor types.

[0115] One or more of the controllers 1636 may receive input (e.g., represented by input data) from the instrument cluster 1632 of the vehicle 1600 and provide output (e.g., represented by output data, display data, etc.) via a human-machine interface (HMI) display 1634, an audible annunciator, a loudspeaker, and / or other components of the vehicle 1600. The output may include information such as vehicle velocity, speed, time, map data (e.g., HD map 1622 of FIG. 16C), position data (e.g., the position of vehicle 1600 on a map, etc.), direction, positions of other vehicles (e.g., occupancy grid), information about objects and object situations as perceived by the controller 1636, etc. For example, the HMI display 1634 may display information regarding the presence of one or more objects (e.g., road signs, warning signs, changing traffic signals, etc.) and / or a driving maneuver that the moving vehicle has performed, is performing, or will perform (e.g., changing lanes now, taking exit 34B in 2 miles, etc.).

[0116] The mobile vehicle 1600 further includes a network interface 1624 that can communicate over one or more networks using one or more wireless antennas 1626 and / or a modem. For example, the network interface 1624 can be capable of communication over LTE, WCDMA, UMTS, GSM, CDMA2000, etc. The wireless antennas 1626 can also enable communication between objects (e.g., mobile vehicles, mobile devices, etc.) in the environment using local area networks such as Bluetooth, Bluetooth LE, Z-Wave, ZigBee, etc., and / or low power wide-area networks (LPWANs) such as LoRaWAN, SigFox, etc.

[0117] 16B is an illustration of camera positions and fields of view of the example autonomous vehicle 1600 of FIG. 16A, according to some embodiments of the present disclosure. The cameras and their respective fields of view are one example embodiment and are not intended to be limiting. For example, additional and / or alternative cameras may be included and / or the cameras may be located at different positions on the vehicle 1600.

[0118] The camera type of the camera may include, but is not limited to, a digital camera that may be adapted for use with the components and / or systems of the mobile vehicle 1600. The camera may be capable of operating at automotive safety integrity level (ASIL) B and / or at another ASIL. The camera type may be capable of any image capture rate, such as 60 frames per second (fps), 1620 fps, 240 fps, 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 instances, the color filter array may include a red clear clear clear (RCCC) color filter array, a red clear clear blue (RCCB) color filter array, a red blue green clear (RBGC) color filter array, a Foveon X3 color filter array, a Bayer sensor (RGGB) color filter array, a monochrome sensor color filter array, and / or another type of color filter array. In some embodiments, clear pixel cameras, such as cameras with RCCC, RCCB, and / or RBGC color filter arrays, may be used in an effort to increase light sensitivity.

[0119] In some instances, one or more of the cameras may 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 mono camera may be installed to provide functions including lane departure warning, traffic sign assist, and intelligent headlamp control. One or more of the cameras (e.g., all of the cameras) may simultaneously record and provide image data (e.g., video).

[0120] One or more of the cameras may be mounted in a mounting part, such as a custom designed (3D printed) part, to filter out stray light and reflections from within the vehicle (e.g., reflections from the dashboard reflected in the windshield mirror) that may interfere with the camera's image data capture capabilities. With reference to a side mirror mounting part, the side mirror part may be custom 3D printed such that the camera mounting plate fits the shape of the side mirror. In some instances, the camera may be integrated into the side mirror. For side view cameras, the cameras may also be integrated into four posts at each corner of the cabin.

[0121] A camera (e.g., a forward-facing camera) with a field of view that includes a portion of the environment in front of the vehicle 1600 may be used for surround view to help identify the forward path and obstacles and, with the aid of one or more controllers 1636 and / or control SoCs, provide information essential to generating an occupancy grid and / or determining a preferred vehicle path. Forward-facing cameras may be used to perform many of the same ADAS functions as LiDAR, including emergency braking, pedestrian detection, and collision avoidance. Forward-facing cameras may also be used for ADAS functions and systems, including other functions such as lane departure warning ("LDW"), autonomous cruise control ("ACC"), and / or traffic sign recognition.

[0122] A variety of cameras may be used in a forward-facing configuration, including, for example, a monocular camera platform including a complementary metal oxide semiconductor (CMOS) color imager. Another example may be a wide-view camera 1670 that may be used to capture objects entering the view from the periphery (e.g., pedestrians, crossing traffic, or bicycles). Although only one wide-view camera is shown in FIG. 16B, there may be any number of wide-view cameras 1670 in the vehicle 1600. In addition, a long-range camera 1698 (e.g., a long-view stereo camera pair) may be used for depth-based object detection, especially for objects for which the neural network has not yet been trained. The long-range camera 1698 may also be used for object detection and classification, as well as basic object tracking.

[0123] One or more stereo cameras 1668 may also be included in the forward-facing configuration. The stereo camera 1668 may include an integrated control unit with an extensible processing unit that may provide programmable logic (FPGA) and a multi-core microprocessor with CAN or Ethernet interfaces integrated on a single chip. Such a unit may be used to generate a 3D map of the vehicle's environment, including distance estimates for all points in the image. An alternative stereo camera 1668 may include a compact stereo vision sensor that may include two camera lenses (one left and one right) and an image processing chip that can measure distances from the vehicle to objects of interest and use the generated information (e.g., metadata) to activate autonomous emergency braking and lane departure warning functions. Other types of stereo cameras 1668 may be used in addition to or instead of those described herein.

[0124] A camera having a field of view that includes a portion of the environment to the side of the moving vehicle 1600 (e.g., a side-view camera) may be used for surround view, providing information used to create and update the occupancy grid and to generate side impact collision warnings. For example, surround cameras 1674 (e.g., four surround cameras 1674 as shown in FIG. 16B) may be positioned on the moving vehicle 1600. The surround cameras 1674 may include wide-view cameras 1670, fish-eye cameras, 360-degree cameras, and / or the like. For example, four fish-eye cameras may be positioned at the front, back, and sides of the moving vehicle. In an alternative arrangement, the moving vehicle may use three surround cameras 1674 (e.g., left, right, and rear) and may utilize one or more other cameras (e.g., a forward-facing camera) as a fourth surround view camera.

[0125] A camera having a field of view that includes the portion of the environment behind the moving vehicle 1600 (e.g., a rear-view camera) may be used for parking assistance, surround view, rear collision warning, and creating and updating an occupancy grid. As described herein, a wide variety of cameras may be used, including, but not limited to, cameras that are also suitable as forward-facing cameras (e.g., long-range and / or mid-range cameras 1698, stereo cameras 1668, infrared cameras 1672, etc.).

[0126] FIG. 16C is a block diagram of an example system architecture of the example autonomous vehicle 1600 of FIG. 16A, according to some embodiments of the present disclosure. It should be understood that this and other arrangements described herein are merely illustrative. Other arrangements and elements (e.g., machines, interfaces, functions, sequences, groupings of functions, etc.) may be used in addition to or instead of those shown, and some elements may be omitted altogether. Furthermore, many of the elements described herein are functional entities that may be implemented as separate or distributed components or in conjunction with other components, and in any suitable combination and location. Various functions described herein as being performed by entities may be implemented by hardware, firmware, and / or software. For example, various functions may be implemented by a processor executing instructions stored in a memory.

[0127] Each of the components, features, and systems of the mobile vehicle 1600 of FIG. 16C are shown as connected via a bus 1602. The bus 1602 may include a controller area network (CAN) data interface (alternatively referred to as a "CAN bus"). The CAN may be a network within the mobile vehicle 1600 that is used to help control various features and functions of the mobile vehicle 1600, such as actuation of brakes, acceleration, braking, steering, windshield wipers, etc. The CAN bus may be configured to have dozens or even hundreds of nodes, each with its own unique identifier (e.g., CAN ID). The CAN bus may be read to find steering angle, ground speed, engine revolutions per minute (RPM), button positions, and / or other vehicle status indicators. The CAN bus may be ASIL B compliant.

[0128] The bus 1602 is described herein as being a CAN bus, but this is not intended to be limiting. For example, FlexRay and / or Ethernet may be used in addition to or as an alternative to a CAN bus. Additionally, a single line is used to represent the bus 1602, but this is not intended to be limiting. There may be any number of buses 1602, which may include, for example, 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 instances, two or more buses 1602 may be used to perform different functions and / or may be used for redundancy. For example, a first bus 1602 may be used for collision avoidance functions and a second bus 1602 may be used for actuation control. In any instance, each bus 1602 may communicate with any of the components of the vehicle 1600, and two or more buses 1602 may communicate with the same components. In some examples, each SoC 1604, each controller 1636, and / or each computer in the vehicle may have access to the same input data (e.g., input from sensors in the vehicle 1600) and may be connected to a common bus, such as a CAN bus.

[0129] The mobile vehicle 1600 may include one or more controllers 1636, such as those described herein with respect to FIGURE 16A. The controller 1636 may be used for a variety of functions. The controller 1636 may be coupled to any of a variety of other components and systems of the mobile vehicle 1600 and may be used for control of the mobile vehicle 1600, artificial intelligence of the mobile vehicle 1600, infotainment for the mobile vehicle 1600, and / or the like.

[0130] The vehicle 1600 may include a system on a chip (SoC) 1604. The SoC 1604 may include a CPU 1606, a GPU 1608, a processor 1610, a cache 1612, an accelerator 1614, a data store 1616, and / or other components and features not shown. The SoC 1604 may be used to control the vehicle 1600 in a variety of platforms and systems. For example, the SoC 1604 may be coupled in a system (e.g., the vehicle 1600 system) having an HD map 1622 that can obtain map refreshes and / or updates via a network interface 1624 from one or more servers (e.g., server 1678 of FIG. 16D ).

[0131] The CPU 1606 may include a CPU cluster or CPU complex (alternatively referred to as a "CCPLEX"). The CPU 1606 may include multiple cores and / or L2 caches. For example, in some embodiments, the CPU 1606 may include eight cores in a coherent multiprocessor configuration. In some embodiments, the CPU 1606 may include four dual-core clusters, each with its own dedicated L2 cache (e.g., a 2M B L2 cache). The CPU 1606 (e.g., a CCPLEX) may be configured to support concurrent cluster operation allowing any combination of the CPU 1606 clusters to be active at any given time.

[0132] The CPU 1606 may implement power management capabilities including one or more of the following features: individual hardware blocks 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 by 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 1606 may further implement an enhanced algorithm for managing power states where allowed power states and expected wake-up times are specified and hardware / microcode determines the best power state for entering the cores, clusters, and CCPLEXes. The processing cores may support simplified power state entry sequences in software where work is offloaded to the microcode.

[0133] The GPU 1608 may include an integrated GPU (alternatively referred to herein as an "iGPU"). The GPU 1608 may be programmable and efficient for parallel workloads. In some instances, the GPU 1608 may use an enhanced tensor instruction set. The GPU 1608 may include one or more streaming microprocessors, where each streaming microprocessor may include an L1 cache (e.g., an L1 cache having at least 96 KB storage capacity) and two or more of the streaming microprocessors may share a cache (e.g., an L2 cache having 512 KB storage capacity). In some embodiments, the GPU 1608 may include at least eight streaming microprocessors. The GPU 1608 may use a computational application programming interface (API). In addition, the GPU 1608 may use one or more parallel computing platforms and / or programming models (e.g., NVIDIA's CUDA).

[0134] The GPU 1608 may be power optimized for best performance in automotive and embedded use cases. For example, the GPU 1608 may be fabricated on FinFETs (Fin field-effect transistors). However, this is not intended to be limiting and the GPU 1608 may be fabricated using other semiconductor fabrication processes. Each streaming microprocessor may incorporate several mixed-precision processing cores partitioned into multiple blocks. For example, and without limitation, 64 PF32 cores and 32 PF64 cores may be partitioned into four processing blocks. In such an example, each processing block may be assigned 16 FP32 cores, 8 FP64 cores, 16 INT32 cores, two mixed-precision NVIDIA tensor cores for deep learning matrix operations, an L0 instruction cache, a warp scheduler, a dispatch unit, and / or a 64KB register file. In addition, the streaming microprocessor may include independent parallel integer and floating-point data paths to provide efficient execution of workloads having a mix of computational and addressing operations. Streaming microprocessors may include independent thread scheduling capabilities to allow finer grain synchronization and coordination among concurrent threads. Streaming microprocessors may include a combined L1 data cache and shared memory unit to improve performance while simplifying programming.

[0135] The GPU 1608 may include high bandwidth memory (HBM) and / or a 16GB HBM2 memory subsystem to provide about 900GB / sec peak memory bandwidth in some instances. In some instances, synchronous graphics random-access memory (SGRAM), such as graphics double data rate type five synchronous random-access memory (GDDR5), may be used in addition to or in place of the HBM memory.

[0136] The GPU 1608 may include unified memory technology including access counters to enable more accurate movement of memory pages to the processors that access them most frequently, thereby improving the efficiency of storage ranges shared between processors. In some instances, address translation service (ATS) support may be used to enable the GPU 1608 to directly access the CPU 1606 page tables. In such instances, when the GPU 1608 memory management unit (MMU) experiences a miss, an address translation request may be sent to the CPU 1606. In response, the CPU 1606 may consult its page tables for a virtual-to-real mapping of the address and send the translation back to the GPU 1608. As such, the unified memory technology may enable a single unified virtual address space for both the CPU 1606 and GPU 1608 memories, thereby simplifying GPU 1608 programming and porting of applications to the GPU 1608.

[0137] In addition, the GPU 1608 may include access counters that can record the frequency of accesses of the GPU 1608 to the memory of other processors. The access counters can help ensure that memory pages are moved to the physical memory of the processor that is accessing the page most frequently.

[0138] The SoC 1604 may include any number of caches 1612, including those described herein. For example, the cache 1612 may include an L3 cache available to both the CPU 1606 and the GPU 1608 (e.g., connected to both the CPU 1606 and the GPU 1608). The cache 1612 may include a write-back cache that can record line state, such as by using a cache coherence protocol (e.g., MEI, MESI, MSI, etc.). The L3 cache may include 4 MB or more, depending on the embodiment, although smaller cache sizes may be used.

[0139] The SoC 1604 may include an arithmetic logic unit (ALU) that may be utilized in performing processing for any of the various tasks or operations (e.g., processing DNNs) of the vehicle 1600. In addition, the SoC 1604 may include a floating point unit (FPU) (or other math co-processor or numeric co-processor type) for performing mathematical operations within the system. For example, the SoC 1604 may include one or more FPUs integrated as execution units within the CPU 1606 and / or GPU 1608.

[0140] The SoC 1604 may include one or more accelerators 1614 (e.g., hardware accelerators, software accelerators, or a combination thereof). For example, the SoC 1604 may include a hardware acceleration cluster, which may include optimized hardware accelerators and / or large on-chip memory. The large on-chip memory (e.g., 4MB of SRAM) may enable the hardware acceleration cluster to accelerate neural networks and other operations. The hardware acceleration cluster may be used to complement the GPU 1608 and to offload some of the tasks of the GPU 1608 (e.g., to free up more cycles for the GPU 1608 to perform other tasks). As an example, the accelerator 1614 may be used for target workloads that are stable enough to be suitable for acceleration (e.g., perception, convolutional neural networks (CNNs), etc.). As used herein, the term "CNN" may include all types of CNNs, including region-based or regional convolutional neural networks (RCNNs) and Faster RCNNs (e.g., as used for object detection).

[0141] The accelerator 1614 (e.g., a hardware acceleration 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 inference. The TPU may be an accelerator configured and optimized to perform image processing functions (e.g., CNN, RCNN, etc.). The DLA may further be optimized for a specific set of neural network types and floating point operations, as well as inference. The design of the DLA may provide more performance per millimeter than a general-purpose GPU, greatly exceeding the performance of a CPU. The TPU may perform several functions, including, for example, single instance convolution functions, supporting INT8, INT16, and FP16 data types for both features and weights, as well as post-processor functions.

[0142] The DLA can quickly and efficiently run neural networks, in particular CNNs, on processed or unprocessed data for any of a variety of functions, including but not limited to: CNNs for object identification and detection using data from camera sensors, CNNs for distance estimation using data from camera sensors, CNNs for emergency vehicle detection and identification using data from microphones, CNNs for face recognition and moving vehicle owner identification using data from camera sensors, and / or CNNs for security and / or safety related events.

[0143] The DLA can perform any function of the GPU 1608, and by using an inference accelerator, for example, a designer can target either the DLA or the GPU 1608 for any function. For example, a designer can focus on processing CNNs and floating point operations on the DLA and offload other functions to the GPU 1608 and / or other accelerators 1614.

[0144] The accelerator 1614 (e.g., hardware acceleration cluster) may include a programmable vision accelerator (PVA), which may alternatively be referred to herein as a computer vision accelerator. The PVA may be designed and configured to accelerate computer vision algorithms for advanced driver assistance systems (ADAS), autonomous driving, and / or augmented reality (AR) and / or virtual reality (VR) applications. The PVA may provide a balance between performance and flexibility. For example, each PVA may include, but is not limited to, any number of reduced instruction set computer (RISC) cores, direct memory access (DMA), and / or any number of vector processors, for example.

[0145] The RISC cores may interact with an image sensor (e.g., an image sensor of any of the cameras described herein), an image signal processor, and / or the like. Each RISC core may include any amount of memory. The RISC cores may use any of a number of protocols, depending on the embodiment. In some instances, the RISC cores may run a real-time operating system (RTOS). The RISC cores may be implemented using one or more integrated circuit devices, application specific integrated circuits (ASICs), and / or memory devices. For example, the RISC cores may include an instruction cache and / or tightly coupled RAM.

[0146] The DMA may allow components of the PVA to access system memory independent of the CPU 1606. The DMA may support any number of features used to provide optimizations to the PVA, including but not limited to supporting multi-dimensional addressing and / or circular addressing. In some instances, the DMA may support up to six or more dimensions of addressing, which may include block width, block height, block depth, horizontal block stepping, vertical block stepping, and / or depth stepping.

[0147] The vector processor may be a programmable processor that may be designed to efficiently and flexibly execute computer vision algorithm programming and provide signal processing capabilities. In some instances, the PVA may include a PVA core and two vector processing subsystem partitions. The PVA core may include a processor subsystem, a DMA engine (e.g., two DMA engines), and / or other peripherals. The vector processing subsystem may act as the primary processing engine of the PVA and may include a vector processing unit (VPU), an instruction cache, and / or a vector memory (e.g., VMEM). The VPU core may include a digital signal processor, such as, for example, a single instruction, multiple data (SIMD), very long instruction word (VLIW) digital signal processor. The combination of SIMD and VLIW may increase throughput and speed.

[0148] Each vector processor may include an instruction cache and may be coupled to dedicated memory. As a result, in some instances, each vector processor may be configured to execute independently of other vector processors. In other instances, the vector processors included in a particular PVA may be configured to employ data parallelism. For example, in some embodiments, multiple vector processors included in a single PVA may execute the same computer vision algorithm, but on different regions of an image. In other instances, the vector processors included in a particular PVA may execute different computer vision algorithms simultaneously on the same image, or even execute different algorithms on sequential images or portions of an image. In particular, any number of PVAs may be included in a hardware accelerated cluster, and any number of vector processors may be included in each PVA. In addition, the PVAs may include additional error correcting code (ECC) memory to increase overall system security.

[0149] The accelerator 1614 (e.g., a hardware acceleration cluster) may include a computer vision network-on-chip and SRAM to provide high bandwidth, low latency SRAM for the accelerator 1614. In some instances, the on-chip memory may include, for example and without limitation, at least 4 MB of SRAM consisting of eight field configurable memory blocks that may be accessible by both the PVA and DLA. Each pair of memory blocks may include an advanced peripheral bus (APB) interface, configuration circuitry, controllers, and multiplexers. Any type of memory may be used. The PVA and DLA may access the memory through a backbone that provides the PVA and DLA with high speed access to the memory. The backbone may include a computer vision network-on-chip that interconnects the PVA and DLA to the memory (e.g., using the APB).

[0150] The computer vision network-on-chip may include an interface that determines, prior to transmission of any control signals / addresses / data, that both the PVA and DLA provide ready and valid signals. Such an interface may provide separate phases and separate channels for transmitting control signals / addresses / data, as well as burst type communication for continuous data transfer. This type of interface may follow ISO26262 or IEC61508 standards, although other standards and protocols may be used.

[0151] In some examples, the SoC1604 may include a real-time ray tracing hardware accelerator, such as described in U.S. Patent Application Serial No. 16 / 101,232, filed August 10, 2018. The real-time ray tracing hardware accelerator may be used to quickly and efficiently determine the location and scale of objects (e.g., within a world model) to generate real-time visualization simulations for RADAR signal interpretation, for acoustic propagation synthesis and / or analysis, for simulation of SONAR systems, for general wave propagation simulation, for comparison against LiDAR data for localization and / or other functions, and / or for other uses. In some examples, one or more tree traversal units (TTUs) may be used to perform one or more ray tracing related operations.

[0152] The accelerator 1614 (e.g., a hardware accelerator cluster) has a variety of applications for autonomous driving. The PVA may be a programmable vision accelerator that can be used for critical processing stages in ADAS and autonomous vehicles. The capabilities of the PVA are well suited to the domain of algorithms that require predictable processing at low power and low latency. In other words, the PVA works well with semi-dense or dense regular computations on small data sets that require predictable execution times with low latency and low power. Hence, in the context of a platform for autonomous vehicles, the PVA is designed to run classic computer vision algorithms, since the PVA is efficient in object detection and operating on integer computations.

[0153] For example, according to one embodiment of the present technology, the PVA is used to perform computer stereo vision. A semi-global matching based algorithm may be used in some instances, but this is not intended to be limiting. Many applications for Level 3-5 autonomous driving require motion estimation / stereo matching on the fly (e.g., structure from motion, pedestrian recognition, lane detection, etc.). The PVA can perform computer stereo vision functions with input from two monocular cameras.

[0154] In some instances, PVA may 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 instances, PVA is used for time of flight depth processing, for example, by processing raw time of flight data to provide processed time of flight data.

[0155] DLA can be used to implement any type of network to enhance control and driving safety, including, for example, a neural network that outputs a measure of the confidence of each object detection. Such confidence value can be interpreted as a probability or as providing a relative "weight" of each detection compared to other detections. This confidence value allows the system to make further decisions regarding which detections should be considered true positive detections rather than false positive detections. For example, the system can set a confidence threshold and consider only detections above the threshold as true positive detections. In an automatic emergency braking (AEB) system, a false positive detection would cause a moving vehicle to automatically perform emergency braking, which is clearly undesirable. Therefore, only the most confident detections should be considered to trigger AEB. DLA can implement a neural network that regresses the confidence value. The neural network may receive as its inputs 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 1666 outputs that correlate with the vehicle 1600 orientation, distance, and 3D position estimates of the object obtained from the neural network and / or other sensors (e.g., a LiDAR sensor 1664 or a RADAR sensor 1660), and so forth.

[0156] The SoC 1604 may include a data store 1616 (e.g., a memory). The data store 1616 may be an on-chip memory of the SoC 1604 and may store the neural network to be executed by the GPU and / or DLA. In some instances, the data store 1616 may have a capacity large enough to store multiple instances of the neural network for redundancy and safety. The data store 1612 may comprise an L2 or L3 cache 1612. References to the data store 1616 may include references to memory associated with the PVA, DLA, and / or other accelerators 1614, as described herein.

[0157] The SoC 1604 may include one or more processors 1610 (e.g., embedded processors). The processors 1610 may include a boot and power management processor, which may be a dedicated processor and subsystem for handling boot power and management capabilities and associated security enforcement. The boot and power management processor may be part of the SoC 1604 boot sequence and may provide run-time power management services. The boot power and management processor may provide clock and voltage programming, assist with system low power state transitions, manage the SoC 1604 thermal and temperature sensors, and / or manage the SoC 1604 power states. Each temperature sensor may be implemented as a ring oscillator whose output frequency is proportional to temperature, and the SoC 1604 may use the ring oscillator to detect the temperature of the CPU 1606, the GPU 1608, and / or the accelerator 1614. If the temperature is determined to exceed a threshold, the boot and power management processor may enter a temperature fault routine, place the SoC 1604 in a lower power state and / or place the vehicle 1600 in a Chauffeur safe shutdown mode (e.g., bring the vehicle 1600 to a safe shutdown).

[0158] The processor 1610 may further include a set of embedded processors that may perform the functions of an audio processing engine. The audio processing engine may be an audio subsystem that allows full hardware support of multi-channel audio through multiple interfaces and a wide and flexible range of audio I / O interfaces. In some instances, the audio processing engine is a dedicated processor core that includes a digital signal processor with dedicated RAM.

[0159] The processor 1610 may further include an always-on processor engine that can provide the necessary hardware features to support low power sensor management and wake use cases. The always-on processor engine may include a processor core, tightly coupled RAM, support peripherals (e.g., timers and interrupt controllers), various I / O controller peripherals, and routing logic.

[0160] The processor 1610 may further include a safety cluster engine that includes a processor subsystem dedicated to handling safety management of automotive applications. The safety cluster engine may include two or more processor cores, tightly coupled RAM, supporting peripherals (e.g., timers, interrupt controllers, etc.), and / or routing logic. In a safety mode, the two or more cores may operate in lockstep mode and function as a single core with comparison logic to detect any differences between their operations.

[0161] The processor 1610 may further include a real-time camera engine, which may include a dedicated processor subsystem for handling real-time camera management.

[0162] The processor 1610 may further include a high dynamic range signal processor, which may include an image signal processor, which is a hardware engine that is part of the camera processing pipeline.

[0163] The processor 1610 may include a video image compositor, which may be a processing block (e.g., implemented in a microprocessor) that implements video post-processing functions required by the video playback application to produce a final image for the player window. The video image compositor may perform lens distortion correction on the wide-view camera 1670, on the surround camera 1674, and / or on the in-cabin surveillance camera sensor. The in-cabin surveillance camera sensor is preferably monitored by a neural network running on another instance of the advanced SoC, configured to identify in-cabin events and respond appropriately. The in-cabin system may perform lip reading to activate cellular service and make phone calls, dictate emails, change the destination of the vehicle, activate or change the vehicle's infotainment system and settings, or provide voice-activated web surfing. Certain features are available to the driver only when operating in autonomous mode and are disabled otherwise.

[0164] The video image synthesizer may include enhanced temporal noise reduction for both spatial and temporal noise reduction. For example, when motion occurs in the video, the noise reduction reduces the weight of information provided by adjacent frames and appropriately weights the spatial information. When an image or part of an image does not contain motion, the temporal noise reduction performed by the video image synthesizer may use information from previous images to reduce noise in the current image.

[0165] The video image composer may also be configured to perform stereo rectification on the input stereo lens frames. The video image composer may further be used for user interface compositing when the operating system desktop is in use and the GPU 1608 is not required to continuously render new surfaces. Even when the GPU 1608 is powered on and actively performing 3D rendering, the video image composer may be used to offload the GPU 1608 for improved performance and responsiveness.

[0166] The SoC1604 may further include a mobile industry processor interface (MIPI) camera serial interface for receiving video and input from a camera, a high-speed interface, and / or a video input block that may be used for camera and related pixel input functions. The SoC1604 may further include an input / output controller that may be controlled by software and that may be used to receive I / O signals that are not committed to a specific role.

[0167] The SoC 1604 may further include a wide range of peripheral interfaces to enable communication with peripherals, audio codecs, power management, and / or other devices. The SoC 1604 may be used to process data from cameras (e.g., connected via gigabit multimedia serial links and Ethernet), sensors (e.g., LiDAR sensors 1664, RADAR sensors 1660, etc., which may be connected via Ethernet), data from the bus 1602 (e.g., vehicle 1600 speed, steering wheel position, etc.), and GNSS sensors 1658 (e.g., connected via Ethernet or CAN bus). The SoC 1604 may further include a dedicated high-performance mass storage controller that may include its own DMA engine and may be used to offload the CPU 1606 from routine data management tasks.

[0168] The SoC1604 may be an end-to-end platform with a flexible architecture that spans levels 3-5 of automation, thereby providing a comprehensive functional safety architecture that leverages and efficiently uses computer vision and ADAS techniques for diversity and redundancy, and provides a platform for a flexible, reliable driving software stack along with deep learning tools. The SoC1604 may be faster, more reliable, and more energy- and space-efficient than traditional systems. For example, when the accelerator 1614 is combined with the CPU 1606, the GPU 1608, and the data store 1616, the SoC1604 may provide a fast and efficient platform for levels 3-5 of autonomous vehicles.

[0169] Thus, the present technology provides capabilities and functionality that cannot be achieved by conventional systems. For example, computer vision algorithms may be executed on a CPU, which may be configured using a high-level programming language, such as the C programming language, to execute a wide variety of processing algorithms across a wide variety of visual data. However, CPUs often cannot meet the performance requirements of many computer vision applications, such as those related to execution time and power consumption. In particular, many CPUs cannot execute complex object detection algorithms in real time, which is a requirement for in-vehicle ADAS applications and for practical Level 3-5 autonomous vehicles.

[0170] In contrast to conventional systems, by providing a CPU complex, a GPU complex, and a hardware acceleration cluster, the techniques described herein allow multiple neural networks to run simultaneously and / or sequentially and the results to be combined to enable Level 3-5 autonomous driving capabilities. For example, a CNN running on the DLA or dGPU (e.g., GPU1620) can include text and word recognition that allows the supercomputer to read and understand traffic signs, including signs for which the neural network was not specifically trained. The DLA can further include neural networks that can identify, interpret, and provide a semantic understanding of the signs and pass the semantic understanding to a route planning module running on the CPU complex.

[0171] As another example, multiple neural networks may be run simultaneously, as required for level 3, 4, or 5 operation. For example, a warning sign consisting of "Caution: Flashing lights indicate icy conditions" along with a lightning flash may be interpreted by several neural networks, independently or collectively. The sign itself may be identified as a traffic sign by a first deployed neural network (e.g., a trained neural network), and the text "Flashing lights indicate icy conditions" may be interpreted by a second deployed neural network that informs the vehicle's route planning software (preferably running on a CPU complex) that icy conditions exist when a flashing light is detected. The flashing light may be identified by running a third deployed neural network through multiple frames, informing the vehicle's route planning software of the presence (or absence) of the flashing light. All three neural networks may be run simultaneously, such as within the DLA and / or on the GPU 1608.

[0172] In some instances, a CNN for facial recognition and vehicle owner identification can use data from the camera sensor to identify the presence of a legitimate driver and / or owner of the vehicle 1600. 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, to stop operation of the vehicle when the owner leaves the vehicle. In this manner, the SoC 1604 provides security against theft and / or vehicle takeover.

[0173] In another example, a CNN for emergency vehicle detection and identification can detect and identify emergency vehicle sirens using data from the microphone 1696. In contrast to conventional systems that use general classifiers to detect sirens and manually extract features, the SoC 1604 uses CNNs for classification of environmental and urban sounds, as well as classification of visual data. In a preferred embodiment, the CNN running on the DLA is trained to identify the relative terminal velocity of emergency vehicles (e.g., by using the Doppler effect). The CNN can also be trained to identify emergency vehicles specific to the local area in which the mobile vehicle is operating, as identified by the GNSS sensor 1658. Thus, for example, when operating in Europe, the CNN will attempt to detect European sirens, and when in the United States, the CNN will attempt to identify only North American sirens. After an emergency vehicle is detected, the control program may be used to execute emergency vehicle safety routines, such as slowing down the mobile vehicle, stopping the mobile vehicle at the side of the road, parking the mobile vehicle, and / or idling the mobile vehicle, with the assistance of the ultrasonic sensor 1662, until the emergency vehicle has passed.

[0174] The vehicle may include a CPU 1618 (e.g., a separate CPU, or a dCPU) that may be coupled to the SoC 1604 via a high-speed interconnect (e.g., PCIe). The CPU 1618 may include, for example, an X86 processor. The CPU 1618 may be used to perform any of a variety of functions including, for example, reconciling potentially inconsistent results between the ADAS sensors and the SoC 1604 and / or monitoring the status and health of the controller 1636 and / or the infotainment SoC 1630.

[0175] The vehicle 1600 may include a GPU 1620 (e.g., a discrete GPU, or a dGPU) that may be coupled to the SoC 1604 via a high-speed interconnect (e.g., NVIDIA's NVLINK). The GPU 1620 may provide additional artificial intelligence functionality, such as by running redundant and / or different neural networks, and may be used to train and / or update neural networks based on input (e.g., sensor data) from sensors of the vehicle 1600.

[0176] The mobile vehicle 1600 may further include a network interface 1624 that may include one or more wireless antennas 1626 (e.g., one or more wireless antennas for different communication protocols, such as a cellular antenna, a Bluetooth antenna, etc.). The network interface 1624 may be used to enable wireless connections with the cloud over the Internet (e.g., with a server 1678 and / or other network devices), with other mobile vehicles, and / or with computing devices (e.g., passenger client devices). To communicate with other mobile vehicles, a direct link may be established between the two mobile vehicles and / or an indirect link may be established (e.g., through a network and via the Internet). A direct link may be provided using a mobile vehicle-to-mobile communication link. The mobile vehicle-to-mobile communication link may provide the mobile vehicle 1600 information about mobile vehicles in its vicinity (e.g., mobile vehicles in front of, beside, and / or behind the mobile vehicle 1600). This feature may be part of a collaborative adaptive cruise control feature of the vehicle 1600 .

[0177] The network interface 1624 may include a SoC that provides modulation and demodulation functions and enables the controller 1636 to communicate over a wireless network. The network interface 1624 may include a radio frequency front end for upconversion from baseband to radio frequency and downconversion from radio frequency to baseband. The frequency conversion may be performed through well known processes and / or may be performed using a superheterodyne process. In some instances, the radio frequency front end functions may be provided by a separate chip. The network interface may include wireless functions for communicating via LTE, WCDMA, UMTS, GSM, CDMA2000, Bluetooth, Bluetooth LE, Wi-Fi, Z-Wave, ZigBee, LoRaWAN, and / or other wireless protocols.

[0178] The vehicle 1600 may further include a data store 1628, which may include off-chip (e.g., off-SoC 1604) storage. The data store 1628 may include one or more memory elements, including RAM, SRAM, DRAM, VRAM, flash, hard disk, and / or other components and / or devices capable of storing at least one bit of data.

[0179] The vehicle 1600 may further include a GNSS sensor 1658. The GNSS sensor 1658 (e.g., a GPS, an aided GPS sensor, a differential GPS (DGPS) sensor, etc.) aids in mapping, perception, occupancy grid generation, and / or route planning functions. Any number of GNSS sensors 1658 may be used, including, but not limited to, a GPS using a USB connector with an Ethernet to serial (RS-232) bridge, for example.

[0180] The mobile vehicle 1600 may further include a RADAR sensor 1660. The RADAR sensor 1660 may be used by the mobile vehicle 1600 for long-range mobile vehicle detection, even in darkness and / or severe weather conditions. The RADAR functional safety level may be ASIL B. In some instances, the RADAR sensor 1660 may use the CAN and / or bus 1602 for control and to access object tracking data (e.g., to transmit data generated by the RADAR sensor 1660), with access to Ethernet to access raw data. A wide variety of RADAR sensor types may be used. For example, and without limitation, the RADAR sensor 1660 may be suitable for front, rear, and side RADAR use. In some instances, a pulsed Doppler RADAR sensor is used.

[0181] The RADAR sensor 1660 may include different configurations, such as long range with a narrow field of view, short range with a wide field of view, and short range side coverage. In some instances, the long range RADAR may be used for adaptive cruise control functions. Long range RADAR systems may provide a wide field of view achieved by two or more independent scans, such as within a 250m range. The RADAR sensor 1660 may help distinguish between static and moving objects and may be used by ADAS systems for emergency brake assist and forward collision warning. The long range RADAR sensor may include a monostatic multimodal RADAR with multiple (e.g., six or more) fixed RADAR antennas and high speed CAN and FlexRay interfaces. In one instance with six antennas, the center four antennas may create a focused beam pattern designed to record the surroundings of the moving vehicle 1600 at high speeds with minimal interference from traffic in adjacent lanes. The other two antennas may widen the field of view, allowing for rapid detection of moving vehicles entering or leaving the lane of the moving vehicle 1600.

[0182] As an example, a medium-range RADAR system may include a range of up to 1660m (front) or 80m (rear) and a field of view of up to 42 degrees (front) or 1650 degrees (rear). A short-range RADAR system may include, but is not limited to, a RADAR sensor designed to be mounted on either end of a rear bumper. When mounted on either end of a rear bumper, such a RADAR sensor system can create two beams that constantly monitor the blind spots behind and adjacent to a moving vehicle.

[0183] Short-range RADAR systems may be used in ADAS systems for blind spot detection and / or lane change assist.

[0184] The mobile vehicle 1600 may further include ultrasonic sensors 1662. The ultrasonic sensors 1662, which may be located at the front, rear, and / or sides of the mobile vehicle 1600, may be used for parking assist and / or for creating and updating an occupancy grid. A wide variety of ultrasonic sensors 1662 may be used, with different ultrasonic sensors 1662 being used for different ranges of detection (e.g., 2.5m, 4m). The ultrasonic sensors 1662 may operate at a functional safety level of ASIL B.

[0185] The mobile vehicle 1600 may include a LiDAR sensor 1664. The LiDAR sensor 1664 may be used for object and pedestrian detection, emergency braking, collision avoidance, and / or other functions. The LiDAR sensor 1664 may be functional safety level ASIL B. In some instances, the mobile vehicle 1600 may include multiple (e.g., 2, 4, 6, etc.) LiDAR sensors 1664 that may use Ethernet (e.g., to provide data to a Gigabit Ethernet switch).

[0186] In some instances, the LiDAR sensor 1664 may have the ability to provide a list of objects and their distances in a 360-degree field of view. A commercially available LiDAR sensor 1664 may have an advertised range of about 1600 m, for example, with an accuracy of 2 cm to 3 cm, and with support for a 1600 Mbps Ethernet connection. In some instances, one or more non-protruding LiDAR sensors 1664 may be used. In such instances, the LiDAR sensor 1664 may be implemented as a small device that may be integrated into the front, rear, sides, and / or corners of the moving vehicle 1600. In such instances, the LiDAR sensor 1664 may have a range of 200 m even for low-reflecting objects and provide up to 120 degrees horizontal and 35 degrees vertical field of view. A front-mounted LiDAR sensor 1664 may be configured for a horizontal field of view between 45 degrees and 135 degrees.

[0187] In some instances, LiDAR technology such as 3D Flash LiDAR may also be used. 3D Flash LiDAR uses a laser flash as a transmitter to illuminate the surroundings of the moving vehicle up to about 200 m. The Flash LiDAR unit includes a receptor that records the laser pulse transit time and the reflected light on each pixel, which in turn corresponds to the range from the moving vehicle to the object. Flash LiDAR may allow a highly accurate and distortion-free image of the surroundings to be generated with every laser flash. In some instances, four Flash LiDAR sensors may be deployed, one on each side of the moving vehicle 1600. Available 3D Flash LiDAR systems include solid-state 3D steering array LiDAR cameras that have no moving parts other than the blower (e.g., non-scanning LiDAR devices). Flash LiDAR devices may use 5 nanosecond Class I (eye-safe) laser pulses per frame and may capture reflected laser light in the form of a 3D range point cloud and coregistered intensity data. By using flash LiDAR, and because flash LiDAR is a solid-state device with no moving parts, the LiDAR sensor 1664 may be less susceptible to motion blur, vibration, and / or shock.

[0188] The mobile vehicle may further include an IMU sensor 1666. In some instances, the IMU sensor 1666 may be positioned at the center of the rear axle of the mobile vehicle 1600. The IMU sensor 1666 may include, for example, but not limited to, an accelerometer, a magnetometer, a gyroscope, a magnetic compass, and / or other sensor types. In some instances, such as in a 6-axis application, the IMU sensor 1666 may include an accelerometer and a gyroscope, while in a 9-axis application, the IMU sensor 1666 may include an accelerometer, a gyroscope, and a magnetometer.

[0189] In some embodiments, the IMU sensor 1666 may be implemented as a miniature, high-performance GPS-Aided Inertial Navigation System (GPS / INS) that combines micro-electro-mechanical system (MEMS) inertial sensors, highly sensitive GPS receivers, and advanced Kalman filtering algorithms to provide estimates of position, velocity, and attitude. As such, in some instances, the IMU sensor 1666 may enable the vehicle 1600 to estimate heading without requiring input from a magnetic sensor by directly observing and correlating changes in velocity from the GPS to the IMU sensor 1666. In some instances, the IMU sensor 1666 and the GNSS sensor 1658 may be combined in a single integrated unit.

[0190] The vehicle may include microphones 1696 placed in and / or around the vehicle 1600. The microphones 1696 may be used for emergency vehicle detection and identification, among other things.

[0191] The vehicle may further include any number of camera types, including stereo cameras 1668, wide view cameras 1670, infrared cameras 1672, surround cameras 1674, long and / or mid-range cameras 1698, and / or other camera types. The cameras may be used to capture image data around the entire exterior of the vehicle 1600. The type of cameras used depends on the embodiment and requirements of the vehicle 1600, and any combination of camera types may be used to achieve the required coverage around the vehicle 1600. 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. The cameras may support, by way of example and without limitation, Gigabit Multimedia Serial Link (GMSL) and / or Gigabit Ethernet. Each camera is described in further detail herein in conjunction with Figures 16A and 16B.

[0192] The mobile vehicle 1600 may further include a vibration sensor 1642. The vibration sensor 1642 may measure vibration of a component of the mobile vehicle, such as an axle. For example, a change in vibration may indicate a change in the surface of the road. In another example, when two or more vibration sensors 1642 are used, the difference in vibration may be used to determine friction or slippage of the road surface (e.g., when the difference in vibration is between a powered axle and a free-running axle).

[0193] The mobile vehicle 1600 may include an ADAS system 1638. In some instances, the ADAS system 1638 may include a SoC. The ADAS system 1638 may include autonomous / adaptive / automatic cruise control (ACC), cooperative adaptive cruise control (CACC), forward crash warning (FCW), automatic emergency braking (AEB), lane departure warning (LDW), lane keep assist (LKA), blind spot warning (BSW), rear cross-traffic warning (RCTW), collision warning system (CWS), lane centering (LC), and / or other features and functions.

[0194] The ACC system may use a RADAR sensor 1660, a LiDAR sensor 1664, and / or a camera. The ACC system may include longitudinal ACC and / or lateral ACC. The longitudinal ACC monitors and controls the distance to the vehicle directly ahead of the vehicle 1600 and automatically adjusts the vehicle speed to maintain a safe distance from the vehicle ahead. The lateral ACC performs distance keeping and advises the vehicle 1600 to change lanes when necessary. The lateral ACC is related to other ADAS applications such as LCA and CWS.

[0195] CACC uses information from other moving vehicles, which may be received from other moving vehicles via a wireless link via the network interface 1624 and / or the wireless antenna 1626, or indirectly via a network connection (e.g., via the Internet). A direct link may be provided by a vehicle-to-vehicle (V2V) communication link, while an indirect link may be an infrastructure-to-vehicle (I2V) communication link. In general, a V2V communication concept provides information about the immediately preceding moving vehicle (e.g., the moving vehicle directly ahead of the moving vehicle 1600, in the same lane as the moving vehicle 1600), while an I2V communication concept provides information about the traffic further ahead. A CACC system may include either or both I2V and V2V information sources. Given the information of moving vehicles ahead of the moving vehicle 1600, CACC may be more reliable and may potentially make traffic flow smoother and reduce congestion on the roads.

[0196] FCW systems are designed to warn the driver of hazards so that the driver can take corrective action. FCW systems use forward-facing cameras and / or RADAR sensors 1660 coupled to dedicated processors, DSPs, FPGAs, and / or ASICs electrically coupled to driver feedback such as displays, speakers, and / or vibrating components. FCW systems can provide warnings in the form of audio, visual alerts, vibrations, and / or quick brake pulses.

[0197] An AEB system can detect an imminent forward collision with another moving vehicle or other object and automatically apply the brakes if the driver does not take corrective action within specified time or distance parameters. The AEB system can use a forward-facing camera and / or RADAR sensor 1660 coupled to a dedicated processor, DSP, FPGA, and / or ASIC. When the AEB system detects a hazard, the AEB system 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 as part of an effort to prevent, or at least mitigate, the effects of the predicted collision. The AEB system can include techniques such as Dynamic Brake Support and / or Collision Imminent Braking.

[0198] The LDW system provides visual, audible, and / or tactile warnings, such as vibrations of the steering wheel or seat, to alert the driver when the mobile vehicle 1600 crosses a lane marking. The LDW system does not activate when the driver indicates an intentional lane departure by activating a turn signal. The LDW system may use a forward-facing camera coupled to a dedicated processor, DSP, FPGA, and / or ASIC electrically coupled to driver feedback, such as a display, speaker, and / or vibrating components.

[0199] The LKA system is a modification of the LDW system, which provides steering input or braking to correct the vehicle 1600 if it begins to wander out of its lane.

[0200] The BSW system detects and warns the driver of a moving vehicle in the vehicle's blind spot. The BSW system can provide visual, audible, and / or tactile warnings to indicate that merging or changing lanes is unsafe. The system can provide additional warnings when the driver uses turn signals. The BSW system can use rear-facing cameras and / or RADAR sensors 1660 coupled to dedicated processors, DSPs, FPGAs, and / or ASICs electrically coupled to driver feedback, e.g., displays, speakers, and / or vibration components.

[0201] The RCTW system may provide visual, audible, and / or tactile notifications when an object is detected outside the range of the rear camera when the vehicle 1600 is backing up. Some RCTW systems include AEB to ensure vehicle brakes are applied to avoid a collision. The RCTW system may use one or more rear-facing RADAR sensors 1660 coupled to a dedicated processor, DSP, FPGA, and / or ASIC electrically coupled to driver feedback, e.g., a display, speaker, and / or vibration components.

[0202] Because conventional ADAS systems alert the driver and allow the driver to determine whether a safety condition truly exists and act accordingly, conventional ADAS systems can be prone to producing false positive results that are usually not catastrophic but can be annoying and distracting to the driver. However, in an autonomous vehicle 1600, the vehicle 1600 itself must decide whether to listen to the results from the primary computer or the secondary computer (e.g., the first controller 1636 or the second controller 1636) when the results are conflicting. For example, in some embodiments, the ADAS system 1638 may be a backup and / or secondary computer to provide perception information to a 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. Output from the ADAS system 1638 may be provided to a supervisory MCU. When the outputs from the primary and secondary computers conflict, the supervisory MCU must decide how to reconcile the conflict to ensure safe operation.

[0203] In some instances, the primary computer may be configured to provide a reliability score to the overseer MCU indicating the reliability of the primary computer in a selected outcome. If the reliability score exceeds a threshold, the overseer MCU may follow the instructions of the primary computer regardless of whether the secondary computers provide conflicting or inconsistent results. If the reliability score does not meet the threshold, and if the primary and secondary computers show different (e.g., conflicting) results, the overseer MCU may arbitrate between the computers to determine the appropriate outcome.

[0204] The supervisory MCU may be configured to execute a neural network that is trained and configured to determine conditions under which the secondary computer provides a false alarm based on the outputs from the primary and secondary computers. Thus, the neural network in the supervisory MCU may learn when the output of the secondary computer may be trusted and when it may not be trusted. For example, when the secondary computer is a RADAR-based FCW system, the neural network in the supervisory MCU may learn when the FCW identifies a metal object that is not actually a hazard, such as a sewer grate or manhole cover, that triggers an alarm. Similarly, when the secondary computer is a camera-based LDW system, the neural network in the supervisory MCU may learn to ignore the LDW when a bicyclist or pedestrian is present and lane departure is, in fact, the safest maneuver. In embodiments that include a neural network executing on the supervisory MCU, the supervisory MCU may include at least one of a DLA or a GPU suitable for executing the neural network with associated memory. In a preferred embodiment, the director MCU may comprise and / or be included as a component of the SoC1604.

[0205] In other instances, the ADAS system 1638 may include a secondary computer that performs ADAS functions using classical rules of computer vision. As such, the secondary computer may use classical computer vision rules (if-then) and the presence of a neural network in the supervisory MCU may improve reliability, safety and performance. For example, the diverse implementations and intentional non-identity may make the overall system more fault-tolerant, especially to failures caused by software (or software-hardware interface) functions. For example, if there is a software bug or error in the software running on the primary computer and non-identical software code running on the secondary computer provides the same overall result, the supervisory MCU may have greater confidence that the overall result is correct and that a bug in the software or hardware on the primary computer has not caused a critical error.

[0206] In some instances, the output of the ADAS system 1638 may be fed to a perception block of the primary computer and / or a dynamic driving task block of the primary computer. For example, if the ADAS system 1638 indicates a forward collision warning due to an object directly ahead, the perception block may use this information when identifying the object. In other instances, the secondary computer may have its own neural network that is trained as described herein, thus reducing the risk of false positives.

[0207] The mobile vehicle 1600 may further include an infotainment SoC 1630 (e.g., an in-vehicle infotainment system (IVI)). Although shown and described as an SoC, the infotainment system need not be an SoC and may include two or more separate components. The infotainment SoC 1630 may include a combination of hardware and software that may be used to provide audio (e.g., music, personal digital assistant, navigation instructions, news, radio, etc.), video (e.g., TV, movies, streaming, etc.), telephony (e.g., hands-free calling), network connectivity (e.g., LTE, Wi-Fi, etc.), and / or information services (e.g., navigation system, rear park assist, wireless data system, mobile vehicle related information such as fuel level, total distance traveled, brake fuel level, oil level, door opening / closing, air filter information, etc.) to the mobile vehicle 1600. For example, the infotainment SoC 1630 may be a radio, a disc player, a navigation system, a video player, USB and Bluetooth connectivity, a car computer, in-car entertainment, Wi-Fi, steering wheel audio controls, hands-free voice control, a heads-up display (HUD), an HMI display 1634, 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 1630 may further be used to provide information (e.g., visual and / or audible) to a user of the vehicle, such as information from an ADAS system 1638, autonomous driving information such as planned vehicle maneuvers, trajectory, surrounding environment information (e.g., intersection information, vehicle information, road information, etc.), and / or other information.

[0208] The infotainment SoC 1630 may include GPU functionality. The infotainment SoC 1630 may communicate with other devices, systems, and / or components of the mobile vehicle 1600 via the bus 1602 (e.g., CAN bus, Ethernet, etc.). In some instances, the infotainment SoC 1630 may be coupled to a supervisory MCU such that the infotainment system's GPU can perform some self-drive functions in the event that the primary controller 1636 (e.g., the primary and / or backup computer of the mobile vehicle 1600) fails. In such instances, the infotainment SoC 1630 can place the mobile vehicle 1600 in a Chauffeur safe stop mode, as described herein.

[0209] The mobile vehicle 1600 may further include an instrument cluster 1632 (e.g., a digital dash, an electronic instrument cluster, a digital instrument panel, etc.). The instrument cluster 1632 may include a controller and / or a supercomputer (e.g., a separate controller or a supercomputer). The instrument cluster 1632 may include a set of instruments such as a speedometer, fuel level, oil pressure, tachometer, odometer, turn signals, gear shift position indicator, seat belt warning light, parking brake warning light, engine malfunction light, airbag (SRS) system information, lighting controls, safety system controls, navigation information, etc. In some instances, information may be displayed and / or shared between the infotainment SoC 1630 and the instrument cluster 1632. In other words, the instrument cluster 1632 may be included as part of the infotainment SoC 1630, or vice versa.

[0210] 16D is a system diagram of communication between a cloud-based server and an example autonomous vehicle 1600 of FIG. 16A in accordance with some embodiments of the present disclosure. The system 1676 may include a server 1678, a network 1690, and a mobile vehicle including the mobile vehicle 1600. The server 1678 may include a number of GPUs 1684(A)-1684(H) (collectively referred to herein as GPUs 1684), PCIe switches 1682(A)-1682(H) (collectively referred to herein as PCIe switches 1682), and / or CPUs 1680(A)-1680(B) (collectively referred to herein as CPUs 1680). The GPUs 1684, CPUs 1680, and PCIe switches may be interconnected with a high-speed interconnect, such as, but not limited to, an NVLINK interface 1688 and / or a PCIe connection 1686 developed by NVIDIA. In some instances, the GPUs 1684 are connected via an NVLINK and / or NVSwitch SoC, and the GPUs 1684 and PCIe switch 1682 are connected via a PCIe interconnect. Although eight GPUs 1684, two CPUs 1680, and two PCIe switches are illustrated, this is not intended to be limiting. Depending on the embodiment, each server 1678 may include any number of GPUs 1684, CPUs 1680, and / or PCIe switches. For example, the servers 1678 may each include 8, 16, 32, and / or more GPUs 1684.

[0211] The server 1678 can receive image data from the mobile vehicle over the network 1690, the image data representing images showing unexpected or changed road conditions, such as recently started road construction. The server 1678 can transmit the neural network 1692, the updated neural network 1692, and / or the map information 1694, including information about traffic and road conditions, to the mobile vehicle over the network 1690. The updates to the map information 1694 can include updates to the HD map 1622, such as information about construction sites, potholes, detours, flooding, and / or other obstacles. In some instances, the neural network 1692, the updated neural network 1692, and / or the map information 1694 may have arisen from new training and / or experience represented in data received from any number of mobile vehicles in the environment and / or based on training performed at a data center (e.g., using the server 1678 and / or other servers).

[0212] The server 1678 may be used to train a machine learning model (e.g., a neural network) based on the training data. The training data may be generated by a moving vehicle and / or generated in a simulation (e.g., using a game engine). In some instances, the training data is tagged (e.g., if the neural network benefits from supervised learning) and / or undergoes other pre-processing, while in other instances, the training data is not tagged and / or pre-processed (e.g., if the neural network does not require supervised learning). The training may be performed according to any one or more classes of machine learning techniques, including, but not limited to, the following classes: supervised training, semi-supervised training, unsupervised training, self-learning, reinforcement learning, federated learning, transfer learning, feature learning (including principal component and cluster analysis), multi-linear subspace learning, manifold learning, representation learning (including preliminary dictionary learning), rule-based machine learning, anomaly detection, and variations or combinations thereof. After the machine learning model has been traced, the machine learning model may be used by the vehicle (e.g., transmitted to the vehicle via network 1690) and / or the machine learning model may be used by server 1678 to remotely monitor the vehicle.

[0213] In some instances, the server 1678 can receive data from the moving vehicles and apply the data to state-of-the-art real-time neural networks for real-time intelligent inference. The server 1678 can include deep learning supercomputers and / or dedicated AI computers powered by GPUs 1684, such as the DGX and DGX Station machines developed by NVIDIA. However, in some instances, the server 1678 can include a deep learning infrastructure that uses only CPU-powered data centers.

[0214] The deep learning infrastructure of the server 1678 may be capable of rapid real-time inference and may use that capability to evaluate and validate the health of the processor, software, and / or associated hardware within the mobile vehicle 1600. For example, the deep learning infrastructure may receive periodic updates from the mobile vehicle 1600, such as sequences of images and / or objects that the mobile vehicle 1600 has located within the images of that sequence (e.g., via computer vision and / or other machine learning object classification techniques). The deep learning infrastructure may run its own neural networks to identify objects and compare them with objects identified by the mobile vehicle 1600, and if the results do not match and the infrastructure concludes that the AI ​​within the mobile vehicle 1600 is not functioning properly, the server 1678 may send a signal to the mobile vehicle 1600 to infer control, notify passengers, and command the mobile vehicle's 1600 failsafe computer to complete a safe parking maneuver.

[0215] For inference, the server 1678 may include a GPU 1684 and one or more programmable inference accelerators (e.g., NVIDIA's TensorRT). The combination of a GPU-powered server and inference acceleration can enable real-time responsiveness. In other instances, such as when less performance is required, servers powered by CPUs, FPGAs, and other processors may be used for inference.

[0216] Exemplary Computing Device 17 is a block diagram of an example computing device 1700 suitable for use in implementing some embodiments of the present disclosure. The computing device 1700 may include an interconnection system 1702 that indirectly or directly couples the following devices: memory 1704, one or more central processing units (CPUs) 1706, one or more graphic processing units (GPUs) 1708, a communication interface 1710, input / output (I / O) ports 1712, input / output components 1714, a power supply 1716, one or more presentation components 1718 (e.g., displays), and one or more logic units 1720.

[0217] Although the various blocks in FIG. 17 are shown connected via interconnect system 1702 with lines, this is not intended to be limiting and is merely for clarity. For example, in some embodiments, a presentation component 1718, such as a display device, may be considered an I / O component 1714 (e.g., if the display is a touch screen). As another example, CPU 1706 and / or GPU 1708 may include memory (e.g., memory 1704 may represent a storage device in addition to the memory of GPU 1708, CPU 1706, and / or other components). In other words, the computing devices of FIG. 17 are merely exemplary. Categories such as “workstations,” “servers,” “laptops,” “desktops,” “tablets,” “client devices,” “mobile devices,” “handheld devices,” “gaming consoles,” “electronic control units (ECUs),” “virtual reality systems,” and / or other device or system types are all intended to be within the scope of the computing devices of FIG. 17 and therefore are not to be differentiated.

[0218] The interconnect system 1702 may represent one or more links or buses, such as an address bus, a data bus, a control bus, or a combination thereof. The interconnect system 1702 may include one or more bus or link types, such as an industry standard architecture (ISA) bus, an extended industry standard architecture (EISA) bus, a video electronics standards association (VESA) bus, a peripheral component interconnect (PCI) bus, a peripheral component interconnect express (PCIe) bus, and / or another type of bus or link. In some embodiments, there are direct connections between components. As an example, the CPU 1706 may be directly connected to the memory 1704. Additionally, the CPU 1706 may be directly connected to the GPU 1708. When there is a direct or point-to-point connection between components, the interconnect system 1702 may include a PCIe link to implement the connection. In these instances, a PCI bus need not be included in the computing device 1700 .

[0219] Memory 1704 may include any of a variety of computer-readable media. Computer-readable media may be any available media that can be accessed by computing device 1700. Computer-readable media may include both volatile and nonvolatile media, and removable and non-removable media. By way of example, and not limitation, computer-readable media may comprise computer storage media and communication media.

[0220] Computer storage media may include both 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 1704 may store computer readable instructions (e.g., representing programs and / or 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 disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that may be used to store the desired information and that may be accessed by computing device 1700. As used herein, computer storage media does not include the signals themselves.

[0221] Computer storage media may embody computer-readable instructions, data structures, program modules, and / or other data types in a modulated data signal such as a carrier wave or other transport mechanism and include 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 may 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.

[0222] The CPU 1706 may be configured to execute at least some of the computer readable instructions to control one or more components of the computing device 1700 to perform one or more of the methods and / or processes described herein. The CPU 1706 may include one or more (e.g., 1, 2, 4, 8, 28, 72, etc.) cores each capable of processing multiple software threads simultaneously. The CPU 1706 may include any type of processor, and may include different types of processors depending on the type of computing device 1700 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 1700, the processor may be an Advanced RISC Machines (ARM) processor implemented using Reduced Instruction Set Computing (RISC) or an x86 processor implemented using Complex Instruction Set Computing (CISC). The computing device 1700 may include one or more CPUs 1706 within one or more microprocessors or auxiliary coprocessors, such as computational coprocessors.

[0223] In addition to or in lieu of the CPU 1706, the GPU 1708 may be configured to execute at least some of the computer readable instructions to control one or more components of the computing device 1700 to perform one or more of the methods and / or processes described herein. One or more of the GPUs 1708 may be integrated GPUs (e.g., with one or more of the CPUs 1706 and / or one or more of the GPUs 1708 may be discrete GPUs. In an embodiment, one or more of the GPUs 1708 may be a co-processor or co-processors of the CPUs 1706. The GPUs 1708 may be used by the computing device 1700 to render graphics (e.g., 3D graphics) or perform general purpose computations. For example, the GPUs 1708 may be configured to perform general purpose computing on a GPU (GPGPU). A graphics processing unit (GPU) 1708 may be used for processing graphics data. The GPU 1708 may include hundreds or thousands of cores capable of processing hundreds or thousands of software threads simultaneously. The GPU 1708 may generate pixel data for an output image in response to rendering commands (e.g., rendering commands from the CPU 1706 received via a host interface). The GPU 1708 may include graphics memory, e.g., display memory, for storing the pixel data or any other suitable data, e.g., GPGPU data. The display memory may be included as part of the memory 1704. GPU 1708 may include two or more GPUs operating in parallel (e.g., via links). The links may connect the GPUs directly (e.g., using NVLINK) or may connect the GPUs via a switch (e.g., using NVSwitch). When coupled together, each GPU 1708 may generate pixel data or GPGPU data for a different portion of the output or for a different output (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 the other GPUs.

[0224] In addition to or instead of the CPU 1706 and / or GPU 1708, the logic unit 1720 may be configured to execute at least some of the computer readable instructions to control one or more of the computing devices 1700 to perform one or more of the methods and / or processes described herein. In an embodiment, the CPU 1706, the GPU 1708, and / or the logic unit 1720 may execute any combination of the methods, processes, and / or portions thereof, either discretely or jointly. One or more of the logic units 1720 may be part of and / or integrated with one or more of the CPU 1706 and / or GPU 1708, and / or one or more of the logic units 1720 may be discrete components to or otherwise external to the CPU 1706 and / or GPU 1708. In an embodiment, one or more of the logic units 1720 may be a co-processor of one or more of the CPU 1706 and / or GPU 1708.

[0225] Examples of logic unit 1720 include one or more processing cores and / or components thereof, such as a Tensor Core (TC), a Tensor Processing Unit (TPU), a Pixel Visual Core (PVC), a Vision Processing Unit (VPU), a Graphics Processing Cluster (GPC), a Texture Processing Cluster (TPC), a Streaming Multiprocessor (SM), a Tree Traversal 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 Peripheral Component Interconnect Express (PCIe) element, and / or the like.

[0226] Communications interface 1710 may include one or more receivers, transmitters, and / or transceivers that enable computing device 1700 to communicate with other computing devices over electronic communications networks, including wired and / or wireless communications. Communications interface 1710 may include components and functionality to 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., communicating over Ethernet or InfiniBand), a low power wide area network (e.g., LoRaWAN, SigFox, etc.), and / or the Internet.

[0227] The I / O ports 1712 may enable the computing device 1700 to be logically coupled to other devices, including I / O components 1714, presentation components 1718, and / or other components, some of which may be built-in (e.g., integrated) to the computing device 1700. Exemplary I / O components 1714 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 1714 may provide a natural user interface (NUI) that processes air gestures, voice, or other physiological input generated by a user. In some cases, the input may be sent to an appropriate network element for further processing. The NUI may implement any combination of voice recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition on and next to the screen, air gestures, head and eye tracking, and touch recognition in conjunction with the display of the computing device 1700 (as described in more detail below). The computing device 1700 may include a depth camera, such as a stereoscopic camera system, an infrared camera system, an RGB camera system, touch screen technology, and combinations thereof, for gesture detection and recognition. Additionally, the computing device 1700 may include an accelerometer or gyroscope (e.g., as part of an inertia measurement unit (IMU)) to enable detection of motion. In some instances, the output of the accelerometer or gyroscope may be used by the computing device 1700 to render immersive augmented or virtual reality.

[0228] The power supply 1716 may include a hardwired power supply, a battery power supply, or a combination thereof. The power supply 1716 may provide power to the computing device 1700 to enable the components of the computing device 1700 to operate.

[0229] The presentation component 1718 may include a display (e.g., a monitor, a touch screen, a television screen, a heads-up display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. The presentation component 1718 can receive data from other components (e.g., GPU 1708, CPU 1706, etc.) and output data (e.g., as images, video, sound, etc.).

[0230] The present disclosure may be described in the general context of computer code or machine usable instructions, including computer executable instructions, such as program modules, being executed by a computer or other machine, such as a personal digital assistant or other handheld device. Generally, program modules, including routines, programs, objects, components, data structures, etc., refer to code that performs particular tasks or implements particular abstract data types. The present disclosure may be implemented in a variety of configurations, including handheld devices, consumer electronics, general purpose computers, more specialized computing devices, etc. The present disclosure may also be implemented in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.

[0231] As used herein, the term "and / or" with respect to two or more elements should be construed to mean only one element or a combination of elements. For example, "element A, element B, and / or element C" may include element A only, element B only, element C only, elements A and element B, elements A and element C, elements 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. Furthermore, "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.

[0232] The subject matter of the present disclosure has been described with specificity to meet statutory requirements. However, the description itself is not intended to limit the scope of the present disclosure. Instead, the inventors intend that the claimed subject matter may be implemented in other ways, including different steps or combinations of steps similar to those described in this document, in conjunction with other current or future technologies. Furthermore, although the terms "step" and / or "block" may be used herein to connote different elements of the method used, these terms should not be construed as implying any particular order between the various steps disclosed herein unless and when the order of the individual steps is explicitly described.

Claims

1. A step of determining first data representing a first view of an environment from accumulated sensor data from one or more LiDAR sensors of an ego actor over a period of time, the first data being based at least on converting the accumulated sensor data into motion-corrected sensor data corresponding to a location of the ego actor at a particular time, and projecting the motion-corrected sensor data into a two-dimensional (2D) image space to generate the first data representing a LiDAR range image of the first view of the environment; extracting classification data representative of one or more classifications of objects or scenes shown in the first view based on at least the first data using one or more neural networks (NN); generating transformed classification data representative of the one or more classifications in the second view of the environment based at least on projecting the one or more classifications from the first view onto a second view; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data; A method comprising:

2. The method of claim 1 , wherein the first view is a perspective view and the second view is a top-down view.

3. 2. The method of claim 1 , wherein the first data representing the first view of the environment comprises a projection of a LiDAR point cloud representing a perspective view of the environment, and wherein the projecting the one or more classifications from the first view onto the second view comprises projecting the one or more classifications from the perspective view onto a top-down view of the environment using the LiDAR point cloud.

4. The method of claim 1, wherein the step of determining the first data includes projecting a LiDAR point cloud onto the LiDAR range image.

5. The method of claim 1, wherein the LiDAR range image has a height in pixels corresponding to several horizontal scan lines of the LiDAR sensor that captured the sensor data.

6. The method of claim 1 , wherein the projecting the one or more classifications from the first view onto the second view comprises applying a differentiable transformation to 3D positions associated with the classification data.

7. A method for detecting a first view of an environment, comprising: determining first data from sensor data, the first data representing a first view of the environment; extracting classification data representative of one or more classifications of objects or scenes shown in the first view based on at least the first data using one or more neural networks (NN); generating transformed classification data representative of the one or more classifications in the second view of the environment based at least on projecting the one or more classifications from the first view onto a second view; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data; A method comprising: The sensor data represents a LiDAR point cloud, the transformed classification data represents one or more confidence maps in the second view, and the method further comprises: generating third data representing one or more height maps based at least on projecting the LiDAR point cloud onto the second view; forming a tensor comprising a first set of one or more channels storing the transformed classification data representing the one or more confidence maps and a second set of one or more channels storing the third data representing the one or more height maps; extracting second classification data representative of one or more second classifications in the second view and fourth data representative of object instance geometry of the one or more objects from the tensor using the one or more neural networks; The method further comprising:

8. A method of detecting an environment, comprising: determining first data from sensor data, the first data representing a first view of the environment; extracting classification data representative of one or more classifications of objects or scenes shown in the first view based on at least the first data using one or more neural networks (NN); generating transformed classification data representative of the one or more classifications in the second view of the environment based at least on projecting the one or more classifications from the first view onto a second view; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data; decoding the output of the one or more neural networks to generate candidate boundary shapes for the one or more objects; identifying the second data representative of the one or more bounding shapes of the one or more objects based on at least one of filtering or clustering candidate bounding boxes to remove duplicate candidates from the candidate bounding boxes; assigning a class label for each of the one or more boundary shapes based on the output of the one or more neural networks; A method comprising:

9. A method for detecting a first view of an environment, comprising: determining first data from sensor data, the first data representing a first view of the environment; extracting classification data representative of one or more classifications of objects or scenes shown in the first view based on at least the first data using one or more neural networks (NN); generating transformed classification data representative of the one or more classifications in the second view of the environment based at least on projecting the one or more classifications from the first view onto a second view; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data; Including, 20. The method of claim 19, further comprising: decoding the output of the one or more neural networks to generate candidate boundary shapes for the one or more objects; identifying the second data representing the one or more boundary shapes of the one or more objects based on performing at least one of non-maximum suppression of noisy applications or density-based spatial clustering to remove duplicate candidates from the candidate boundary shapes; A method comprising:

10. A method for detecting an environment comprising: determining first data from sensor data, the first data representing a first view of the environment; extracting classification data representative of one or more classifications of objects or scenes shown in the first view based on at least the first data using one or more neural networks (NN); generating transformed classification data representative of the one or more classifications in the second view of the environment based at least on projecting the one or more classifications from the first view onto a second view; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data; Including, 13. The method of claim 12, wherein an output of the one or more NNs comprises a tensor storing regressed geometry data for each detected object, and wherein the step of determining second data representing the one or more boundary shapes comprises generating one or more 3D boundary shapes of the one or more objects from the regressed geometry data.

11. 10. The method of claim 1, further comprising training the one or more NNs using training data generated using annotation tracking to track annotated objects between two or more frames of corresponding sensor data.

12. 10. The method of claim 1, further comprising: training the one or more NNs using training data generated using links between object trajectories generated for a particular object from corresponding sensor data from two or more sensors.

13. receiving LiDAR data from one or more LiDAR sensors in an environment; generating first data from the LiDAR data representing a perspective view of the environment; generating classification data from the first data using one or more neural networks (NNs), the classification data representing one or more classifications in the perspective views; generating transformed classification data representative of the one or more classifications in the top-down view of the environment by projecting the one or more classifications in the perspective view to a top-down view using the LiDAR data; generating, using the one or more neural networks, second data representative of one or more boundary shapes of one or more objects detected in the environment based at least on the transformed classification data in the top-down view; Including, said generating said first data representing said perspective view of said environment further comprises: accessing accumulated sensor data from the one or more LiDAR sensors of an ego actor, the data having been accumulated over a period of time; converting the accumulated sensor data into motion-corrected sensor data corresponding to the location of the ego actor at a particular time; projecting the motion-corrected sensor data into a two-dimensional (2D) image space to generate the first data representing a LiDAR range image of the perspective view of the environment; A method comprising:

14. 14. The method of claim 13, wherein the one or more NNs include a first stage configured to evaluate the first data representing the perspective view and a second stage configured to evaluate the transformed classification data representing the top-down view.

15. The method of claim 13 , wherein the second data further represents a class label for each of the one or more bounding shapes of the one or more objects.

16. A method for detecting an image of an environment, comprising: determining image data from the sensor data, the image data representing a first view of the environment; generating classification data representative of one or more classifications from the image data representative of an image of the first view of the environment using one or more neural networks (NN); associating the classification data with corresponding three-dimensional (3D) locations identified from corresponding sensor data to generate labeled sensor data; projecting the labeled sensor data onto a second view of the environment to generate transformed classification data representative of the one or more classifications in the second view; generating, using said one or more neural networks (NNs), second data representative of one or more boundary shapes of one or more objects detected in said environment based at least on said transformed classification data; Including, The sensor data includes accumulated sensor data from one or more LiDAR sensors of an ego actor accumulated over a period of time, and the step of determining the image data representing the first view of the environment includes: converting the accumulated sensor data into motion-corrected sensor data corresponding to the location of an ego actor at a particular time; projecting the motion-corrected sensor data into a two-dimensional (2D) image space to generate the image data representative of a LiDAR range image of the first view of the environment; A method comprising:

17. The method of claim 16 , wherein the second data further represents the one or more class labels of the one or more objects.

18. 17. The method of claim 16, wherein the generating step of the second data representing the one or more bounding shapes comprises generating the one or more bounding shapes and associated class labels of the object based on second classification data representing one or more second classifications in the second view and third data representing an object instance geometry, the second classification data and the third data being extracted by the one or more NNs.

Citation Information

Patent Citations

  • Information processing device, information processing method, program and mobile unit

    WO2019026714A1