Method for a vehicle, a vehicle and a storage medium

Abstract

The present invention relates to a method, a vehicle and a storage medium for a vehicle. A technique for controlling the motion of a vehicle-based object identified using LiDAR using a control circuit is described. For example, individual classes of points of a point cloud are determined, and objects near the vehicle are identified based on the determined individual classes of points of the point cloud.

Description

Technical Field

[0001] This specification relates to using light detection and ranging ("LiDAR") to identify objects. Background Art

[0002] LiDAR is a technology that uses light to obtain data about physical objects in the line of sight of the light emitter. LiDAR data is typically in the form of a collection of points (also called a point cloud) used to build a representation of the surrounding environment. LiDAR can be used to detect objects in the vicinity of a vehicle. Summary of the invention

[0003] According to a first aspect of the present invention, a method for a vehicle comprises: using at least one processor to receive LiDAR point cloud data associated with a plurality of points in a point cloud; using the at least one processor to generate a first view of the point cloud and a second view of the point cloud based on the plurality of points in the point cloud, wherein the second view is different from the first view; using the at least one processor to provide the first view as an input to a first view neural network and the second view as an input to a second view neural network, wherein the second view neural network is different from the first view neural network; using the at least one processor, for each point in the point cloud, using the first view neural network to generate a first set of class scores indicating an object class and using the second view neural network to generate a second set of class scores indicating the object class, wherein the first set of class scores and the second set of class scores are generated in parallel; using the at least one processor to determine a final determined label of at least one point in the point cloud, wherein the determination is based on the first set of class scores of the at least one point and the second set of class scores of the at least one point; using the at least one processor to identify at least one object near the vehicle based at least in part on the final determined label of the at least one point; and using the at least one processor to control the movement of the vehicle based on the at least one object.

[0004] According to a second aspect of the present invention, a vehicle comprises: at least one LiDAR device capable of generating a LiDAR scan point cloud comprising a plurality of LiDAR data points; and a processing circuit coupled to the LiDAR device, wherein the processing circuit is configured to perform the above method.

[0005] According to a third aspect of the present invention, a non-transitory computer-readable storage medium includes at least one program for execution by at least one processor of a first device, wherein the at least one program includes instructions that, when executed by the at least one processor, cause the first device to perform the above method. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 An example of an autonomous vehicle with autonomous capabilities is shown.

[0007] Figure 2 An example "cloud" computing environment is shown.

[0008] Figure 3 A computer system is shown.

[0009] Figure 4 An example architecture of an autonomous vehicle is shown.

[0010] Figure 5 Shows examples of inputs and outputs that a perception module can use.

[0011] Figure 6 An example of a LiDAR system is shown.

[0012] Figure 7 The LiDAR system is shown in operation.

[0013] Figure 8 Additional details of the operation of the LiDAR system are shown.

[0014] Fig. 9 A flow chart showing an example process for classifying LiDAR points.

[0015] Fig.10 A block diagram showing an example classification network.

[0016] Fig.11 A representation of an example view network is shown.

[0017] Fig.12 A representation of an example fusion network is shown.

[0018] Fig.13 A flow chart showing an example process for operating a vehicle based on classified LiDAR points. DETAILED DESCRIPTION

[0019] In the following description, for the purpose of explanation, many specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent that the present invention can be implemented without these specific details. In other examples, well-known configurations and devices are shown in block diagram form to avoid unnecessarily obscuring the present invention.

[0020] In the accompanying drawings, for ease of description, a specific arrangement or order of schematic elements (such as those representing devices, modules, instruction blocks, and data elements) is shown. However, it should be understood by those skilled in the art that the specific order or arrangement of the schematic elements in the accompanying drawings is not intended to mean that a specific processing order or sequence, or separation of processing processes is required. In addition, the inclusion of schematic elements in the accompanying drawings is not intended to mean that such elements are required in all embodiments, nor is it intended to mean that the features represented by such elements cannot be included in some embodiments or cannot be combined with other elements in some embodiments.

[0021] In addition, in the accompanying drawings, connecting elements, such as solid or dotted lines or arrows, are used to illustrate the connection, relationship or association between two or more other schematic elements, and there is no such connecting element and is not intended to mean that there can be no connection, relationship or association. In other words, the connection, relationship or association between some elements are not shown in the accompanying drawings, so as not to make the present disclosure vague. In addition, for the convenience of illustration, a single connecting element is used to represent multiple connections, relationships or associations between elements. For example, if the connecting element represents the communication of signal, data or instruction, it will be understood by those skilled in the art that this element represents one or more signal paths (for example, bus) that may be needed to affect the communication.

[0022] Reference will now be made in detail to the embodiments, examples of which are illustrated in the accompanying drawings. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the various embodiments described. However, it will be apparent to one of ordinary skill in the art that the various embodiments described may be implemented without these specific details. In other cases, well-known methods, procedures, components, circuits, and networks are not described in detail in order not to unnecessarily obscure aspects of the embodiments.

[0023] Several features described below can each be used independently of one another or in combination with other features. However, any individual feature may not solve any of the problems discussed above, or may only solve one of the problems discussed above. Some of the problems discussed above may not be fully solved by any one of the features described herein. Although a title is provided, information related to a specific title but not found in the section with that title may also be found elsewhere in this specification. This article describes an embodiment according to the following summary:

[0024] 1. General Overview

[0025] 2. System Overview

[0026] 3. Autonomous Vehicle Architecture

[0027] 4. Autonomous Vehicle Input

[0028] 5. Autonomous vehicle planning

[0029] 6. Autonomous Vehicle Control

[0030] 7. Use multi-view fusion to identify objects

[0031] General Overview

[0032] Multiple views of the LiDAR scan are processed in parallel, and the results are combined to generate semantic labels for point clusters included in the LiDAR point cloud. For example, the output from the bird's eye view (BeV) neural network and the output from the range view (RV) neural network are compared and combined (i.e., "fused") to generate a finalized set of labels. The labeled point cloud is then used to identify objects in the vicinity of the vehicle, for example, for navigation purposes.

[0033] Some advantages of the new approach include: using parallel training and deploying separate view networks, also called view neural networks (e.g., the BeV network and RV network described above), to reduce the amount of computational resources required to identify objects near a vehicle. Additionally or alternatively, the approach enables the network to function even when a subset of the separate view networks do not provide outputs as expected. To eliminate boundary errors and improve accuracy, recursive layers for spatial smoothing are added to the individual view networks to learn spatial relationships between objects. Data augmentation from a single LiDAR scan increases the number of available training samples to better train the entire network. Fusing the outputs of the view networks generates more accurate classification results for points than when the views are considered independently.

[0034] System Overview

[0035] Figure 1 An example of an autonomous vehicle 100 having autonomous capabilities is shown.

[0036] As used herein, the term "autonomous capability" refers to a function, feature, or facility that enables a vehicle to operate partially or fully without real-time human intervention, including but not limited to fully autonomous vehicles, highly autonomous vehicles, and conditionally autonomous vehicles.

[0037] As used herein, an autonomous vehicle (AV) is a vehicle with autonomous capabilities.

[0038] As used herein, "vehicle" includes a mode of transport of goods or people. For example, a car, bus, train, airplane, drone, truck, boat, ship, submersible, spacecraft, etc. An unmanned car is an example of a vehicle.

[0039] As used herein, a "track" refers to a path or route that navigates an AV from a first spatiotemporal location to a second spatiotemporal location. In embodiments, the first spatiotemporal location is referred to as an initial location or starting location, and the second spatiotemporal location is referred to as a destination, final location, target, target location, or target location. In some examples, a track consists of one or more road segments (e.g., several sections of a road), and each road segment consists of one or more blocks (e.g., a portion of a lane or intersection). In embodiments, a spatiotemporal location corresponds to a real-world location. For example, a spatiotemporal location is a pickup or drop-off location to get people or cargo on or off the vehicle.

[0040] As used herein, "sensor(s)" includes one or more hardware components for detecting information about the environment surrounding the sensor. Some hardware components may include sensing components (e.g., image sensors, biometric sensors), transmission and / or reception components (e.g., laser or radio frequency wave transmitters and receivers), electronic components (such as analog-to-digital converters), data storage devices (such as RAM and / or non-volatile memory), software or firmware components and data processing components (such as application specific integrated circuits), microprocessors and / or microcontrollers.

[0041] As used herein, a “scene description” is a data structure (e.g., a list) or data stream that includes one or more classified or labeled objects detected by one or more sensors on an AV vehicle, or one or more classified or labeled objects provided by a source external to the AV.

[0042] As used herein, a "road" is a physical area that can be traversed by a vehicle and may correspond to a named thoroughfare (e.g., a city street, an interstate highway, etc.) or may correspond to an unnamed thoroughfare (e.g., a driveway within a house or office building, a section of a parking lot, a section of a vacant parking lot, a dirt road in a rural area, etc.). Because some vehicles (e.g., four-wheel drive pickup trucks, off-road vehicles (SUVs), etc.) are able to traverse a variety of physical areas that are not particularly suitable for vehicle travel, a "road" may be any physical area that is not formally defined as a thoroughfare by a municipality or other governmental or administrative agency.

[0043] As used herein, a "lane" is a portion of a road that can be traversed by a vehicle. Lanes are sometimes identified based on lane markings. For example, a lane may correspond to most or all of the space between lane markings, or only a portion of the space between lane markings (e.g., less than 50%). For example, a road with lane markings that are far apart may accommodate two or more vehicles, so that one vehicle can pass another vehicle without crossing the lane markings, and thus may be interpreted as a lane that is narrower than the space between lane markings, or as having two lanes between lanes. Lanes may also be interpreted in the absence of lane markings. For example, lanes may be defined based on physical features of the environment (e.g., rocks in rural areas and trees along avenues, or natural obstacles that should be avoided, for example, in underdeveloped areas). Lanes may also be interpreted independently of lane markings or physical features. For example, a lane may be interpreted based on an arbitrary path without obstacles in an area that would otherwise lack features that would be interpreted as lane boundaries. In an example scenario, an AV may interpret a lane as an obstacle-free portion through a field or open space. In another example scenario, an AV may interpret a lane through a wide (e.g., wide enough for two or more lanes) road that does not have lane markings. In this scenario, the AV may communicate information about the lane to other AVs so that the other AVs may use the same lane information to coordinate path planning between AVs.

[0044] The term “Over-the-Air (OTA) Client” includes any AV, or any electronic device (e.g., computer, controller, IoT device, Electronic Control Unit (ECU)) embedded in, coupled to, or communicating with an AV.

[0045] The term "Over-the-Air (OTA) Update" means any update, change, deletion or addition to software, firmware, data or configuration settings, or any combination thereof, delivered to an OTA client using proprietary and / or standardized wireless communication technologies including, but not limited to: cellular mobile communications (e.g., 2G, 3G, 4G, 5G), radio wireless area networks (e.g., WiFi) and / or satellite Internet.

[0046] The term “edge node” refers to one or more edge devices coupled to a network that provide a portal for communicating with AVs and can communicate with other edge nodes and cloud-based computing platforms to schedule and deliver OTA updates to OTA clients.

[0047] The term "edge device" refers to a device that implements an edge node and provides a physical wireless access point (AP) to the core network of an enterprise or service provider (such as VERIZON, AT&T). Examples of edge devices include, but are not limited to: computers, controllers, transmitters, routers, routing switches, integrated access devices (IADs), multiplexers, metropolitan area networks (MANs), and wide area networks (WAN) access devices.

[0048] “One or more” includes a function performed by one element, a function performed by multiple elements, such as in a distributed manner, several functions performed by one element, several functions performed by several elements, or any combination of the above.

[0049] It will also be understood that although in some cases, the terms "first," "second," etc. are used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the various described embodiments, a first contact may be referred to as a second contact, and similarly, a second contact may be referred to as a first contact. Both the first contact and the second contact are contacts, but they are not the same contact.

[0050] The terms used in the specification of the various embodiments described herein are only used for the purpose of describing specific embodiments and are not intended to be limiting. As used in the specification of the various embodiments described and the appended claims, the singular forms "a", "an" and "the" are also intended to include plural forms, unless the context clearly indicates otherwise. It will also be understood that "and / or" as used herein refers to and includes any and all possible combinations of one or more related list items. It will also be understood that when the terms "include", "comprises", "have" and / or "have" are used in this specification, the stated features, integers, steps, operations, elements and / or components are specifically stated, but the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof are not excluded.

[0051] As used herein, the term "if" is alternatively understood to mean "when" or "at the time" or "in response to determining that" or "in response to detecting," depending on the context. Similarly, the phrases "if it has been determined" or "if [the stated condition or event] has been detected" are alternatively understood to mean "upon determination" or "in response to determining that" or "upon detecting [the stated condition or event]" or "in response to detecting [the stated condition or event]," depending on the context.

[0052] As used herein, an AV system refers to an AV and the array of hardware, software, stored data, and real-time generated data that supports the operation of the AV. In an embodiment, the AV system is incorporated into the AV. In an embodiment, the AV system is distributed across several locations. For example, some software of the AV system is in a system similar to the following about Figure 2 The described cloud computing environment 200 is implemented on the cloud computing environment.

[0053] In general, this document describes technologies applicable to any vehicle having one or more autonomous capabilities, including fully autonomous vehicles, highly autonomous vehicles, and conditionally autonomous vehicles, such as so-called Level 5, Level 4, and Level 3 vehicles, respectively (see SAE International Standard J3016: Classification and Definitions of Terms Relating to Automated Driving Systems for On-Road Motor Vehicles, which is incorporated by reference in its entirety into this document for more details on vehicle autonomy levels). The technologies described in this document are also applicable to partially autonomous vehicles and driver-assisted vehicles, such as so-called Level 2 and Level 1 vehicles (see SAE International Standard J3016: Classification and Definitions of Terms Relating to Automated Driving Systems for On-Road Motor Vehicles). In embodiments, one or more Level 1, Level 2, Level 3, Level 4, and Level 5 vehicle systems may automatically perform certain vehicle operations (e.g., steering, braking, and using maps) under certain operating conditions based on processing of sensor inputs. The technologies described in this document may benefit any level of vehicle ranging from fully autonomous vehicles to human-operated vehicles.

[0054] Autonomous vehicles have advantages over vehicles that require human drivers. One advantage is safety. For example, in 2016, the United States experienced 6 million automobile accidents, 2.4 million injuries, 40,000 deaths, and 13 million vehicle crashes, with an estimated social cost of more than $910 billion. From 1965 to 2015, the number of U.S. traffic fatalities has decreased from about 6 to about 1 per 100 million miles traveled, in part due to additional safety measures deployed in vehicles. For example, it is believed that an extra half-second of warning related to an impending collision mitigates 60% of front-to-rear collisions. However, passive safety features (e.g., seat belts, airbags) may have reached their limits in improving this number. Therefore, active safety measures such as automatic control of vehicles are a possible next step to improve these statistics. Because human drivers are believed to be the cause of severe pre-crash events in 95% of crashes, automated driving systems have the potential to achieve better safety outcomes by, for example: reliably identifying and avoiding emergency situations better than humans; making better decisions than humans, obeying traffic laws better than humans, and predicting future events better than humans; and reliably controlling a vehicle better than humans.

[0055] refer to Figure 1 , the AV system 120 causes the vehicle 100 to operate along a trajectory 198, through an environment 190 to a destination 199 (sometimes referred to as a final location), while avoiding objects (e.g., natural obstacles 191, vehicles 193, pedestrians 192, cyclists, and other obstacles) and complying with road rules (e.g., operating rules or driving preferences).

[0056] In an embodiment, the AV system 120 includes a device 101 for receiving and operating commands from a computer processor 146. The term "operating command" is used to refer to an executable instruction (or set of instructions) that causes a vehicle to perform an action (e.g., a driving maneuver). The operating command may include, but is not limited to, instructions for causing the vehicle to start moving forward, stop moving forward, start moving backward, stop moving backward, speed up, slow down, make a left turn, and make a right turn. In an embodiment, the computer processor 146 and the following reference Figure 3 The processor 304 is similarly described. Examples of devices 101 include steering controls 102, brakes 103, gears, an accelerator pedal or other acceleration control mechanism, windshield wipers, side door locks, window controls, and turn indicators.

[0057] In an embodiment, the AV system 120 includes sensors 121 for measuring or inferring attributes of the state or condition of the vehicle 100, such as the position, linear and angular velocity and acceleration of the AV, and heading (e.g., the direction of the front end of the vehicle 100). Examples of sensors 121 are GPS, an inertial measurement unit (IMU) that measures both linear acceleration and angular rate of the vehicle, wheel rate sensors for measuring or estimating wheel slip, wheel brake pressure or brake torque sensors, engine torque or wheel torque sensors, and steering angle and angular rate sensors.

[0058] In an embodiment, the sensors 121 also include sensors for sensing or measuring properties of the environment of the AV, such as monocular or stereo cameras 122 in the visible, infrared, or thermal (or both) spectrum, LiDAR 123, RADAR, ultrasonic sensors, time-of-flight (TOF) depth sensors, velocity sensors, temperature sensors, humidity sensors, and precipitation sensors.

[0059] In an embodiment, the AV system 120 includes a data storage unit 142 and a memory 144 for storing machine instructions associated with a computer processor 146 or data collected by the sensor 121. In an embodiment, the data storage unit 142 is associated with the following Figure 3190. In an embodiment, the memory 144 is similar to the main memory 306 described below. In an embodiment, the data storage unit 142 and the memory 144 store historical, real-time, and / or predictive information about the environment 190. In an embodiment, the stored information includes maps, driving performance, traffic congestion updates, or weather conditions. In an embodiment, data related to the environment 190 is transmitted from the remote database 134 to the vehicle 100 via a communication channel.

[0060] In an embodiment, the AV system 120 includes a communication device 140 for transmitting measured or inferred properties of the state and condition of other vehicles (such as position, linear and angular velocity, linear and angular acceleration, and linear and angular heading) to the vehicle 100. These devices include vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication devices and devices for wireless communication through point-to-point or ad hoc networks or both. In an embodiment, the communication device 140 communicates across the electromagnetic spectrum (including radio and optical communications) or other media (e.g., air and acoustic media). The combination of vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications (and in some embodiments, one or more other types of communications) is sometimes referred to as vehicle-to-everything (V2X) communications. V2X communications typically comply with one or more communication standards for communication with and between autonomous vehicles.

[0061] In an embodiment, the communication device 140 includes a communication interface. For example, a wired, wireless, WiMAX, Wi-Fi, Bluetooth, satellite, cellular, optical, near field, infrared, or radio interface. The communication interface transmits data from the remote database 134 to the AV system 120. In an embodiment, the remote database 134 is embedded in a Figure 2 The communication device 140 transmits data collected from the sensor 121 or other data related to the operation of the vehicle 100 to the remote database 134. In an embodiment, the communication device 140 transmits information related to teleoperation to the vehicle 100. In some embodiments, the vehicle 100 communicates with other remote (e.g., "cloud") servers 136.

[0062] In an embodiment, the remote database 134 also stores and transmits digital data (e.g., data such as roads and street locations). Such data is stored in a memory 144 on the vehicle 100 or transmitted from the remote database 134 to the vehicle 100 via a communication channel.

[0063] In an embodiment, the remote database 134 stores and transmits historical information (e.g., speed and acceleration profiles) related to driving attributes of the vehicle that has previously traveled along the trajectory 198 at similar times of day. In one implementation, such data may be stored in a memory 144 on the vehicle 100 or transmitted from the remote database 134 to the vehicle 100 via a communication channel.

[0064] A computer processor 146 located onboard the vehicle 100 algorithmically generates control actions based on both real-time sensor data and a priori information, allowing the AV system 120 to perform its autonomous driving capabilities.

[0065] In an embodiment, the AV system 120 includes a computer peripheral device 132 coupled to the computer processor 146 for providing information and reminders to a user of the vehicle 100 (e.g., an occupant or a remote user) and receiving input from the user. In an embodiment, the peripheral device 132 is similar to the one described below with reference to Figure 3 Display 312, input device 314 and cursor control 316 are discussed. The coupling may be wireless or wired. Any two or more of the interface devices may be integrated into a single device.

[0066] In an embodiment, the AV system 120 receives and enforces the privacy level of the occupant, for example, specified by the occupant or stored in a profile associated with the occupant. The privacy level of the occupant determines how to permit the use of specific information associated with the occupant (e.g., occupant comfort data, biometric data, etc.) stored in the occupant profile and / or stored on the cloud server 136 and associated with the occupant profile. In an embodiment, the privacy level specifies specific information associated with the occupant that is deleted once the ride is completed. In an embodiment, the privacy level specifies specific information associated with the occupant and identifies one or more entities authorized to access the information. Examples of designated entities authorized to access information may include other AVs, third-party AV systems, or any entity that can potentially access the information.

[0067] The privacy level of an occupant may be specified at one or more levels of granularity. In an embodiment, the privacy level identifies specific information to be stored or shared. In an embodiment, the privacy level applies to all information associated with the occupant, such that the occupant may specify that her personal information is not to be stored or shared. The designation of entities permitted to access specific information may also be specified at various levels of granularity. The various sets of entities permitted to access specific information may include, for example, other AVs, cloud servers 136, specific third-party AV systems, etc.

[0068] In an embodiment, the AV system 120 or the cloud server 136 determines whether the AV 100 or another entity may access certain information associated with the occupant. For example, a third-party AV system attempting to access occupant input related to a particular spatiotemporal location must obtain authorization, such as from the AV system 120 or the cloud server 136, to access information associated with the occupant. For example, the AV system 120 uses the occupant's specified privacy level to determine whether the occupant input related to the spatiotemporal location can be presented to a third-party AV system, the AV 100, or another AV. This enables the occupant's privacy level to specify which other entities are allowed to receive data related to the occupant's actions or other data associated with the occupant.

[0069] Figure 2 An example "cloud" computing environment is shown. Cloud computing is a service delivery model for enabling convenient, on-demand access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage, applications, virtual machines, and services) over a network. In a typical cloud computing system, one or more large cloud data centers house the machines used to deliver the services provided by the cloud. Now refer to Figure 2 , cloud computing environment 200 includes cloud data centers 204a, 204b, and 204c interconnected by cloud 202. Data centers 204a, 204b, and 204c provide cloud computing services for computer systems 206a, 206b, 206c, 206d, 206e, and 206f connected to cloud 202.

[0070] The cloud computing environment 200 includes one or more cloud data centers. Figure 2 The cloud data center 204a shown in FIG. 2 refers to a cloud (eg, Figure 2 The physical arrangement of servers in a cloud 202 or a specific portion of a cloud (as shown in FIG. 202 or a specific portion of a cloud). For example, servers are physically arranged into rooms, groups, rows, and racks in a cloud data center. A cloud data center has one or more zones, which include one or more server rooms. Each room has one or more rows of servers, and each row includes one or more racks. Each rack includes one or more individual server nodes. In some implementations, servers in zones, rooms, racks, and / or rows are arranged into groups based on the physical infrastructure requirements of the data center facility, including power, energy, heat, heat sources, and / or other requirements. In an embodiment, server nodes are similar to Figure 3 The data center 204a has many computing systems distributed across multiple racks.

[0071] Cloud 202 includes cloud data centers 204a, 204b, and 204c and networks and network resources (e.g., network devices, nodes, routers, switches, and network cables) used to connect cloud data centers 204a, 204b, and 204c and help facilitate access to cloud computing services by computing systems 206a-f. In an embodiment, the network represents any combination of one or more local networks, wide area networks, or internetworks coupled by wired or wireless links deployed using terrestrial or satellite connections. Data exchanged through the network is transmitted using a variety of network layer protocols (such as Internet Protocol (IP), Multi-Protocol Label Switching (MPLS), Asynchronous Transfer Mode (ATM), Frame Relay, etc.). In addition, in embodiments where the network represents a combination of multiple subnetworks, different network layer protocols are used on each underlying subnetwork. In some embodiments, the network represents one or more interconnected internetworks (such as the public Internet, etc.).

[0072] Computing systems 206a-f or cloud computing service consumers are connected to the cloud 202 through network links and network adapters. In an embodiment, computing systems 206a-f are implemented as various computing devices, such as servers, desktops, laptops, tablets, smart phones, Internet of Things (IoT) devices, autonomous vehicles (including cars, drones, shuttles, trains, buses, etc.) and consumer electronics. In an embodiment, computing systems 206a-f are implemented in other systems or as part of other systems.

[0073] Figure 3 Computer system 300 is shown. In implementation, computer system 300 is a special-purpose computing device. The special-purpose computing device is hard-wired to perform these techniques, or includes a digital electronic device such as one or more application-specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) that are permanently programmed to perform the above-mentioned techniques, or may include one or more general-purpose hardware processors that are programmed to perform these techniques according to program instructions in firmware, memory, other memory, or a combination. Such a special-purpose computing device can also combine customized hard-wired logic, ASICs or FPGAs with customized programming to complete these techniques. In various embodiments, the special-purpose computing device is a desktop computer system, a portable computer system, a handheld device, a network device, or any other device that includes hard-wired and / or program logic to implement these techniques.

[0074] In an embodiment, the computer system 300 includes a bus 302 or other communication mechanism for communicating information, and a processor 304 coupled to the bus 302 to process information. The processor 304 is, for example, a general-purpose microprocessor. The computer system 300 also includes a main memory 306, such as a random access memory (RAM) or other dynamic storage device, coupled to the bus 302 to store information and instructions, which are executed by the processor 304. In one implementation, the main memory 306 is used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processor 304. When these instructions are stored in a non-transitory storage medium accessible to the processor 304, the computer system 300 becomes a special-purpose machine that is customized to perform the operations specified in the instructions.

[0075] In an embodiment, computer system 300 also includes a read only memory (ROM) 308 or other static storage device coupled to bus 302 for storing static information and instructions for processor 304. A storage device 310, such as a magnetic disk, optical disk, solid state drive, or three-dimensional cross point memory, is provided and coupled to bus 302 to store information and instructions.

[0076] In an embodiment, the computer system 300 is coupled to a display 312 such as a cathode ray tube (CRT), a liquid crystal display (LCD), a plasma display, a light emitting diode (LED) display, or an organic light emitting diode (OLED) display for displaying information to a computer user via bus 302. An input device 314 including alphanumeric and other keys is coupled to bus 302 for communicating information and command selections to processor 304. Another type of user input device is a cursor controller 316, such as a mouse, a trackball, a touch display, or cursor direction keys, for communicating direction information and command selections to processor 304 and for controlling movement of a cursor on display 312. Such input devices typically have two degrees of freedom in two axes, a first axis (e.g., an x-axis) and a second axis (e.g., a y-axis) that allow the device to specify a position on a plane.

[0077] According to one embodiment, the techniques herein are performed by computer system 300 in response to processor 304 executing one or more sequences of one or more instructions contained in main memory 306. These instructions are read into main memory 306 from another storage medium, such as storage device 310. Execution of the sequences of instructions contained in main memory 306 causes processor 304 to perform the process steps described herein. In alternative embodiments, hard-wired circuitry is used in place of or in combination with software instructions.

[0078] As used herein, the term "storage medium" refers to any non-temporary medium that stores data and / or instructions that cause a machine to operate in a particular manner. Such storage media include non-volatile media and / or volatile media. Non-volatile media include, for example, optical disks, magnetic disks, solid-state drives, or three-dimensional cross-point memories such as storage device 310. Volatile media include dynamic memories such as main memory 306. Common forms of storage media include, for example, floppy disks, floppy disks, hard disks, solid-state drives, tapes or any other magnetic data storage media, CD-ROMs, any other optical data storage media, any physical media with a hole pattern, RAM, PROM and EPROM, FLASH-EPROM, NV-RAM, or any other memory chip or storage box.

[0079] Storage media are distinct from transmission media, but can be used in conjunction with transmission media. Transmission media participate in the transmission of information between storage media. For example, transmission media include coaxial cables, copper wires, and optical fibers, including wires that provide bus 302. Transmission media can also take the form of sound waves or light waves, such as those generated during radio wave and infrared data communications.

[0080] In an embodiment, various forms of media are involved in carrying one or more sequences of one or more instructions to the processor 304 for execution. For example, the instructions are initially executed on a disk or solid-state drive of a remote computer. The remote computer loads the instructions into its dynamic memory and sends the instructions over a telephone line using a modem. The local modem of the computer system 300 receives the data on the telephone line and uses an infrared transmitter to convert the data to an infrared signal. An infrared detector receives the data carried in the infrared signal, and appropriate circuitry places the data on the bus 302. The bus 302 carries the data to the main memory 306, from which the processor 304 retrieves and executes the instructions. The instructions received by the main memory 306 may optionally be stored on the storage device 310 before or after execution by the processor 304.

[0081] Computer system 300 also includes a communication interface 318 coupled to bus 302. Communication interface 318 provides a two-way data communication coupled to a network link 320 connected to a local network 322. For example, communication interface 318 is an integrated services digital network (ISDN) card, a cable modem, a satellite modem, or a modem for providing a data communication connection with a corresponding type of telephone line. As another example, communication interface 318 is a local area network (LAN) card for providing a data communication connection with a compatible LAN. In some implementations, a wireless link is also implemented. In any such implementation, communication interface 318 sends and receives an electrical, electromagnetic or optical signal carrying a digital data stream representing various types of information.

[0082] The network link 320 typically provides data communication to other data devices through one or more networks. For example, the network link 320 provides a connection to a host computer 324 or to a cloud data center or device operated by an Internet Service Provider (ISP) 326 through a local network 322. The ISP 326, in turn, provides data communication services through a worldwide packet data communication network now commonly referred to as the "Internet" 328. Both the local network 322 and the Internet 328 use electrical, electromagnetic, or optical signals that carry digital data streams. The signals through the various networks and the signals on the network link 320 and through the communication interface 318 are example forms of transmission media, where these signals carry digital data to and from the computer system 300. In an embodiment, the network 320 includes the cloud 202 described above or a portion of the cloud 202.

[0083] Computer system 300 sends messages and receives data, including program code, through network(s), network link 320, and communication interface 318. In an embodiment, computer system 300 receives code for processing. The received code is executed by processor 304 as it is received, and / or stored in storage device 310, or other non-volatile storage for later execution.

[0084] Autonomous Vehicle Architecture

[0085] Figure 4 An example of a method for autonomous vehicles (e.g., Figure 1 100). The architecture 400 includes a perception module 402 (sometimes referred to as a perception circuit), a planning module 404 (sometimes referred to as a planning circuit), a control module 406 (sometimes referred to as a control circuit), a positioning module 408 (sometimes referred to as a positioning circuit), and a database module 410 (sometimes referred to as a database circuit). Each module plays a role in the operation of the vehicle 100. Collectively, the modules 402, 404, 406, 408, and 410 may be Figure 1 404, 406, 408, and 410 are each sometimes referred to as a processing circuit (e.g., computer hardware, computer software, or a combination of the two). Any or all of the combinations of modules 402, 404, 406, 408, and 410 are also examples of processing circuits.

[0086] In use, planning module 404 receives data representing destination 412 and determines data representing a trajectory 414 (sometimes referred to as a route) that vehicle 100 may travel in order to reach (e.g., arrive at) destination 412. In order for planning module 404 to determine data representing trajectory 414, planning module 404 receives data from perception module 402, positioning module 408, and database module 410.

[0087] The perception module 402 uses, for example, Figure 1 One or more sensors 121 are shown to identify nearby physical objects. Objects are classified (e.g., grouped into types such as pedestrians, bicycles, cars, traffic signs, etc.), and a scene description including the classified objects 416 is provided to the planning module 404.

[0088] The planning module 404 also receives data representing the AV position 418 from the positioning module 408. The positioning module 408 determines the AV position by using data from the sensor 121 and data from the database module 410 (e.g., geographic data) to calculate the position. For example, the positioning module 408 uses data from a GNSS (Global Navigation Satellite System) sensor and geographic data to calculate the longitude and latitude of the AV. In an embodiment, the data used by the positioning module 408 includes a high-precision map with lane geometry attributes, a map describing the road network connection attributes, a map describing the physical attributes of the lane (such as traffic speed, traffic volume, the number of vehicles and bicycle lanes, lane width, lane traffic direction, or lane marking type and location, or a combination thereof), and a map describing the spatial location of road features (such as intersections, traffic signs, or various types of other driving signals, etc.). In an embodiment, a high-precision map is constructed by adding data to a low-precision map via automatic or manual annotation.

[0089] The control module 406 receives data representing the trajectory 414 and data representing the AV position 418, and operates the control functions 420a-420c of the AV (e.g., steering, throttle, brakes, ignition) in a manner that will cause the vehicle 100 to travel the trajectory 414 to reach the destination 412. For example, if the trajectory 414 includes a left turn, the control module 406 will operate the control functions 420a-420c in the following manner: the steering angle of the steering function will cause the vehicle 100 to turn left, and the throttle and brakes will cause the vehicle 100 to pause and wait for a passing pedestrian or vehicle before making the turn.

[0090] Autonomous Vehicle Input

[0091] Figure 5 The perception module 402 ( Figure 4) used by the inputs 502a-502d (e.g., Figure 1 1) and outputs 504a-504d (e.g., sensor data). One input 502a is a LiDAR (Light Detection and Ranging) system (e.g., Figure 1 123). LiDAR is a technology that uses light (e.g., a beam of light such as infrared light) to obtain data about physical objects in its line of sight. The LiDAR system produces LiDAR data as output 504a. For example, LiDAR data is a collection of 3D or 2D points (also called a point cloud) used to construct a representation of the environment 190.

[0092] Another input 502b is a RADAR (radar) system. RADAR is a technology that uses radio waves to obtain data related to nearby physical objects. RADAR can obtain data related to objects that are not within the line of sight of the LiDAR system. The RADAR system produces RADAR data as output 504b. For example, RADAR data is one or more radio frequency electromagnetic signals used to construct a representation of the environment 190.

[0093] Another input 502c is a camera system. The camera system uses one or more cameras (e.g., a digital camera using a light sensor such as a charge coupled device [CCD]) to obtain information about nearby physical objects. The camera system generates camera data as output 504c. The camera data is typically in the form of image data (e.g., data in an image data format such as RAW, JPEG, PNG, etc.). In some examples, the camera system has multiple independent cameras, such as for the purpose of stereoscopic imaging (stereoscopic vision), which enables the camera system to perceive depth. Although the objects perceived by the camera system are described here as "nearby", this is relative to the AV. In some embodiments, the camera system is configured to "see" distant objects (e.g., as far as 1 km or more in front of the AV). Therefore, in some embodiments, the camera system has features such as sensors and lenses that are optimized for perceiving distant objects.

[0094] Another input 502d is a traffic light detection (TLD) system. The TLD system uses one or more cameras to obtain information about traffic lights, street signs, and other physical objects that provide visual navigation information. The TLD system generates TLD data as an output 504d. TLD data often takes the form of image data (e.g., data in image data formats such as RAW, JPEG, PNG, etc.). The difference between a TLD system and a system including a camera is that the TLD system uses a camera with a wide field of view (e.g., using a wide-angle lens or a fisheye lens) to obtain information about as many physical objects that provide visual navigation information as possible, so that the vehicle 100 can access all relevant navigation information provided by these objects. For example, the viewing angle of the TLD system is about 120 degrees or greater.

[0095] In some embodiments, sensor fusion techniques are used to combine the outputs 504a-504d. Thus, the individual outputs 504a-504d are provided to other systems of the vehicle 100 (e.g., to a Figure 4 The combined output may be provided to other systems in the form of a single combined output or multiple combined outputs of the same type (e.g., using the same combining technique or combining the same outputs, or both) or a single combined output or multiple combined outputs of different types (e.g., using different individual combining techniques or combining different individual outputs, or both). In some embodiments, an early fusion technique is used. An early fusion technique is characterized in that the outputs are combined before one or more data processing steps are applied to the combined output. In some embodiments, a late fusion technique is used. A late fusion technique is characterized in that the outputs are combined after one or more data processing steps are applied to the individual outputs.

[0096] Figure 6 An example of a LiDAR system 602 is shown (eg, Figure 5502a shown). The LiDAR system 602 emits light 604a-604c from a light emitter 606 (e.g., a laser emitter). The light emitted by the LiDAR system is typically not in the visible spectrum; for example, infrared light is often used. Some of the emitted light 604b encounters a physical object 608 (e.g., a vehicle) and reflects back to the LiDAR system 602. (The light emitted from the LiDAR system typically does not penetrate a physical object, such as a physical object that is solid in form.) The LiDAR system 602 also has one or more light detectors 610 for detecting the reflected light. In an embodiment, one or more data processing systems associated with the LiDAR system generate an image 612 representing a field of view 614 of the LiDAR system. The image 612 includes information representing a boundary 616 of the physical object 608. In this way, the image 612 is used to determine the boundaries 616 of one or more physical objects near the AV.

[0097] Figure 7 6. The LiDAR system 602 is shown in operation. In the scenario shown in the figure, the vehicle 100 receives both the camera system output 504c in the form of an image 702 and the LiDAR system output 504a in the form of LiDAR data points 704. In use, the data processing system of the vehicle 100 compares the image 702 to the data points 704. In particular, the physical objects 706 identified in the image 702 are also identified in the data points 704. In this way, the vehicle 100 perceives the boundaries of the physical objects based on the contours and density of the data points 704.

[0098] Figure 8 6 shows additional details of the operation of the LiDAR system 602. As described above, the vehicle 100 detects the boundaries of the physical object based on the characteristics of the data points detected by the LiDAR system 602. Figure 8 As shown, a flat object such as the ground 802 will reflect light 804a-804d emitted from the LiDAR system 602 in a consistent manner. In other words, because the LiDAR system 602 emits light using consistent intervals, the ground 802 will reflect light back to the LiDAR system 602 at the same consistent intervals. As the vehicle 100 travels over the ground 802, the LiDAR system 602 will continue to detect light reflected by the next valid ground point 806 if nothing is blocking the road. However, if an object 808 is blocking the road, the light 804e-804f emitted by the LiDAR system 602 will be reflected from points 810a-810b in a manner that is inconsistent with the expected consistent manner. Based on this information, the vehicle 100 can determine that the object 808 is present.

[0099] Recognize objects using multi-view fusion

[0100] Fig. 9 A flow chart of a process 900 for classifying LiDAR points is shown. Figure 6 As described, the vehicle 100 detects physical objects based on the characteristics of data points 704 in the form of a point cloud detected by the LiDAR system 602. In some embodiments, the data points are processed by one or more neural networks to identify the objects represented by these data points. For example, the point cloud is processed by a neural network to generate semantic labels for point clusters included in the point cloud. The semantic labels are used to distinguish objects in the point cloud. In addition, in some examples, multiple views of the point cloud are processed and fused in parallel to generate a final set of labels.

[0101] During process 900, from LiDAR device 502a ( Figure 5 ) generates point cloud data. In some examples, from LiDAR 502a and camera device 502c ( Figure 5 ) generates point cloud data. In an embodiment, the received point cloud data includes three-dimensional position information of each point in the point cloud. In an embodiment, the point cloud data includes point intensity data. The point intensity data represents the light intensity of the point in the point cloud. In an embodiment, the point intensity data is a real-valued number. In an embodiment utilizing a combination of LiDAR 502a and camera device 502c, the point cloud data includes color data of at least one point in the point cloud. The color data represents color information of the point in the point cloud. In an embodiment, the color data is represented as a tensor including RGB data. Collect (902) available sensor data including point cloud data or camera data.

[0102] Then, the point cloud data is projected onto two two-dimensional surfaces to generate (904) view data. The view data is the projection of the three-dimensional point cloud data onto the two two-dimensional surfaces. For example, the point cloud data is projected onto a bird's eye view (BeV) and a range view (RV). Fig.10 and Fig.11 to describe the view data in detail.

[0103] The view data is provided to a view network. The view network is an encoder-decoder neural network (e.g., with or without point-level encodings) for generating a classification metric based on the input view data. In embodiments where the input view data is an image, the view network outputs a segmentation map of the image. The segmentation map is a matrix of labels such that each element of the segmentation map is a label for a corresponding pixel in the input view image data. Each view network takes the view data as input and computes (906) a set of class scores for each point in the point cloud data. A set of class scores is an n-dimensional vector, where n is the number of predefined classes and where each element of the vector represents a likelihood score for the class to which the point belongs. A class reflects an object type (e.g., vegetation, vehicle, or pedestrian). The following is based on Fig.11 Get more details about the view network.

[0104] The class scores of different groups of points in the point cloud data are obtained and compared. Based on the results of the comparison, the point is then determined (910) to be an uncertain point 950 or a classified point 960. Fig.10 Find out the details about the comparison process.

[0105] The finalized (914) label is a label assigned to a point in the point cloud to indicate that the point is part of the object of the point (e.g., vegetation, a vehicle, or a pedestrian). In most cases, the classified point 960 shows the same dominant class from different sets of class scores, so the finalized label of the classified point 960 is the dominant class from one set of class scores. However, the uncertain point 950 usually does not show the same dominant class from different sets of class scores, and additional processing is required to determine the finalized label of the uncertain point.

[0106] The fusion network (also called a fusion neural network) takes the uncertain point and generates a new set of class scores for the uncertain point by fusing (912). The fusion network is a neural network that fuses the final or intermediate outputs of the view networks for points in the point cloud to generate a more accurate result. In an embodiment, the new set of class scores shows the dominant class used as the label for the final determination (914) of the uncertain point 950. The following is based on Fig.12 Find out the details about the comparison process.

[0107] In an embodiment, the finalized labels of the points in the point cloud are then passed to Figure 4 The sensing circuit 402 shown is used for tasks such as object recognition. Fig.10 Find out the details related to the subsequent tasks.

[0108] Fig.10A block diagram of a classification network 1000 for classifying LiDAR scan points is shown. The classification network 1000 classifies LiDAR scan points from a LiDAR device 1001 (e.g., Figure 5 The system uses point cloud data 1003 obtained by a LiDAR 502a as shown as input, and generates finalized labels 1060 for each point in the point cloud data 1003. The finalized labels 1060 are labels assigned to points in the point cloud that are part of an object (e.g., vegetation, a vehicle, or a pedestrian).

[0109] Based on the order of the subsequent tasks to be performed, the positioning circuit 408 ( Figure 4 ) or sensing circuit 402 ( Figure 4 ) to receive and process the point cloud data. The processing circuit 1002 is used to project the three-dimensional point cloud data 1003 onto two two-dimensional surfaces. Similar to the projection of the earth on a world map, one projection is as if the point cloud is unfolded from a spherical surface onto a plane, which is called range view (RV) data 1020. In other words, the range view (RV) data 1020 is the three-dimensional point cloud data in a spherical coordinate system mapped onto an xy coordinate system, where each x coordinate represents each Angle. Another projection is as if the point cloud is viewed from above, which is called a bird's eye view (BeV) data 1030. In other words, the bird's eye view (BeV) data 1030 is three-dimensional point cloud data in an xyz coordinate system mapped to an xy coordinate system with the z axis removed. The RV data 1020 and BeV data 1030 are in multiple formats. In an example, the RV data 1020 and BeV data 1030 are in the format of an image. In another example, the RV data 1020 and BeV data 1030 are in the format of a matrix.

[0110] RV data 1020 is provided as input to the range view network 1004 or RV network, and BeV data 1030 is provided as input to the bird's eye view network 1006 or BeV network. In an embodiment, both the RV network 1004 and the BeV network 1006 are view networks. Fig.11 Detailed architecture of a view network representing RV network 1004 and BeV network 1006 is shown in RV network 1004. RV network 1004 computes a first set of class scores 1040 for each point in the point cloud. Similarly, BeV network 1006 computes a second set of class scores 1050 for each point in the point cloud.

[0111] RV network 1004 and BeV network 1006 are independent neural networks that do not depend on each other for input or computation. In an embodiment, two different processors are used to evaluate RV network 1004 and BeV network 1006 in parallel. In an embodiment, one of the two different processors used is Figure 3 The processor 304 is shown, while another processor used is Figure 2 202. In another embodiment, both processors are located in the cloud 202, or as elements of the AV. In an embodiment with limited computing resources, the RV network 1004 and the BeV network 1006 are evaluated sequentially, so that both are evaluated using computing resources that can evaluate only one of the RV network 1004 and the BeV network 1006 at the same time.

[0112] The first set of class scores 1040 of the points and the second set of class scores 1050 of the points are provided as input to the score comparator 1008. If the two sets of scores 1040 and 1050 differ by a threshold value, the point is considered to be an uncertain point 950. Otherwise, the point is considered to be a classified point 960. The threshold difference between the two sets of class scores 1040 and 1050 depends on the selected evaluation metric. As described above, the first set of class scores 1040 and the second set of class scores 1050 are each represented by an n-dimensional vector. In an embodiment, the difference between the two sets of class scores 1040 and 1050 is calculated based on the cosine distance between the two vectors. In an embodiment, the difference between the two sets of class scores 1040 and 1050 is calculated based on the Euclidean distance or the L2 norm distance between the two vectors. In an embodiment, the difference between the two sets of class scores 1040 and 1050 is calculated based on the Manhattan distance or the L1 norm distance between the two vectors.

[0113] The fusion network 1010 takes the uncertain point 950 as input and generates a new set of class scores 1220 for the uncertain point 950 based on the first set of class scores 1040 and the second set of class scores 1050, considering that the difference between the two sets of class scores 1040 and 1050 satisfies the threshold difference. The predicted class of the uncertain point is determined based on the new set of class scores 1220. In an embodiment, the predicted class of the uncertain point 950 is the dominant class corresponding to the maximum likelihood score in the new set of class scores 1220. The predicted class is regarded as the final determined label 1060 of the uncertain point 950. According to Fig.12 Details regarding converged networks and example implementations of converged networks are described below.

[0114] In an embodiment, the fusion network 1010 is extended to include a finalization module. For a classified point, the finalization module uses either the first set of class scores 1040 or the second set of class scores 1050 and determines a dominant class. In an embodiment, the dominant class corresponds to the maximum likelihood score in the first set of class scores 1040 or the second set of class scores 1050. The classified point is then assigned a dominant class, which serves as the finalized label 1060 for the classified point. For an uncertain point, the finalized label 1060 for the uncertain point is determined to be the dominant class from the new set of class scores.

[0115] In an embodiment, the final determined labels 1060 of the points in the point cloud are provided to, for example, the perception module 402 for tasks such as object recognition. For example, the perception module 402 applies a point cloud-based object recognition algorithm (such as VoxelNet, etc.) to detect objects in the point cloud. In an embodiment, based on the detected objects, the planning module 404 ( Figure 4 ) outputs a strategy for following the road or avoiding collisions with nearby vehicles.

[0116] Fig.11 A view network 1100 is shown representing both the RV network 1004 and the BeV network 1006. The view network 1100 takes as input view data 1120 representing either the RV data 1020 or the BeV data 1030 and outputs a set of class scores 1110. In an embodiment, the view data 1120 is passed through a series of convolutional layers 1102. A convolutional layer is a layer in a neural network that performs convolutions on the input to the layer. A convolution is an operation that convolves a convolution kernel (e.g., a 5×5 matrix) with an input tensor to produce a new tensor. In an embodiment, the convolutional layer is replaced by a transposed convolutional layer. A transposed convolutional layer is a layer in a neural network that performs upsampling using a transposed convolution on the input to the layer. A transposed convolution may be performed using a convolution on the input with a padded border.

[0117] In an embodiment, multiple maximum pooling layers are embedded between consecutive convolutional layers 1102, each of which extracts the dominant characteristics of the input to each layer. A maximum pooling layer is a layer in a neural network that performs maximum pooling on the input to the layer. Maximum pooling is a pooling operation that calculates the maximum value in each patch (e.g., a 3×3 region of the input to each layer of the input tensor).

[0118] In an embodiment, an activation function is included in some convolutional layers 1102. The activation function is a function that rectifies the output of the layer. For example, the activation function is a sigmoid function or a rectified linear unit (ReLU) function.

[0119] A spatial smoothing layer 1104 implemented using a recursive layer or multiple consecutive recursive layers is embedded between consecutive convolutional layers 1102. A recursive layer is a neural network layer with an internal memory. Whenever there is a new input, the memory is updated. The output is calculated using the current input and the internal memory. As a result, the recursive layer is able to learn sequential relationships in the input.

[0120] During training, in an embodiment, data enhancement is performed on the generated RV data 1020 or BeV data 1030. Data enhancement is the process of generating more training samples based on existing training samples. For example, data enhancement of the RV data 1020 includes splitting the RV data into multiple smaller segments. These smaller segments help the network perform better on small objects. In addition, data enhancement increases the number of available training samples.

[0121] In an embodiment, in the view network 1100, one or more consecutive fully connected layers 1106 are included before the output set of class scores 1110. A fully connected layer is a layer in a neural network, wherein in the layer, the neurons in the fully connected layer have full connections with all outputs in the previous layer. Neurons in a neural network are components with learnable weights and biases. In another embodiment, the fully connected layer 1106 is replaced by a convolutional layer 1102. During training, the weights and biases of each neuron in the neural network are updated so that the actual output of the neural network converges to the expected output of the neural network. In an embodiment, the update is performed via back propagation. Back propagation is the process of propagating the difference between the actual output and the expected output relative to the gradient of each weight of the neural network. The difference between the actual output and the expected output is calculated according to a loss function. The loss function is a measure designed to calculate the error between the actual output and the expected output.

[0122] A set of class scores 1110 output for a point includes likelihood scores for different classes such as: a vegetation score 1112, i.e., a likelihood score that the point is part of some vegetation; a vehicle score 1114, i.e., a likelihood score that the point is part of some vehicle; a pedestrian score 1116, i.e., a likelihood score that the point is part of some pedestrian; and a road surface score 1118, i.e., a likelihood score that the point is part of some road surface; and so on.

[0123] Fig.12 A representation of a fusion network 1200 is shown. The fusion network 1200 takes the uncertainty point 950 as input and outputs a new set of class scores 1220 for the uncertainty point 950.

[0124] During training, a sampler 1202 is used to filter the uncertain points 950 so that only a portion of the uncertain points 950 are evaluated. This increases the speed of training. In an embodiment, the sampler 1202 is implemented using a probability function. In an embodiment, the sampler 1202 is implemented using a filter function. The sampler 1202 is used to specify a threshold for the difference between the two sets of class scores 1040 and 1050. During deployment, all uncertain points 950 are considered.

[0125] In an embodiment, a K-dimensional tree (KD tree) 1204 selects neighboring points 1230 of an uncertain point 950. A KD tree is a multidimensional binary search tree structure for organizing points in a k-dimensional space, which is useful for nearest neighbor search. The neighboring points 1230 are provided as input to a feature extractor and concatenator 1206. The neighboring points 1230 of an uncertain point 950 are points near the uncertain point 950. In an embodiment, the KD tree is replaced by another nearest neighbor search algorithm (such as a linear search, etc.).

[0126] Features are defined as outputs from layers of the view network 1100. In an embodiment, the feature extractor and concatenator 1206 takes the uncertain point 950 as input and extracts features of the neighboring points 1230 and the uncertain point 950 from the corresponding view network 1100 to form concatenated features as input to the point neural network 1210 or point network. In an embodiment, the feature is the output from an intermediate layer of the view network 1100. In an embodiment, the output from the intermediate layer of the view network 1100 is a tensor. In an embodiment, the feature is the output from the final layer of the view network 1100, or a set of class scores 1110 outputted. In an embodiment, the features include unprocessed point cloud data read from a LiDAR device, three-dimensional position information of neighboring points, or point intensity information of neighboring points.

[0127] In an embodiment, the cascade features 1240 from the neighboring points include the original point cloud data, the three-dimensional position information of the neighboring points relative to the uncertain points, or the point intensity information of the neighboring points. In an embodiment, the cascade features 1242 from the uncertain points include the original point cloud data, the three-dimensional position information of the uncertain points, or the point intensity information of the uncertain points.

[0128] Point network 1210 takes as input the concatenated features 1240 from neighboring points and the concatenated features 1242 from uncertain points. In an embodiment, the concatenated features 1240 from neighboring points are input by a multilayer perceptron 1212, which outputs features of some condensed form of the concatenated features 1242 from uncertain points. Multilayer perceptron 1212 is a neural network in which each node of the neural network is a perceptron. Perceptron is an algorithm for learning binary classifiers. In an embodiment, multilayer perceptron 1212 is replaced by a convolutional neural network. In an embodiment, multilayer perceptron 1212 is replaced by a transformer. A transformer is a type of neural network that transforms an input sequence into an output sequence.

[0129] The output from the multi-layer perceptron 1212 is then provided to a max pooling layer 1214. The max pooling layer 1214 extracts the dominant characteristics from the neighboring points. In an embodiment, the dominant characteristics are then concatenated with the concatenated features 1242 from the uncertain points via a concatenation layer 1216. The output from the concatenation layer 1216 is fed to a fully connected layer 1218. In an embodiment, the concatenation layer 1216 is replaced by a layer that is connected or stacked. In an embodiment, the fully connected layer 1218 is replaced by a convolutional layer.

[0130] The output of the fully connected layer is a new set of class scores 1220 for the uncertain point 950. The new set of class scores 1220 includes likelihood scores for different classes such as: vegetation score 1222, i.e., the likelihood score that the point is part of some vegetation; vehicle score 1224, i.e., the likelihood score that the point is part of some vehicle; pedestrian score 1226, i.e., the likelihood score that the point is part of some pedestrian; and road surface score 1228, i.e., the likelihood score that the point is part of some road surface; and so on.

[0131] Typically, based on a selected distance metric such as cosine distance or Euclidean distance, the new set of class scores 1220 for the uncertain point 950 is significantly different from the first set of class scores 1040 for the uncertain point 950 or the second set of class scores 1050 for the uncertain point 950. Therefore, a finalized label 1060 may be assigned to the uncertain point 950 based on the new set of class scores 1220. In an embodiment, the finalized label 1060 for the uncertain point 950 is the class corresponding to the maximum likelihood score in the new set of class scores 1220.

[0132] Fig.13 A flow chart of a process 1300 for operating a vehicle based on classified LiDAR points is shown. In an embodiment, the vehicle is Figure 1 AV 100 is shown. In an implementation, process 1300 is performed by, for example, Figure 3 In an implementation, the process 1300 is performed by a processor such as the processor 304 shown. Figure 4 The processing is performed by the perception module 402, planning module 404, control module 406 or positioning module 408 shown.

[0133] The processor receives (1302) point cloud data. In an implementation, the point cloud data is as follows Fig.10 Point cloud data 1003 generated from LiDAR device 1001 is shown. In an implementation, the LiDAR point cloud data includes information related to a color associated with at least one point included in the point cloud. In an implementation, the LiDAR point cloud data includes point intensity information.

[0134] The processor generates (1304) a first view of the point cloud and a second view of the point cloud based on a plurality of points in the point cloud, wherein the second view is different from the first view. In an implementation, the first view of the point cloud is Fig.10 Range view (RV) data 1020 is shown. In an implementation, the second view of the point cloud is Fig.10 Bird's eye view (BeV) data 1040 is shown.

[0135] The processor provides (1306) the first view as input to a first view neural network and provides (1306) the second view as input to a second view neural network, the second view neural network being different from the first view neural network. In an implementation, the first view is range view (RV) data 1020, and the first view neural network is Fig.10 The range view (RV) network 1004 is shown. In an implementation, the second view is the bird's eye view (BeV) data 1040, and the second view neural network is Fig.10 The range view (RV) network 1006 is shown. In an implementation, the first view neural network or the second view neural network includes at least one recurrent layer, such as Fig.11 The spatial smoothing layer 1104 shown in FIG. 1104 and the like. In an implementation, during training of the first view neural network or the second view neural network, such as for Fig.11 The first view data or the second view data is generated at least in part based on data enhancement.

[0136] The processor generates (1308) a first set of class scores indicating the class of the object using the first view neural network for each point in the point cloud, and generates (1308) a second set of class scores indicating the class of the object using the second view neural network, wherein the first set of class scores and the second set of class scores are generated in parallel. In an embodiment, the first set of class scores is Fig.10 The first set of class scores from RV is shown as 1040. In an embodiment, the second set of class scores is Fig.10 A second set of class scores from BeV is shown 1050. In implementations, at least one class score included in the first set of class scores or the second set of class scores for a particular point corresponds to a predefined class of an object.

[0137] The processor determines (1310) at least one uncertain point in the point cloud, wherein the determination is based on a first set of class scores for the at least one uncertain point and a second set of class scores for the at least one uncertain point. Fig.10 The score comparator 1008 shown compares the first set of class scores and the second set of class scores. In an implementation, at least one uncertainty point is Fig. 9 The uncertainty point 950 is shown. In implementation, as for Fig.10As described, the uncertain point is determined relative to a threshold difference of the class score, wherein the threshold difference is determined based on at least one or both of a probability function and a filter function. In implementation, the point determined to be not an uncertain point is a classified point 960 .

[0138] The processor generates (1312) a third set of class scores based on at least one of the first set of class scores for the at least one uncertain point and the second set of class scores for the at least one uncertain point using a fused neural network, wherein the third set of class scores is based on characteristics of neighboring points of the at least one uncertain point. In an embodiment, the third set of class scores for the at least one uncertain point is Fig.12 The new set of class scores 1220 is shown. In an implementation, at least one class score in the third set of class scores for at least one uncertain point corresponds to a predefined class of the object. In an implementation, the fused neural network includes at least one recursive layer. In an implementation, the fused neural network includes a multilayer perceptron and a convolutional layer (such as Fig.12 The multilayer perceptron 1212 and Fig.11 At least one of the convolutional layers shown, etc.

[0139] The processor determines (1314) a finalized label for the at least one uncertainty point based on the third set of class scores. In an implementation, the finalized label is a label that is consistent with an element in the new set of class scores 1220 (such as Fig.12 The object class (e.g., vegetation, vehicle, pedestrian, or road surface) associated with the vegetation score 1222, vehicle score 1224, pedestrian score 1226, or road surface score 1228 shown in FIG. 12 is a vector of the object class ... Fig.12 The cascade features described above) determine the class score of at least one uncertain point. In an implementation, the cascade features include intermediate outputs of intermediate layers of the first view neural network and the second view neural network, such as from Fig.11 The output of the intermediate layer of the view network 1100, etc. In an implementation, the cascaded features include class scores output from at least one of the first view neural network and the second view neural network, such as Fig.11 The outputted set of class scores 1110 shown in FIG. 1110 and the like. In the implementation, the Fig.10 The finalization module determines a finalized label of at least one classified point.

[0140] The processor identifies (1316) at least one object near the vehicle based at least in part on the final determined label of the at least one uncertainty point. In an implementation, the object is an element (such as Fig.12The identification is based at least in part on a final determined label of at least one classified point.

[0141] The processor controls (1318) the movement of the vehicle. In implementation, the processor is the planning module 404, the control module 406, or the positioning module 408 for controlling the vehicle to follow a planned path for the vehicle to avoid collisions with known objects.

[0142] In the previous description, embodiments of the present invention have been described with reference to many specific details, which may be different due to implementation. Therefore, the description and the accompanying drawings should be regarded as illustrative, rather than restrictive. The only and exclusive indication of the scope of the present invention, and the applicant's expectation that the content of the scope of the present invention is the literal and equivalent scope of the claims issued from this application in the specific form of issuing the claims, including any subsequent amendments. Any definition of the terms used to be included in such claims clearly set forth herein should be based on the meaning of such terms as used in the claims. In addition, when the term "also includes" is used in the previous description or the attached claims, the following of the phrase can be an additional step or entity, or a sub-step / sub-entity of the previously described step or entity.

Claims

1. A method for a vehicle, comprising: Using at least one processor, receiving LiDAR point cloud data associated with a plurality of points in the point cloud; generating, using the at least one processor, a first view of the point cloud and a second view of the point cloud based on the plurality of points in the point cloud, wherein the second view is different from the first view; using the at least one processor, providing the first view as an input to a first view neural network and providing the second view as an input to a second view neural network, the second view neural network being different from the first view neural network; generating, using the at least one processor, for each point in the point cloud, a first set of class scores indicative of an object class using the first view neural network and generating, using the second view neural network, a second set of class scores indicative of the object class, wherein the first set of class scores and the second set of class scores are generated in parallel; determining, using the at least one processor, a final determined label for at least one point in the point cloud, wherein the determination is based on a first set of class scores for the at least one point and a second set of class scores for the at least one point; identifying, using the at least one processor, at least one object in proximity to the vehicle based at least in part on the finalized label of the at least one point; as well as controlling, using the at least one processor, movement of the vehicle based on the at least one object, The first view is a bird's-eye view, namely, BeV, and the second view is a range view, namely, RV.

2. The method according to claim 1, wherein: Determining a finalized label of at least one point in the point cloud includes: determining at least one uncertain point in the point cloud, wherein the determining is based on a first set of class scores for the at least one uncertain point and a second set of class scores for the at least one uncertain point; generating, using a fused neural network, a third group class score for the at least one uncertain point based on at least one of the first group class score for the at least one uncertain point and the second group class score for the at least one uncertain point, wherein the third group class score is based on characteristics of neighboring points of the at least one uncertain point; and Based on the third set of class scores, a final determined label for the at least one uncertainty point is determined using the at least one processor.

3. The method according to claim 1 or 2, wherein: The LiDAR point cloud data includes information related to a color associated with at least one point included in the point cloud.

4. The method according to claim 1 or 2, wherein: The LiDAR point cloud data includes point intensity information.

5. The method according to claim 1 or 2, wherein: At least one class score included in the first set of class scores, the second set of class scores, or the third set of class scores for the particular point corresponds to a predefined class of the object.

6. The method according to claim 2, wherein: At least one of the first view neural network, the second view neural network, and the fused neural network includes at least one recurrent layer.

7. The method according to claim 1 or 2, wherein: Providing the first view data as input to a first view neural network and providing the second view data as input to a second view neural network includes: The first view data is provided as input to a first view neural network and the second view data is provided as input to a second view neural network, the first view data or the second view data being generated at least in part based on data augmentation.

8. The method according to claim 2, wherein: The uncertainty point is determined relative to a threshold difference of the class score, wherein the threshold difference is determined based on at least one of a probability function and a filter function, or both.

9. The method according to claim 2, wherein: The class score of the at least one uncertain point is determined based on concatenated features of neighboring points of the at least one uncertain point.

10. The method according to claim 9, wherein: The cascaded features include intermediate outputs of intermediate layers of the first view neural network and the second view neural network.

11. The method according to claim 9, wherein: The cascaded features include class scores output from at least one of the first view neural network and the second view neural network.

12. The method according to claim 2, wherein: The fused neural network includes at least one of a multilayer perceptron and a convolutional layer.

13. A vehicle, comprising: at least one LiDAR device capable of generating a LiDAR scan point cloud comprising a plurality of LiDAR data points; as well as A processing circuit is coupled to the LiDAR device, wherein the processing circuit is configured to perform the method according to claim 1.

14. A non-transitory computer-readable storage medium comprising at least one program for execution by at least one processor of a first device, the at least one program comprising instructions which, when executed by the at least one processor, cause the first device to perform the method according to claim 1.

15. A computer program product comprising a program for causing a computer to execute the method according to any one of claims 1 to 12.

Images