Systems and methods for ensuring the reliability of probabilistic delays in cloud robotics

By utilizing multiple independent network interfaces and computing nodes through a reliable offloading framework, the problem of unreliable connection between autonomous robots or vehicles and cloud servers is solved, achieving low-latency and high-reliability communication. The pairing of network interfaces and computing nodes is optimized, improving the system's fault tolerance.

CN123123372APending Publication Date: 2026-11-17ROBERT BOSCH GMBH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610681458.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2025-05-16
Filing Date
2026-05-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the connection between autonomous robots or vehicles and cloud servers is susceptible to network failures and congestion, resulting in unreliable delays and an inability to achieve real-time or near-real-time computational offloading.

Method used

A reliable offloading framework is adopted, which performs packet replication and transmission through multiple independent network interfaces and computing nodes. The connectivity between autonomous robots or vehicles and cloud servers is realized by using agent computing nodes based on the communication framework, optimizing the pairing of network interfaces and computing nodes, and reducing latency and failure probability.

Benefits of technology

It significantly reduces latency between autonomous vehicles or robots and cloud servers, lowers the probability of failure, ensures reliable and low-latency communication, and enhances proactive measures against latency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123372A_ABST
    Figure CN123123372A_ABST
Patent Text Reader

Abstract

Systems and methods for ensuring probabilistic latency reliability in cloud robotics are provided. A method for transmitting communication based on a communication framework (e.g., Robot Operating System (ROS2) communication) between an autonomous robot or vehicle and a remote cloud server via a proxy computing node is disclosed. The reliable offloading framework orchestrates the transmission and reception of data packets via multiple network interfaces, thus reducing latency in communication between the autonomous robot or vehicle and the cloud server, and also improving the reliability of those types of communication. The pairing between the corresponding network interfaces (both on the cloud server side and the autonomous robot or vehicle side) is optimized to minimize the probability of missing deadlines. The probability of missing deadlines is also significantly reduced by providing a replicated server with uncorrelated failures.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to orchestrating reliable and low-latency communication between autonomous robots or vehicles connected to a cloud. Background Technology

[0002] In the past, cloud connectivity services in mobility and robotics have often suffered from unreliability due to network congestion, network failures, or server outages. To make offloading computation feasible for more demanding applications, the latency reliability of cloud services also needs to be extended accordingly. As used in this paper, improved latency reliability can be defined as reducing the mean end-to-end latency of communication with cloud services, thus reducing the maximum end-to-end latency, and reducing the jitter of this latency over time.

[0003] Past implementations not optimized for latency reliability typically referred to connectivity stacks in robots or automobiles that used a single internet connection to a single endpoint in the cloud. This endpoint could be a single server, or it could be a load balancer with a single Uniform Resource Locator (URL) that distributes requests to other servers in the background. However, this resulted in multiple single points of failure (SPOF), including instances where the entire system failed and experienced high latency when a communication link failed or became congested, or when an endpoint or server failed or was overutilized to the point where failure was inevitable. Summary of the Invention

[0004] In one embodiment, a method for establishing connectivity between a proxy computing node and a cloud server is provided. The method includes: generating a connectivity thread for a corresponding network interface within an autonomous vehicle, the network interface being configured to send and receive data packets to and from a cloud server, wherein the cloud server orchestrates computer vision operations of the autonomous vehicle; generating a Linux-based socket for the network interface; executing a cloud-deployed Session Traversal Utility (STUN) protocol for network address translation to output a cloud-accessible network Internet Protocol (IP) address for the corresponding network interface in the network interface; establishing connectivity between the corresponding network interface in the network interface and a corresponding computing node of the cloud server based on the cloud-accessible network IP address; and, based on the established connectivity, sending and receiving communication based on a communication framework via the proxy computing node located within the autonomous vehicle for performing computer vision-related operations of the autonomous vehicle.

[0005] In another embodiment, a method is provided for transmitting communication based on a communication framework via an agent computing node. The method includes: receiving a request from a communication framework-specific application located within sensors of an autonomous vehicle and to an agent computing node located within the autonomous vehicle to send a data packet to a cloud server, wherein the cloud server orchestrates computer vision-related operations of the autonomous vehicle; mapping two or more network interfaces to corresponding computing nodes of the cloud server based on an optimized combination of network interface and computing node pairings, wherein the two or more network interfaces are intended for transmitting the data packet; copying the data packet and subsequently transmitting the data packet via the two or more network interfaces; receiving communication based on the communication framework for the agent computing node via a given one of the two or more network interfaces; and providing the communication based communication to an actuator of the autonomous vehicle for performing a given computer vision-related operation of the autonomous vehicle.

[0006] In another embodiment, a system includes a processor and a memory, the memory containing instructions that, when executed by the processor, cause the processor to perform these steps.

[0007] In another embodiment, a non-transitory computer-readable medium includes instructions that, when executed by the processor, cause the processor to perform these steps. Attached Figure Description

[0008] Figure 1 The illustration shows a high-level overview of a software framework according to some embodiments, which enables reliable and low-latency communication between cloud-connected mobile robots or vehicles when offloading latency-sensitive, stateless tasks.

[0009] Figure 2 The illustration depicts a workflow, according to some embodiments, for establishing connectivity between an autonomous robot or vehicle and one or more computing nodes of a cloud server, and for transmitting communication based on a communication framework (e.g., communication based on a robot operating system (ROS)) via a proxy computing node.

[0010] Figure 3A This is a flowchart illustrating a process for establishing connectivity between an autonomous vehicle and one or more computing nodes of a cloud server, according to some embodiments.

[0011] Figure 3B This is a flowchart illustrating a process, according to some embodiments, of transmitting communication based on a communication framework between an autonomous vehicle and one or more computing nodes of a cloud server via a proxy computing node.

[0012] Figure 4AIt is a geographic map, which, according to some embodiments, illustrates... Figures 4A-4F The first case study shown illustrates an example driving route taken by an autonomous vehicle, which uses the systems and methods described herein to orchestrate ROS2-based communication between the autonomous vehicle and one or more compute nodes of a cloud server.

[0013] Figure 4B The diagram illustrates along Figure 4A The driving route shown is at one point in time where a handover resulted in a service interruption due to a lack of use of the systems and methods described herein.

[0014] Figure 4C The illustration shows the path along according to some embodiments. Figure 4A At the same time point of the driving route shown, the handover did not cause service interruption due to the use of the systems and methods described herein.

[0015] Figure 4D illustrates the situation when using a single service provider (e.g., Verizon) and not applying the systems and methods described herein to cross-regional handovers occurring along the driving route. Figure 4A Multiple services were disrupted along the driving routes shown.

[0016] Figure 4E illustrates the situation when using another single service provider (e.g., AT&T) and not applying the systems and methods described herein to a cross-regional handover occurring along the driving route. Figure 4A Several other services were disrupted along the driving routes shown.

[0017] Figure 4F The illustration shows, according to some embodiments, along... Figure 4A The reduction in service interruptions along the driving route during the application of the cross-zone handover shown in the document, due to the use of the systems and methods described herein.

[0018] Figure 5A The illustration shows a second case study implementation of the system and method described herein according to some embodiments (in... Figures 5A-5E As shown in the figure, the two computing nodes of the cloud server orchestrate computer vision-related tasks for the autonomous robot.

[0019] Figure 5B The illustration shows a comparison between using the systems and methods described herein (i.e., the reliable offloading framework in the figure) and not using the systems and methods described herein (i.e., Wi-Fi and 5G in the figure) according to some embodiments.

[0020] Figure 5C The diagram illustrates when... Figure 5AThe second case study introduced in this paper describes the accumulated latency and latency frequency when the system and methods described in this paper were not applied during implementation, but only Wi-Fi connectivity was used instead.

[0021] Figure 5D The diagram illustrates when... Figure 5A The second case study introduced in this paper describes another amount of latency and latency frequency accumulated when the system and methods described in this paper were not applied during implementation, but only 5G connectivity was used instead.

[0022] Figure 5E The illustration shows, according to some embodiments, when in Figure 5A The second case study introduced in this paper describes a significantly reduced latency and latency frequency when the system and methods described in this paper were applied during implementation. Detailed Implementation

[0023] Embodiments of this disclosure are described herein. However, it is to be understood that the disclosed embodiments are merely examples and other embodiments may take various and alternative forms. The drawings are not necessarily to scale; some features may be exaggerated or minimized to show detail of particular components. Therefore, the specific structural and functional details disclosed herein are not to be construed as limiting, but merely as a representative basis for teaching those skilled in the art to adopt the embodiments in various ways. As will be understood by those skilled in the art, the various features illustrated and described with reference to any of the accompanying drawings may be combined with features illustrated in one or more other drawings to produce embodiments not explicitly illustrated or described. The combinations of illustrated features provide representative embodiments of typical applications. However, for a particular application or implementation, various combinations and modifications of features consistent with the teachings of this disclosure may be desired.

[0024] Unless the context clearly indicates otherwise, as used herein, “a,” “an,” and “the” refer to both the singular and plural references. As an example, “processor” programmed to perform various functions refers to a processor programmed to perform each of the various functions, or multiple processors collectively programmed to perform each of the various functions.

[0025] As described above, when implementing cloud robotics services for autonomous robots and vehicles, past implementations have been affected by multiple single points of failure (SPOFs), which inevitably cause high latency between the autonomous robot or vehicle and the cloud-based computing services and instructions that should be sent and received in real time or near real time.

[0026] Past implementations have attempted to address this issue in various ways. For example, some approaches would handle server failures and some degree of network congestion by configuring multiple redundant servers and switching between them after a failure is detected. However, these attempted solutions fail to address the "first hop" of network connectivity (such as a Wi-Fi or 5G network interface link), instead assuming that this link is reliable and will remain reliable, and provides sufficient network characteristics such as bandwidth and latency. In another example, other approaches handle server failures simply by replicating the connection across the network (web), failing to properly address network connectivity.

[0027] Other past implementations (such as Apple's Wi-Fi Assist) attempted to resolve the first hop in network connectivity by switching between network interfaces in the event of network degradation. However, such implementations only react to network failures and the switching between network interfaces takes a long time (e.g., within a few seconds). Switching in this way often also triggers the establishment of a new connection. In the event that a network interface is missing, applications either do not react at all and stop working, or they need to restart and / or reload some aspect of themselves (e.g., restart a stream, re-establish a connection to the server, etc.).

[0028] In other past implementations (such as Multipath Transmission Control Protocol (TCP)), multiple local network interfaces were used to transmit a single input data stream to a single server with slightly more reliability by creating multiple TCP connections via multiple paths and then distributing the traffic across those so-called substreams. In multipath TCP, the focus is on utilizing a single server for delivery, allowing SpOF (Split-Point-of-Flight) to occur within the system itself. Multipath TCP also does not perform packet replication by default; instead, a single packet is delivered only on a single substream. This is also a significant limitation, as actions against latency are only reactive rather than proactive.

[0029] In contrast, the systems and methods described herein involve a probabilistically latency-reliable cloud robotics platform that operates on unreliable, commercially available infrastructure. Even when using unreliable, commercially available infrastructure (defined herein as inexpensive, latency-unreliable, and error-prone computing, networking, and storage infrastructure), the probabilistically latency-reliable cloud robotics platform significantly reduces latency between autonomous vehicles or robots and their cloud-based servers(s). Furthermore, by providing replicated resources with uncorrelated failures, the probability of failure is reduced exponentially.

[0030] For ease of discussion in this paper, the systems and methods described herein concerning probabilistic latency-reliable cloud robotics platforms are also referred to as reliable offloading frameworks. A reliable offloading framework can be defined as a cloud robotics framework that uses multiple independent networks for communication between autonomous robots or vehicles and fault-independent computing resources in cloud servers, based on a communication framework (ROS2-based communication), thereby achieving probabilistic latency reliability on commercial cloud infrastructure. It should be understood that the use of any communication framework for cloud robotics (such as ROS2, Message Queuing Telemetry Transport (MQTT), or Open Platform Communication Unified Architecture (OPC-UA)) is implied as included in the description herein. The type of communication framework is agnostic to the reliable offloading framework.

[0031] Furthermore, the reliable offloading framework is configured to discover and connect to multiple non-default network interfaces on autonomous robots or vehicles, and simultaneously route messages to the same service deployed in up to multiple cloud data centers via these same non-default network interfaces. It then uses the first response received from the replicated service as an instruction for that autonomous robot or vehicle within the context of stateless computer vision-related operations. Therefore, as long as at least one replica and network remain operational and responsive, the reliable offloading framework significantly improves the probability of obtaining timely responses to and from those fundamental cloud robotics components. The reliable offloading framework optimizes the selection of network interfaces for cloud-deployed robotics algorithms by minimizing the probability of missing deadlines.

[0032] The following description continues the introduction of a reliable offloading framework, defined as a cloud robotics framework that applies independent network interfaces and cloud services for latency reliability within the context of cloud robotics applications. Implementations of the reliable offloading framework during (1) establishing connectivity and (2) orchestrating communication based on a communication framework between the autonomous robot or vehicle and the cloud server are then described. Additionally, an algorithm that can be executed to determine the optimal selection of network interface and computing node pairings is discussed. This disclosure then demonstrates the versatility of the systems and methods described herein for integration into various technologies and larger systems, such as for human tracking via autonomous robots and for autonomous driving implementations.

[0033] Figure 1 The illustration shows a high-level overview of a software framework according to some embodiments, which enables reliable and low-latency communication between cloud-connected mobile robots or vehicles when offloading latency-sensitive, stateless tasks.

[0034] like Figure 1As shown, mobile autonomous robots and autonomous vehicles (such as autonomous robot 102) can connect to cloud services for various applications. In the cloud robotics environment 100, autonomous robot 102 is configured to connect to clouds 122 and 128 via multi-interface route 108.

[0035] By applying paradigms and software frameworks to achieve probabilistically latency-reliable distributed services in unreliable public networks using commercial hardware, the Reliable Offload Framework enables computationally intensive algorithms to be offloaded from autonomous robots 102 to clouds 122 and 128, while also providing Quality of Service (QoS) standards in terms of latency and availability. The Reliable Offload Framework utilizes multiple layers of redundancy and replication to achieve fault tolerance and latency reliability, thus eliminating SpOF (Split-Point-of-Flight).

[0036] As shown in the cloud robotics environment 100, the reliable offloading framework is configured to transparently proxy communication based on a communication framework, such as ROS2-based communication. Program instructions are configured to send communication-based communications to replicated cloud virtual machines (VMs) (such as Cloud 122 and Cloud 128) via multiple network interfaces (such as Wi-Fi network interface 114 and 5G network interface 116). The reliable offloading framework proxies 118 and 124 then provide those communications to compute nodes 120 and 126 of the corresponding cloud servers, which are responsible for orchestrating stateless computer vision-related operations for the autonomous robot 102.

[0037] For example, camera 104 may resemble a sensor located within autonomous robot 102, configured to provide image-based data samples to clouds 122 and 128 via a reliable offloading framework. More specifically, image-based data samples from camera 104 or some other sensor of autonomous robot 102 can be sent from a communication framework-specific application to an agent computing node located within autonomous robot 102. The information may also include requests to send data packets (or, in this case, image-based data samples) to a cloud server responsible for orchestrating stateless computer vision-related tasks and operations of autonomous robot 102.

[0038] A proxy computing node located within the autonomous robot 102 then copies the data packet and subsequently transmits it via at least two network interfaces, such as Wi-Fi network interface 114 and 5G network interface 116. Proxy computing nodes 118 and 124 then receive the data packet and relay it to computing nodes 120 and 126 in clouds 122 and 128. Similarly, based on data packets received from camera 104, computing nodes 120 and 126 can then prepare and request the transmission of communication based on a communication framework to the autonomous robot 102, such as program instructions for actuators (e.g., controller 106) of the autonomous robot 102. Proxy computing nodes 118 and 124 then copy those program instructions and transmit them to a multi-interface router 108 via at least two network interfaces, such as Wi-Fi network interface 114 and 5G network interface 116, so that the program instructions can be routed to controller 106. This process can continue during the operational deployment of the autonomous robot 102.

[0039] Furthermore, the first response to receiving the multi-interface route 108 is the set of program instructions provided to the controller 106. For example, if the program instructions are first received via the Wi-Fi network interface 114, they are then provided to the controller 106 without waiting for repeated instructions from the 5G network interface 116. This ensures the latency reliability of the communication framework.

[0040] A reliable offloading framework acts as a proxy, enabling unmodified applications based on the communication framework to seamlessly integrate with the communication framework illustrated in the cloud robotics environment 100. Latency reliability is achieved by replicating requests, data packets, and program instructions to cloud servers in different regions via multiple independent network interfaces, and subsequently routing the first response back. Furthermore, by selecting different combinations of network interface and compute node pairings, several potentially related failures are reduced.

[0041] In the example above, image-based data samples from camera 104 are used as an example of the type of data packet that can be copied and transmitted using the methods described herein. However, various other sensors located within the autonomous robot 102 also imply coverage by “data packets” such as audio-based data samples, video-based data samples, or text-based data samples.

[0042] The cloud robotics environment 100 also illustrates the delay-aware and deadline-aware interface – the cloud matcher 112. The reliable offloading framework is also configured to map two or more network interfaces to corresponding compute nodes on the cloud server based on an optimized combination of network interface and compute node pairings. This mapping minimizes the probability of missing deadlines. These mappings are used when data packets are sent.

[0043] To determine which of the available network interfaces should be used to connect to which compute node of a given cloud server and / or to a different cloud server, the reliable offloading framework is configured to minimize the probability of missing a delayed deadline by applying a deadline-aware interface-cloud matcher 112. The deadline-aware interface-cloud matcher 112 algorithm is described below.

[0044] Assume the system consists of n possible cloud servers. The system comprises autonomous robots and includes the capability to connect to those cloud servers. One network interface. Using binary decision variables. To represent network interface Should it be used with cloud servers? Connection, where , And delay When using a network interface With cloud servers The probability of missing the deadline when connecting.

[0045] Therefore, the objective function can be written as: It is subject to the following rules.

[0046] First, the binary decision variables are defined as: Second, the corresponding autonomous robot (or vehicle) network interface should be connected to the corresponding cloud server: Third, the capacity of the number of connections to the corresponding cloud server is defined as follows: .

[0047] This optimization problem can be solved using integer programming to find the optimal allocation of the autonomous robot's (or vehicle's) network interface to the cloud server, which minimizes latency while satisfying constraints. Furthermore, Computation time and network time can be measured and utilized independently. This reduces profiling overhead. For example, if the system uses a type of cloud server for computation, network data can be collected by swapping the network interface to the cloud server.

[0048] like Figure 1The diagram further illustrates that the robot onboard unit 110 includes various components of an autonomous robot (such as 102). Because the description herein relates to the use of both autonomous robots and autonomous vehicles within the context of cloud robotics, it should be understood that a similar diagram with a camera 104 sensor located within an autonomous vehicle can be similarly operated using a reliable offloading framework.

[0049] Therefore, the following description of the interaction between computer-controlled machines and control systems can be applied to the systems and methods described herein.

[0050] Computer-controlled machines may include one or more actuators and one or more sensors. Such sensors are configured to sense conditions of the computer-controlled machine. For example, a given sensor may be configured to sense ID and / or OOD data, and a corresponding processor may be configured to determine whether the data is ID or OOD based on the teachings herein. A given sensor may be configured to encode the sensed conditions into a sensor signal and transmit the sensor signal to a control system. Non-limiting examples of such sensors include cameras, video sensors, radar, lidar, ultrasonic and motion sensors, temperature sensors, etc. In one embodiment, the sensor is an optical sensor configured to sense an optical image of the environment near the computer-controlled machine.

[0051] The corresponding control system is configured to receive sensor signals from the computer-controlled machine. As described below, the control system can also be configured to calculate actuator control commands based on the sensor signals and transmit the actuator control commands to the actuator of the computer-controlled machine.

[0052] The control system may include a receiving unit. The receiving unit may be configured to receive sensor signals from sensors and convert the sensor signals into input signals. In an alternative embodiment, the sensor signal is directly used as the input signal. It is received without a receiving unit. Each input signal This can be a portion of each sensor signal. The receiving unit can be configured to process each sensor signal to generate each input signal. Input signal It can include data corresponding to images recorded by a given sensor.

[0053] Control systems include computationally intensive robotics algorithms and machine learning models. For example, a control system can be configured to use machine learning algorithms, such as object detection models and classifiers, to process input signals. Classified into one or more labels. Computationally intensive robotics algorithms and machine learning models are configured by parameters (e.g., parameters). Parameterization. Parameters It can be stored in and provided by non-volatile memory. Computationally intensive robotics algorithms and machine learning models are configured to work from input signals. Determine the output signal Each output signal Includes each input signal Assigning information to one or more labels. Computationally intensive robotics algorithms and machine learning models can output signals. The signal is transmitted to the conversion unit. The conversion unit is configured to convert the output signal... The commands are converted into actuator control commands. The control system is configured to transmit actuator control commands to the actuator, which is configured to actuate a computer-controlled machine in response to the actuator control commands. In another embodiment, the actuator is configured to directly base its output signal on the actuator control commands. To actuate computer-controlled machines.

[0054] When an actuator receives an actuator control command, the actuator is configured to perform an action corresponding to the relevant actuator control command. The actuator may include control logic configured to transform the actuator control command into a second actuator control command for controlling the actuator. In one or more embodiments, instead of or in addition to an actuator, the actuator control command may also be used to control a display.

[0055] The control system also includes a processor and memory. The processor may include one or more processors. The memory may include one or more memory devices.

[0056] The non-volatile storage device of the control system may include one or more persistent data storage devices, such as hard disk drives, optical drives, magnetic tape drives, non-volatile solid-state devices, cloud storage devices, or any other device capable of persistently storing information. The processor may include one or more devices selected from high-performance computing (HPC) systems including high-performance cores, microprocessors, microcontrollers, digital signal processors, microcomputers, central processing units, field-programmable gate arrays, programmable logic devices, state machines, logic circuits, analog circuits, digital circuits, or any other device that manipulates signals (analog or digital) based on computer-executable instructions residing in memory. The memory may include a single memory device or multiple memory devices, including but not limited to random access memory (RAM), volatile memory, non-volatile memory, static random access memory (SRAM), dynamic random access memory (DRAM), flash memory, cache memory, or any other device capable of storing information.

[0057] The processor can be configured to load and execute computer-executable instructions residing in a non-volatile memory device that embody one or more machine learning algorithms. The non-volatile memory device may include one or more operating systems and applications. The non-volatile memory device may store content compiled and / or interpreted from computer programs created using various programming languages ​​and / or technologies, including but not limited to (and individually or in combination) Java, C, C++, C#, Objective C, Fortran, Pascal, JavaScript, Python, Perl, and PL / SQL.

[0058] The program code embodied in the algorithms and / or methodologies described herein can be distributed individually or collectively as a program product in various different forms. The program code can be distributed using a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to perform one or more aspects of the embodiments. The computer-readable storage medium (which is inherently non-transitory) can include tangible media that are volatile and non-volatile, and removable and non-removable, implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. The computer-readable storage medium can also include RAM, ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other solid-state memory technologies, portable optical disc read-only memory (CD-ROM) or other optical storage devices, magnetic tape, magnetic tape, disk storage devices or other magnetic storage devices, or any other medium that can be used to store desired information and can be read by a computer. The computer-readable program instructions can be downloaded from the computer-readable storage medium to a computer, another type of programmable data processing device, or another device, or downloaded via a network to an external computer or external storage device.

[0059] Computer-readable program instructions stored in a computer-readable medium can be used to direct a computer, other type of programmable data processing apparatus, or other device to operate in a particular manner, such that the instructions stored in the computer-readable medium produce an article of writing including instructions that implement the functions, actions, and / or operations specified in the flowchart or diagram. In some alternative embodiments, the functions, actions, and / or operations specified in the flowchart and diagram may be reordered, processed sequentially, and / or concurrently, consistent with one or more embodiments. Furthermore, any flowchart and / or diagram may include more or fewer nodes or blocks than those nodes or blocks consistently illustrated in one or more embodiments.

[0060] The process, method, or algorithm may be embodied, in whole or in part, using suitable hardware components such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), state machines, controllers, or other hardware components or devices, or a combination of hardware, software, and firmware components.

[0061] Returning to the example of a control system configured to control at least partially autonomous vehicles in the cloud robotics environment 100, the description below can also be applied to the systems and methods described herein.

[0062] In such embodiments, the autonomous vehicle 102 includes an actuator 106 and a sensor 104. The sensor 104 may include one or more video sensors, cameras, radar sensors, ultrasonic sensors, lidar sensors, and / or position sensors (e.g., GPS). One or more of these specific sensors may be integrated into the autonomous vehicle 102. In the context of sign recognition and processing, for example, the sensor 104 is a camera mounted to or integrated into the autonomous vehicle 102. Instead of the one or more specific sensors identified above, or in addition to the one or more specific sensors identified above, the sensor 104 may also include a software module configured to determine the state of the actuator 106 upon execution. A non-limiting example of the software module includes a weather information software module configured to determine the current or future weather conditions near or at locations other than the autonomous vehicle 102.

[0063] The computer vision and planning 120 and 126 of the remote control system of the autonomous robot or vehicle 102 can be configured to depend on the input signal. To detect targets near the autonomous vehicle 102. In such embodiments, the output signal... This can include information characterizing nearby targets for the autonomous vehicle 102. This information can be used to determine actuator control commands. These actuator control commands can be used to avoid collisions with detected targets. See also the section in this document. Figures 5A-5E .

[0064] In embodiments where the autonomous vehicle 102 is at least partially autonomous, the actuator 106 may be embodied in the brakes, propulsion system, engine, transmission system, or steering mechanism of the autonomous vehicle 102. Actuator control commands can be determined to control the actuator 106 so that the autonomous vehicle 102 avoids collisions with detected targets. In scenarios where adversarial attacks may occur, the system can be further trained to better detect targets or recognize changes in lighting conditions or angles of sensors or cameras on the autonomous vehicle 102.

[0065] In other embodiments where the autonomous robot 102 is at least partially autonomous, the autonomous robot 102 may be a mobile robot configured to perform one or more functions, such as flying, swimming, diving, and walking. The mobile robot may be a lawnmower or a cleaning robot that is at least partially autonomous. In such embodiments, actuator control commands may be determined, thereby controlling the mobile robot's propulsion, steering, and / or braking units to prevent collisions with identified targets.

[0066] In another embodiment, the autonomous robot 102 is at least partially autonomous in the form of a gardening robot. In such embodiments, the autonomous robot 102 may use optical sensors (such as sensor 104) to determine the state of plants in the environment near the autonomous robot 102. The actuator 106 may be a nozzle configured to spray chemicals. Based on the identified plant species and / or the identified plant state, actuator control commands may be determined to cause the actuator 106 to spray the plants with an appropriate amount of suitable chemicals.

[0067] Figure 2 The illustration depicts a workflow, according to some embodiments, for establishing connectivity between an autonomous robot or vehicle and one or more computing nodes of a cloud server, and for transmitting communication based on a communication framework (e.g., ROS2-based communication) via a proxy computing node.

[0068] In some embodiments, the cloud robotics environment 200 illustrates a reliable offloading framework for handling fault-tolerant requests. The following paragraphs first detail the workflow of establishing connectivity between at least one agent compute node and at least one compute node of the cloud server using the reliable offloading framework. Then, the following paragraphs detail another workflow of orchestrating communication based on a communication framework via the agent compute node between the reliable offloading framework and at least one compute node of the cloud server. For ease of discussion in the following paragraphs, ROS2-based communication is discussed. However, it should be understood that MQTT- or OPC-UA-based communication can be similarly applied to the following description, as the reliable offloading framework is agnostic to the type of communication framework used.

[0069] Furthermore, the specific embodiment in cloud robotics environment 200 continues the example of an autonomous robot 202 receiving program instructions from cloud servers 238 and 248. However, it should be understood that in other embodiments, autonomous vehicles may be configured to receive program instructions from cloud servers 238 and 248 in a similar manner, and the specific embodiment in cloud robotics environment 200 is not intended to be inherently limiting for the deployment configuration of the reliable offloading framework.

[0070] In addition, the above description of the sensors and actuators for the autonomous robot 102 also implies the application of sensor 204 and actuator 206 to the autonomous robot 202.

[0071] During the connection establishment phase, the reliable offloading framework is configured to generate connectivity threads for the corresponding network interfaces located within the autonomous robot 202. These network interfaces are configured to enable the sending and receiving of data packets between the autonomous robot 202 and cloud servers 238 and 248. The generation of connectivity threads may also include generating Linux-based sockets for the network interfaces, which forces the network interfaces to bind to the sockets, even if the specific network interface is not the default network interface.

[0072] Furthermore, the network interfaces selected to participate in the connectivity establishment process are fault-independent network interfaces. For example, the interruption of the Wi-Fi network interface 224 is independent of the functionality of the 5G network interface 228, and vice versa. In addition, the stability of the Wi-Fi network interface 224 does not directly affect the stability of the cloud NIC 232, and so on.

[0073] like Figure 2 As shown, a connectivity thread 226 is generated for Wi-Fi network interface 224, and a connectivity thread 230 is generated for 5G network interface 228. To reduce latency, the reliable offloading framework is configured to generate connectivity threads for at least two network interfaces, allowing copied data packets to be transmitted. So far, the reliable offloading framework has been discussed from the perspective of the autonomous robot 202. However, and as... Figure 2 As shown, additional connectivity threads are also generated from the perspective of cloud servers 238 and 248, including a connectivity thread 234 generated for cloud network interface card (NIC) 232 and another connectivity thread generated for cloud NIC 236, and so on.

[0074] Next, the Session Traversal Utility (STUN) protocol for network address translation deployed in the cloud is executed on connectivity server 218 to output the cloud-accessible network Internet Protocol (IP) address of the corresponding network interface among the network interfaces described in the previous paragraphs. By executing the STUN protocol, cloud-accessible network addresses from non-default network interfaces are ensured, regardless of whether the network interface is local or virtual (such as in the case of a private VLAN for autonomous robot 202).

[0075] A given period of time may elapse during the execution of the STUN protocol. The response fails to output after that period. Figure 2The cloud-accessible network IP address of a given network interface shown in the diagram has its corresponding connectivity thread disconnected. This particular network interface will not be used or attempted for future packet transmissions until connectivity is re-established.

[0076] For other network interfaces that have obtained cloud-accessible network IP addresses, then establish connectivity between the autonomous robot 202 and the corresponding cloud server.

[0077] For example Figure 2 As shown, the establishment of connectivity, and thus the establishment of connectivity threads between autonomous robot 202 and cloud servers 238 and 248, can be orchestrated via connectivity server 218. In some embodiments, connectivity server 218 may be located near autonomous robot 202. Furthermore, for ease of discussion, addressing and discovery actions are labeled herein as addressing service 220 and discovery service 222. Once a connectivity thread announces itself to discovery service 222 using its cloud-accessible network IP address, discovery service 222 is configured to run another protocol to facilitate the exchange of discovered addresses. Additionally, both addressing service 220 and discovery service 222 can correspond to autonomous robot 202 and cloud servers 238 and 248 via the public internet or in a private network setup. Because connectivity server 218 is used to establish connectivity, no subsequent data packets pass through connectivity server 218, thus ensuring that server failure is not caused by the use of connectivity server 218.

[0078] The following paragraphs now detail the workflow of the reliable offloading framework for orchestrating ROS2-based communication between at least one agent compute node and at least one compute node of the cloud server via the agent compute node.

[0079] This workflow occurs at a point in time after which connectivity has been established between Wi-Fi network interface 224 and cloud NIC 232 via connectivity threads 226 and 234, and connectivity has been established between 5G network interface 228 and cloud NIC 236 via connectivity thread 230 and another connectivity thread. Furthermore, autonomous robot 202 is now ready to provide data packets from sensor 204 to cloud servers 238 and 248, and is ready to receive program instructions for stateless computer vision-related operations on actuator 206.

[0080] As described above, sensor 204 is configured to provide a request to send data packets to a cloud server via a ROS2 application. Depending on the sensor type, the data packets may include image-based, audio-based, video-based, or text-based data samples. The request to send the data packets is received by a proxy computing node (such as proxy node 208, proxy 216).

[0081] Next, the request is extracted as a serialized request with a unique identifier. In some embodiments, the unique identifier is extracted from a ROS2 middleware layer (e.g., 'rmw'). The proxy node 208 is then configured to store the unique identifier and the request to send the data packet.

[0082] Subsequently, mappings are performed between two or more network interfaces and their corresponding compute nodes on at least one cloud server to prepare for connectivity threads used to send request packets. As described above regarding the deadline-aware interface-cloud matcher 112, agent 216 optimizes the pairing of network interfaces and compute nodes. For ease of discussion herein, pairing is made between Wi-Fi network interface 224 and cloud NIC 232, and another pairing is made between 5G network interface 228 and cloud NIC 236. To ensure latency reliability, at least two pairings are used to send and receive packets between autonomous robot 202 and cloud servers 238 and 248.

[0083] Then, the optimizer 212 and repeater 214 of the proxy 216 cause the data packet to be copied and subsequently transmitted to the cloud NIC 232 via the Wi-Fi network interface 224, and to the cloud NIC 236 via the 5G network interface 228. In some embodiments, the repeater 214 is also configured to register unique identifiers and requests with callbacks for responses and callbacks for timeouts in case program instructions are not subsequently received from the cloud server 238 or the cloud server 248 in a timely manner.

[0084] On the cloud server 238 and 248 side, data packets are received by agent 246, which then provides the data packets to the corresponding computing node, such as computing node 240 provided to cloud server 238, via repeater 244 and agent node 242. Computing node 240 then executes computationally intensive algorithms or machine learning models for computer vision-related tasks of autonomous robot 202 using the received data packets. Once computing node 240 outputs program instructions in a ROS2-based communication format for autonomous robot 202, agent 246 is then configured to similarly copy the program instructions and transmit them via cloud NICs 232 and 236.

[0085] When agent 216 receives a given ROS2-based communication (such as via Wi-Fi 224 or 5G network interface 228), agent 216 is configured to verify that the ROS2-based communication includes the same unique identifier that matches the original unique identifier of the original request. Once verified, agent 216 provides the ROS2-based communication to actuator 206 for performing computer vision-related operations of autonomous robot 202.

[0086] When a first response is received (such as via Wi-Fi 224), and if the unique identifier does match, agent 216 marks the given request as complete. Upon receiving a second, third, or subsequent response via another network interface (such as via 5G network interface 228), those redundant or duplicate responses are discarded because the request has already been marked as complete.

[0087] Furthermore, upon receiving the first response, agent 216 is additionally configured to verify the elapsed time or a period less than the originally set timeout callback. Timeout handler 210 monitors the elapsed time.

[0088] Furthermore, the reliable offloading framework is configured to handle packet processing by generating independent connectivity threads for the corresponding network interfaces and to ensure efficient internal processing in preparation for potential packet retransmissions due to packet loss. By performing asynchronous operations throughout the entire packet processing pipeline, the reliable offloading framework continues processing packets from other network interfaces without delay. Additionally, leveraging zero-copy constraints, packets are copied and transmitted via two or more network interfaces, where packets are not duplicated in memory during processing. This minimizes the overhead associated with using multiple network interfaces, thus maintaining efficiency and preserving the integrity of fault-tolerant network interfaces.

[0089] The reliable offloading framework uses asynchronous operations throughout the packet processing pipeline, allowing the agent to continue processing packets from other interfaces without latency. Furthermore, the reliable offloading framework leverages the zero-copy property to process packets, meaning that packet data is not duplicated in memory during processing. This minimizes the overhead associated with using multiple interfaces, maintaining efficiency and upholding the integrity of the independence assumption.

[0090] Figure 3A This is a flowchart illustrating a process for establishing connectivity between an autonomous vehicle and one or more computing nodes of a cloud server, according to some embodiments.

[0091] In some embodiments, process 300 runs in parallel with the above for Figure 2The provided description states that the reliable offloading framework establishes connectivity between at least one agent compute node and at least one compute node of the cloud server.

[0092] In block 302, a connectivity thread is generated for the corresponding network interface within the autonomous vehicle or robot (such as autonomous robot 102 or 202) to prepare for sending data packets to and receiving data packets from at least one cloud server configured to orchestrate stateless computer vision-related operations of the autonomous robot.

[0093] During the generation of those connectivity threads, Linux-based sockets are generated for the corresponding network interfaces, and the STUN protocol for cloud deployment is executed to output the cloud-accessible network IP address of the corresponding network interface, as illustrated in blocks 304 and 306.

[0094] Using those cloud-accessible IP addresses, connectivity is thus established between the network interface and the corresponding pair of compute nodes of a given cloud server, as shown in block 308. In some embodiments, the establishment of connectivity may include the establishment of a bidirectional peer-to-peer connection or some similar type of connection.

[0095] Figure 3B This is a flowchart illustrating a process, according to some embodiments, of transmitting communication based on a communication framework between an autonomous vehicle and one or more computing nodes of a cloud server via a proxy computing node.

[0096] In some embodiments, process 350 runs in parallel with the above for Figure 2 The provided description describes a reliable offloading framework that orchestrates ROS2-based communication or some other communication framework-based communication between at least one agent compute node and at least one compute node of the cloud server. Furthermore, process 350 illustrates the workflow from the perspective of an autonomous vehicle.

[0097] In block 352, the reliable offloading framework receives requests from application-specific communication frameworks located within the sensors of the autonomous vehicle. The proxy computing node is configured to generate a unique identifier for the request and similarly record the data packets requested to be sent to the cloud server.

[0098] In block 354, the deadline-aware interface-cloud matcher 112 maps two or more network interfaces located in the autonomous vehicle to the corresponding network interfaces and computing nodes of the cloud server.

[0099] In block 356, the data packet is copied and then transmitted along two or more selected network interfaces. After a period of time, the reliable offloading framework receives communication based on the communication framework, which indicates program instructions for the actuator or controller devices of the autonomous vehicle, as indicated in block 358.

[0100] In block 360, the communication based on the communication framework is then provided to the corresponding actuator, communication framework-specific application, or other controller device to enable the autonomous vehicle to perform computer vision-related operations.

[0101] Figures 4A-5E This section showcases the practicality and versatility of a reliable offloading framework for cloud robotics applications. Figures 4A-4F In this study, the latency reliability of the reliable offloading framework was demonstrated using cloud-based operations in a high-mobility driving setup with two 5G interfaces. Then, in... Figures 5A-5E In this paper, an indoor human tracking environment with Wi-Fi and 5G is discussed. The reliable offloading framework is quantified by long-tailed abnormal latency errors with 99th percentile (P99) latency and operates using the top 1% latency.

[0102] Figures 4A-4C This demonstrates how to maintain high-quality service in driving scenarios involving high mobility. More specifically, autonomous vehicles (such as Waymo, Cruise, and Robotaxi) use cloud-based operations to remotely take control in emergency situations, such as when a car is stuck and onboard algorithms cannot resolve the issue. Such cloud operations may involve remote operators, remote data servers, or more complex models.

[0103] Figure 4A It is a geographic map illustrating an example driving route taken by an autonomous vehicle within the first case study implementation.

[0104] This geographic map depicts a 50-mile drive along a standard U.S. highway from Sunnyvale, CA to Berkeley, CA over a 90-minute period, during which an autonomous vehicle is traveling along the route. The car's speed is maintained by Tesla's Full Autopilot. During the drive, 1280x820 resolution images are streamed as JPEGs from an Intel RealSense 435i camera connected to a laptop computer. The laptop is connected via physical cables to two different 5G hotspots provided by Verizon and AT&T. The cloud operator's decision-making process is simulated using object detection leveraging YOLOv8. End-to-end latency, from the moment an image is captured to the time a response is received, is measured.

[0105] Figure 4B The diagram illustrates along Figure 4A The driving route shown is at one point in time where a handover resulted in a service interruption due to a lack of use of the systems and methods described herein. Figure 4C The illustration shows the path along according to some embodiments. Figure 4A At the same time point of the driving route shown, the handover did not cause service interruption due to the use of the systems and methods described herein.

[0106] Figure 4D illustrates the situation when using a single service provider (e.g., Verizon) and not applying the systems and methods described herein to cross-regional handovers occurring along the driving route. Figure 4A The driving route shown in Figure 4E exhibits multiple service disruptions. Figure 4E illustrates the situation when using a different single service provider (e.g., AT&T) and the systems and methods described herein are not applied to handovers occurring along the driving route. Figure 4A Several other services were disrupted along the driving routes shown.

[0107] Figure 4F The illustration shows, according to some embodiments, along... Figure 4A The reduction in service interruptions along the driving route during the application of the cross-zone handover shown in the document, due to the use of the systems and methods described herein.

[0108] The first case study demonstrates that even when one of the two 5G service providers experiences low coverage and base station handover, the reliable offloading framework effectively delivers smooth quality of service. Figures 4A-4F In the specific experiment shown, the reliable offloading framework improves the abnormally long-tailed P99 (99th percentile) latency by up to 3.7 times.

[0109] In comparison, AT&T has higher mean and P99 latency in some extreme cases. The Reliable Offload Framework effectively utilizes both service providers, offering more reliable latency. The Reliable Offload Framework improves P99 latency by 3.7x (4829.35ms vs. 1303.00ms) for AT&T and 2.4x (3208.04ms vs. 1303.00ms) for Verizon. Compared to AT&T (791.02ms) and Verizon (550.28ms), the Reliable Offload Framework improves mean latency by 289.44ms.

[0110] Figure 5A The illustration shows a second case study implementation of the system and method described herein according to some embodiments (in... Figures 5A-5EAs shown in the figure, the two computing nodes of the cloud server orchestrate computer vision-related tasks for the autonomous robot.

[0111] like Figure 5A As shown, the reliable offloading framework is deployed on a physical Stretch 3 robot 510 and is used for indoor human tracking tasks for a human 512. Robot 510 is configured with a Wi-Fi network interface 506 and a 5G Dongle 508, which are used to send and receive data packets to and from cloud servers 502 and 504. Even when robot 510 is in a fully covered Wi-Fi and 5G environment (such as environment 500), the reliable offloading framework improves the robot's responsiveness by reducing mean latency by 36% and anomalous long-tail latency by 33%.

[0112] More specifically, Robot 510 was configured to run Ubuntu 22.04 and ROS2 Humble. Figure 5A In the given scenario shown, robot 510 is assigned to follow human 512 along an indoor route around the office.

[0113] like Figure 5B , 5C As shown in 5D and 5E, even in environments with full coverage of Wi-Fi and 5G network interfaces, the use of the reliable offloading framework improved the P99 long-tail latency by 33% (160.00ms vs. 193.52ms). The reliable offloading framework consistently improved the average latency by 36% (71.39ms vs. 97.19ms).

[0114] While exemplary embodiments have been described above, they are not intended to describe all possible forms covered by the claims. The terms used in this specification are descriptive and not limiting, and it is to be understood that various changes may be made without departing from the spirit and scope of this disclosure. As previously stated, features of various embodiments may be combined to form further embodiments of the invention that may not be explicitly described or illustrated. While various embodiments may have been described as providing advantages over or being preferred with respect to one or more desired characteristics compared to other embodiments or prior art implementations, those skilled in the art will recognize that one or more features or characteristics may be compromised to achieve desired overall system properties depending on the specific application and implementation. These properties may include, but are not limited to, cost, strength, durability, lifecycle cost, merchantability, appearance, packaging, size, serviceability, weight, manufacturability, ease of assembly, etc. Therefore, while any embodiment may be described as less desirable with respect to one or more characteristics than other embodiments or prior art, these embodiments do not exceed the scope of this disclosure and may be desirable for a particular application.

Claims

1. A computer-implemented method for transmitting communication based on a communication framework via a proxy computing node, the method comprising: The application receives a request from a communication framework located within the sensors of the autonomous vehicle to send data packets to a cloud server located within the autonomous vehicle, wherein the cloud server orchestrates computer vision-related operations of the autonomous vehicle. Based on the optimized combination of network interface and computing node pairing, two or more network interfaces are mapped to corresponding computing nodes of the cloud server, wherein the two or more network interfaces are to be used to send the data packets; The data packet is copied and then transmitted via the two or more network interfaces; Receive communication based on the communication framework for the proxy computing node via one of the two or more network interfaces; and The communication based on the communication framework is provided to the actuators of the autonomous vehicle to perform given computer vision-related operations of the autonomous vehicle.

2. The computer-implemented method according to claim 1, wherein mapping the two or more network interfaces to the corresponding computing nodes of the cloud server comprises: Based on multiple network interfaces configured to connect to multiple computing nodes of the cloud server, for corresponding combinations of network interface and computing node pairings, the probability of missing the delayed deadline is minimized; and Constraint-based optimization selects two or more network interfaces to be mapped.

3. The computer-implemented method according to claim 2, wherein the constraints include: The limitation that one of the plurality of network interfaces can only connect to one of the plurality of computing nodes at a time; as well as The total number of network interface and computing node pairs is less than or equal to the total number of the plurality of computing nodes.

4. The computer-implemented method of claim 1, wherein the data packet includes image-based data samples, audio-based data samples, video-based data samples, or text-based data samples collected from the sensor.

5. The computer-implemented method of claim 1, wherein the request to send the data packet relates to stateless computer vision-related operations of the autonomous vehicle.

6. The computer-implemented method according to claim 1, wherein the two or more network interfaces are fault-independent network interfaces, including one or more of the following: Wi-Fi interface; 5G interface; or Cloud Network Interface Card (NIC).

7. The computer-implemented method of claim 1, wherein copying the data packet and subsequently transmitting the data packet via the two or more network interfaces is handled using a zero-copy property constraint.

8. The computer-implemented method according to claim 1, further comprising: Receive communication based on the communication framework for the proxy computing node via another network interface of the two or more network interfaces; Based on the fact that the communication based on the communication framework has already been received via one of the two or more network interfaces, it is determined that the communication based on the communication framework received via another of the two or more network interfaces is redundant; and Discard the redundant communication based on the communication framework.

9. The computer-implemented method according to claim 1, further comprising: In response to receiving a request to send the data packet, the request is extracted into a serialized request with an identifier; Store the identifier; and In response to receiving communication based on a communication framework for the proxy computing node via one of the two or more network interfaces, the communication based on the communication framework is verified to include an identifier that matches the identifier of the request.

10. The computer-implemented method according to claim 9, further comprising: Before the copying and subsequent transmission, a callback for the timeout is determined; In addition to the identifier, the callback for timeout is also stored; and In response to receiving communication based on a communication framework for the proxy computing node via one of the two or more network interfaces, the verification time elapsed less than the callback for timeout.

11. The computer-implemented method according to claim 1, further comprising: Generate connectivity threads for the two or more network interfaces within the autonomous vehicle; Generate Linux-based sockets for the two or more network interfaces; The STUN protocol, a session traversal utility for network address translation deployed in the cloud, is executed to output cloud-accessible network Internet Protocol (IP) addresses for the two or more network interfaces; and Based on the cloud-accessible network IP address, a connection is established between the two or more network interfaces and the corresponding computing nodes of the cloud server.

12. A computer-implemented method for establishing connectivity between a proxy computing node and a cloud server, the method comprising: A connectivity thread is generated for the corresponding network interface within the autonomous vehicle, the network interface being configured to send and receive data packets to and from a cloud server, wherein the cloud server orchestrates the computer vision operations of the autonomous vehicle. Generate Linux-based sockets for the network interface; The STUN protocol, a session traversal utility for network address translation deployed in the cloud, is executed to output the cloud-accessible network Internet Protocol (IP) address of the corresponding network interface in the network interface. Based on the cloud-accessible network IP address, a connection is established between the corresponding network interface in the network interface and the corresponding computing node of the cloud server. and Based on the established connectivity, communication based on the communication framework is sent and received via an agent computing node located within the autonomous vehicle to perform computer vision-related operations of the autonomous vehicle.

13. The computer-implemented method according to claim 12, further comprising: In response to the failure to output the cloud-accessible network IP address of a given network interface, disconnect the connection thread in the connectivity thread corresponding to the given network interface.

14. The computer-implemented method of claim 12, wherein the network interface is a fault-independent network interface, and the network interface includes one or more of the following: Wi-Fi interface; 5G interface; or Cloud Network Interface Card (NIC).

15. The computer-implemented method of claim 12, wherein sending and receiving communication based on the communication framework comprises: The communication framework, located within the sensors of the autonomous vehicle, is used by a specific application to receive requests from the agent computing node to send data packets to the cloud server. Based on the optimized combination of network interface and computing node pairing, two or more network interfaces are mapped to the corresponding computing nodes of the cloud server, wherein the two or more network interfaces are to be used to send the data packets. The data packet is copied and then transmitted via two or more of the network interfaces. Receive communication based on a communication framework for the proxy computing node via one of two or more of the network interfaces. and The communication based on the communication framework is provided to the actuators of the autonomous vehicle to perform given computer vision-related operations of the autonomous vehicle.

16. The computer-implemented method of claim 15, wherein mapping two or more network interfaces to corresponding computing nodes of the cloud server comprises: Based on multiple network interfaces configured to connect to multiple computing nodes of the cloud server, for corresponding combinations of network interface and computing node pairings, the probability of missing the delayed deadline is minimized; and Constraint-based optimization selects two or more network interfaces to be mapped.

17. A non-transitory computer-readable medium storing program instructions that, when executed on or by a processor, cause the processor to perform the following operations: The application receives a request from a communication framework located within the sensors of the autonomous vehicle to send data packets to a cloud server located within the autonomous vehicle, wherein the cloud server orchestrates computer vision-related operations of the autonomous vehicle. Based on the optimized combination of network interface and computing node pairing, two or more network interfaces are mapped to corresponding computing nodes of the cloud server, wherein the two or more network interfaces are to be used to send the data packets; The data packet is copied and then transmitted via the two or more network interfaces; Receive communication based on the communication framework for the proxy computing node via one of the two or more network interfaces; and The communication based on the communication framework is provided to the actuators of the autonomous vehicle to perform given computer vision-related operations of the autonomous vehicle.

18. The non-transitory computer-readable medium of claim 17, wherein the program instructions further cause the processor to: Receive communication based on the communication framework for the proxy computing node via another network interface of the two or more network interfaces; Based on the fact that the communication based on the communication framework has already been received via one of the two or more network interfaces, it is determined that the communication based on the communication framework received via another of the two or more network interfaces is redundant; and This causes the redundant communication based on the communication framework to be discarded.

19. The non-transitory computer-readable medium of claim 17, wherein the program instructions further cause the processor to: In response to receiving a request to send the data packet, the request is extracted into a serialized request with an identifier; Store the identifier; and In response to receiving communication based on a communication framework for the proxy computing node via one of the two or more network interfaces, the communication based on the communication framework is verified to include an identifier that matches the identifier of the request.

20. The non-transitory computer-readable medium of claim 19, wherein the program instructions further cause the processor to perform the following operations: Before the copying and subsequent transmission, a callback for the timeout is determined; In addition to the identifier, the callback for the timeout is also stored; and In response to receiving communication based on a communication framework for the proxy computing node via one of the two or more network interfaces, the verification time elapsed less than the callback for timeout.