Systems and methods for data security in autonomous vehicles

By adopting a group signature scheme in autonomous vehicles, privacy and security issues in collaborative machine learning networks are solved, system accuracy and privacy are improved, and data security and model efficient updates are ensured.

CN115812315BActive Publication Date: 2024-12-03HARMAN INT IND INC
View PDF 11 Cites 0 Cited by

Patent Information

Application Number
CN202080102717.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-07-21
Filing Date
2020-09-15
Publication Date
2024-12-03
Estimated Expiration
2040-09-15

AI Technical Summary

Technical Problem

In autonomous vehicles, privacy issues and security threats exist in collaborative machine learning networks, and unauthorized entities may access sensitive data or provide incorrect data, affecting the accuracy of machine learning models and the privacy of the system.

Method used

Using the group signature scheme, the group public key and private key are generated through the group manager. The vehicle uses the private key to digitally sign the message and verifies the signature through the group public key to ensure the anonymity and security of the message, and at the same time, identify and remove non-existent Authorized vehicles, thereby improving the privacy and security of the system.

Benefits of technology

It improves the accuracy of autonomous vehicle perception, path planning and control systems, enhances the privacy and security of the system, prevents unauthorized vehicles from participating in collaborative learning, and ensures the correctness of data and efficient updates of machine learning models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115812315B_ABST
    Figure CN115812315B_ABST
Patent Text Reader

Abstract

An implementation scheme for a group security scheme for autonomous vehicles participating in a collaborative machine learning method is disclosed. For example, a method includes: generating a digital signature in a vehicle based on a first key and a second key, both of which are received from a group manager; and transmitting a message signed with the digital signature to a collaborator, the message including coefficients of a local machine learning model of the vehicle. In this way, the accuracy of the local machine learning model of the vehicle can be improved while improving the privacy of the vehicle.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to Indian Patent Application No. 202011031177, filed on July 21, 2020, entitled “SYSTEMS AND METHODS FOR DATA SECURITY IN AUTONOMOUS VEHICLES”. The entire contents of the above application are hereby incorporated herein by reference for all purposes. Technical Field

[0003] The present disclosure relates to the field of autonomous (eg, self-driving) vehicles. Background Art

[0004] Some motor vehicles may include systems that operate autonomously (e.g., self-driving) and / or partially autonomously. For example, a vehicle may operate without user input (e.g., a fully autonomous vehicle) or with limited user input (e.g., a partially autonomous vehicle). For example, the vehicle's autonomous driving system may at least partially control the vehicle's acceleration, braking, and steering systems to drive the vehicle. For example, a vehicle user may select a desired destination, and the vehicle's autonomous driving system may autonomously control acceleration, braking, and steering to guide the vehicle to the desired destination. In some examples, the autonomous driving system includes at least one machine learning model for controlling vehicle systems and monitoring the environment around the vehicle. For example, the vehicle's machine learning model may monitor data from multiple sensors, such as by detecting pedestrians and construction zones, in order to visualize and sense the surrounding environment. As another example, the vehicle's machine learning model may adjust braking and acceleration based on traffic patterns. The vehicle's machine learning model may be pre-trained using training data from various driving situations, and may be further trained in real time during vehicle operation. In addition, in order to improve the accuracy of the machine learning model, some autonomous vehicles may collaborate with other autonomous vehicles, such as by sharing local data with a remote server and receiving an updated machine learning model. This technology may be referred to as collaborative machine learning. Sharing vehicle data and machine learning models among multiple autonomous vehicles through collaborative machine learning can increase the intelligence of machine learning models by transferring learned knowledge from a local machine learning model in one vehicle to another machine learning model on a distributed network.

[0005] However, in some examples, such sharing may lead to privacy issues. For example, transmitted vehicle data that may include sensitive user information may be accessed by unauthorized entities. As another example, hostile or unauthorized users may provide counterproductive data, which may affect machine learning training. For example, by providing erroneous data, the accuracy of the machine learning model provided to the vehicle participating in the collaborative machine learning may be reduced. In addition, such privacy issues may hinder the adaptability of collaborative learning techniques. For example, due to privacy concerns, vehicle owners or users may be reluctant to participate in collaborative learning, which may reduce the accuracy of their vehicle's local machine learning model. Summary of the invention

[0006] The inventors herein have recognized the aforementioned problems and have developed systems and methods for at least partially solving the above problems. For example, a method includes: generating a digital signature in a vehicle based on a first key and a second key, both of which are received from a group manager; and transmitting a message signed with the digital signature to a collaborator, the message including coefficients and parameters of a local machine learning model of the vehicle.

[0007] In this way, the accuracy of the vehicle's autonomous vehicle perception, path planning, and control systems can be improved while reducing the occurrence of privacy issues. For example, by implementing a collaborative machine learning strategy, multiple vehicles can provide local machine learning models to a cloud server, and the cloud server can build an updated global machine learning model based on the local machine learning models from multiple vehicles. For example, because the updated global machine learning model combines data from multiple vehicles, the updated global machine learning model may be more accurate than each of the local machine learning models. The cloud server can distribute the updated global model to multiple vehicles so that the vehicle group can implement the updated global model. In addition, by managing the group via a group signature scheme, each message (e.g., a message including a local machine learning model) can include a digital signature based on a group public key and a vehicle secret key, which can increase system security, increase system privacy, and allow unauthorized vehicles to be removed from the vehicle group. Overall, system privacy and accuracy can be improved, which in turn can improve customer satisfaction with autonomous vehicle perception, path planning, and control systems.

[0008] The above advantages and other advantages and features of the present specification will be apparent from the following detailed description taken alone or in conjunction with the accompanying drawings.

[0009] It should be understood that the above summary is provided to introduce a series of concepts further described in the detailed description in a simplified form. It is not meant to identify the key or essential features of the claimed subject matter, the scope of which is clearly defined by the claims following the detailed description. In addition, the claimed subject matter is not limited to implementations that solve any disadvantages noted above or in any part of this disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The present disclosure may be better understood by reading the following description of non-limiting embodiments with reference to the accompanying drawings, in which:

[0011] Figure 1 shows an exemplary partial view of a vehicle cabin according to one or more embodiments of the present disclosure;

[0012] Figure 2 A block diagram illustrating an exemplary onboard computing system of a vehicle according to one or more embodiments of the present disclosure is shown;

[0013] Figure 3 An exemplary autonomous vehicle system in communication with a remote server is shown.

[0014] Figure 4 An exemplary system for an autonomous vehicle using continuous machine learning is shown.

[0015] Figure 5 An exemplary system for an autonomous vehicle using continuous machine learning including an adversarial actor is shown.

[0016] Figure 6 A table including a number of symbols for components of a group signature scheme is shown.

[0017] Figure 7 An exemplary system for an autonomous vehicle using continuous machine learning and a group signature scheme is shown.

[0018] Figure 8 A flow chart of a method for establishing and managing collaborative learning for autonomous vehicles using a group signature scheme is shown.

[0019] Fig. 9 A flow chart of a method for operating an autonomous vehicle using collaborative learning and a group signature scheme is shown. DETAILED DESCRIPTION

[0020] As described above, autonomous vehicles can collaboratively share machine learning models to improve the accuracy of machine learning models, and group signature schemes can be employed to maintain data security. Figure 1 The vehicle shown in FIG. 1 may include an autonomous mode for autonomously operating the vehicle. The autonomous mode may operate through autonomous vehicle perception, path planning, and control systems, such as Figure 2 To improve the accuracy of the machine learning models for autonomous vehicle perception, path planning, and control systems, the vehicle can provide local vehicle data to a remote server and receive updated machine learning models, such as Figure 3 In addition, Figure 4 In the exemplary system shown, multiple autonomous vehicles can participate in a collaborative machine learning network, where multiple autonomous vehicles can upload local machine models to a remote server and can receive updated models from the remote server. However, such collaborative machine learning networks may allow hostile participants to adversely affect the system, such as Figure 5 To reduce the occurrence of privacy and / or security issues, a group signature scheme can be provided, such as Figure 6 and Figure 7 For example, according to Figure 8 In this paper, we propose a method to apply a group signature scheme to groups of vehicles using collaborative machine learning. Fig. 9 A method is shown for individual vehicles to participate in a collaborative machine learning approach using a group signature scheme.

[0021] Figure 1 The interior of a vehicle cabin 100 of a vehicle 102 in which a driver and / or one or more passengers may be seated is shown. The vehicle 102 may be a road car, among other types of vehicles. Specifically, Figure 1 The vehicle 102 may be a motor vehicle including drive wheels (not shown) and an engine 104. In some examples, the engine 104 may be an internal combustion engine. In other examples, the engine 104 may be an electric engine, or may include a hybrid power assembly. For example, the vehicle 102 may include a hybrid propulsion system including an energy conversion device operable to absorb energy from vehicle motion and / or the engine, and convert the absorbed energy into an energy form suitable for storage by an energy storage device. The vehicle 102 may include an all-electric vehicle incorporating a fuel cell, a solar energy capture element, and / or other energy storage system for powering the vehicle.

[0022] In addition, vehicle 102 can be an autonomous vehicle. In some examples, vehicle 102 is a fully autonomous vehicle (e.g., a fully autonomous vehicle) that is configured to drive without user input. For example, vehicle 102 can independently control vehicle systems to guide the vehicle to a desired location, and can sense environmental features to guide the vehicle. In some examples, vehicle 102 is a partially autonomous vehicle. In some examples, vehicle 102 can have an autonomous mode and a non-autonomous mode, in which the vehicle operates without user input and in which the user guides the vehicle. In addition, in some examples, although the autonomous vehicle perception, path planning, and control system can control the vehicle primarily in an autonomous mode, the user can input commands to adjust the vehicle operation, such as a command to change the vehicle speed, a command to brake, a command to turn, and the like.

[0023] For example, the vehicle 102 may include multiple vehicle systems, including a braking system for providing braking, an engine system for providing power to the vehicle's wheels, a steering system for adjusting the direction of the vehicle, a transmission system for controlling the engine's gear selection, an exhaust system for treating exhaust gases, etc. In addition, the vehicle 102 includes an onboard computing system 109. The onboard computing system 109 includes an autonomous vehicle perception, path planning, and control system for at least partially controlling the vehicle systems during autonomous driving. For example, when operating in an autonomous mode, the autonomous vehicle perception, path planning, and control system can monitor the vehicle's surroundings via multiple sensors (e.g., cameras, motion sensors, GPS signals, etc.). Based on the desired destination and the vehicle's surroundings, the autonomous vehicle perception, path planning, and control system can adjust multiple vehicle systems to navigate the vehicle to the destination, for example by adjusting steering, braking, and speed. Reference will be made to Figure 2 Describe autonomous vehicle perception, path planning, and control systems in more detail (e.g., Figure 2 Autonomous vehicle perception, path planning and control systems 261).

[0024] As shown, the dashboard 106 may include various displays and controls accessible to human users (also referred to as passengers) of the vehicle 102. For example, the dashboard 106 may include user input devices, such as a touch screen 108 of an onboard computing system 109, an audio system control panel, and an instrument cluster 110. The touch screen 108 may receive user input to the onboard computing system 109 for controlling audio output, visual display output, user preferences, control parameter selection, and the like. In some examples, the dashboard 106 may include input devices for users to switch the vehicle between autonomous mode and non-autonomous mode. For example, the vehicle includes an autonomous mode in which an autonomous vehicle perception, path planning, and control system at least partially independently operate the vehicle, and a non-autonomous mode in which a vehicle user operates the vehicle. The vehicle user may switch between the two modes via user input to the dashboard 106. In addition, in some examples, the dashboard 106 may include one or more controls for autonomous vehicle perception, path planning, and control systems, such as for selecting a destination, setting a desired vehicle speed, setting navigation preferences (e.g., a preference for highways over city streets), and the like. In addition, additional user interfaces not shown may exist in other portions of the vehicle, such as near at least one passenger seat. For example, the vehicle may include a row of rear seats with at least one touch screen that controls the onboard computing system 109 .

[0025] Figure 2 A block diagram of an onboard computing system 109 configured and / or integrated within a vehicle 102 is shown. The onboard computing system 109 may perform one or more of the methods described herein in some embodiments. The onboard computing system may include or be coupled to various vehicle systems, subsystems, hardware components, and software applications and systems that are integrated or can be integrated into the vehicle 102 to enhance the in-vehicle experience of the driver and / or passengers. In addition, the onboard computing system may be coupled to a system for providing autonomous vehicle control. For example, the onboard computing system includes an autonomous vehicle perception, path planning, and control system 261 for autonomously controlling the vehicle.

[0026] The vehicle computing system 109 may include one or more processors, including an operating system processor 214 and an interface processor 220. The operating system processor 214 may execute an operating system on the vehicle computing system and control input / output, display, playback, and other operations of the vehicle computing system. The interface processor 220 may interface with a vehicle control system 230 via an inter-vehicle system communication module 222.

[0027] The inter-vehicle system communication module 222 may output data to other vehicle subsystems 231 and the autonomous vehicle perception, path planning and control system 261, while also receiving data input from other vehicle components and systems 231, 261, for example, through the vehicle control system 230. When outputting data, the inter-vehicle system communication module 222 may provide a signal corresponding to the output of any state of the vehicle, the vehicle surroundings, or any other information source connected to the vehicle via the bus. The vehicle data output may include, for example, an analog signal (e.g., current speed), a digital signal provided by a separate information source (e.g., a clock, a thermometer, a position sensor such as a global positioning system [GPS] sensor, etc.), a digital signal propagated through a vehicle data network (e.g., an engine controller area network [CAN] bus through which engine-related information may be transmitted, a climate control CAN bus through which climate control-related information may be transmitted, and a multimedia data network through which multimedia data may be transmitted between multimedia components in the vehicle). For example, the vehicle data output may be output to the autonomous vehicle perception, path planning and control system 261, and the autonomous vehicle perception, path planning and control system 261 may adjust the vehicle subsystem 231 based on the vehicle data output. For example, the onboard computing system 109 may retrieve from the engine CAN bus the current speed of the vehicle estimated by the wheel sensors, the power status of the vehicle via the vehicle's battery and / or power distribution system, the ignition status of the vehicle, etc. In addition, other interfacing methods such as Ethernet may also be used without departing from the scope of the present disclosure.

[0028] A storage device 208 may be included in the vehicle computing system 109 to store data such as instructions that can be executed by the processors 214 and 220 in a non-volatile form. The storage device 208 may store application data, including pre-recorded sounds, to enable the vehicle computing system 109 to run applications to connect to a cloud-based server and / or collect information for transmission to a cloud-based server. The application may retrieve information collected by vehicle systems / sensors, input devices (e.g., user interface 218), data stored in a volatile storage device 219A or a non-volatile storage device (e.g., memory) 219B, devices that communicate with the vehicle computing system (e.g., a mobile device connected via a Bluetooth link), etc. The vehicle computing system 109 may also include a volatile memory 219A. The volatile memory 219A may be a random access memory (RAM). Non-transitory storage devices, such as non-volatile storage device 208 and / or non-volatile memory 219B, may store instructions and / or code that, when executed by a processor (e.g., operating system processor 214 and / or interface processor 220), control the onboard computing system 109 to perform one or more of the actions described in the present disclosure.

[0029] One or more additional sensors may be included in the sensor subsystem 210 of the onboard computing system 109. For example, the sensor subsystem 210 may include a plurality of cameras, such as a rear-view camera for assisting a user in parking the vehicle and / or a cabin camera for identifying a user (e.g., using facial recognition and / or user gestures). The plurality of cameras may also include at least one camera for environmental sensing during autonomous operation, such as a camera for detecting upcoming road obstacles, a camera for detecting lane changes, etc. The sensor subsystem 210 of the onboard computing system 109 may communicate with and receive input from various vehicle sensors and may further receive user input. For example, inputs received by the sensor subsystem 210 may include transmission gear position, transmission clutch position, accelerator pedal input, brake input, transmission selector position, vehicle speed, engine speed, mass air flow through the engine, ambient temperature, intake air temperature, etc., as well as inputs from climate control system sensors (e.g., heat transfer fluid temperature, antifreeze temperature, fan speed, passenger compartment temperature, desired passenger compartment temperature, ambient humidity, etc.), audio sensors that detect voice commands issued by a user, and key fob sensors that receive commands from a key fob of the vehicle and optionally track the geographic location / proximity of the key fob of the vehicle. For example, the autonomous vehicle perception, path planning, and control system 261 may use data from the sensor subsystem 210 in order to sense the vehicle environment and control vehicle systems. While some vehicle system sensors may communicate with the sensor subsystem 210 alone, other sensors may communicate with both the sensor subsystem 210 and the vehicle control system 230, or may communicate with the sensor subsystem 210 indirectly through the vehicle control system 230.

[0030] The navigation subsystem 211 of the onboard computing system 109 may generate and / or receive navigation information, such as location information (e.g., via a GPS sensor and / or other sensors from the sensor subsystem 210), route guidance, traffic information, point of interest (POI) identification, and / or provide other navigation services to the user. The navigation subsystem 211 may include input / output 280, including analog-to-digital converters, digital inputs, digital outputs, network outputs, radio frequency transmission devices, etc. In some examples, the navigation subsystem 211 may interface with a vehicle control system 230, such as an autonomous vehicle perception, path planning, and control system 261. For example, the navigation subsystem 211 may output map data, traffic data, suggested routes, etc. to the autonomous vehicle perception, path planning, and control system 261 to facilitate autonomous vehicle operation.

[0031] The external device interface 212 of the vehicle-mounted computing system 109 may be coupled to and / or communicate with one or more external devices 150 located outside the vehicle 102. Although the external devices are shown as being located outside the vehicle 102, it will be understood that they may be temporarily accommodated in the vehicle 102, for example, when a user is operating the external devices while operating the vehicle 102. In other words, the external devices 150 are not integrated with the vehicle 102. The external devices 150 may include a mobile device 128 (e.g., connected via Bluetooth, NFC, WIFI direct or other wireless connection 130) or an alternative device 252 with Bluetooth functionality. The mobile device 128 may be a mobile phone, a smart phone, a wearable device / sensor, or other portable electronic device that can communicate with the vehicle-mounted computing system via wired and / or wireless communications. Other external devices include external services 246. For example, the external device may include an outboard device that is separate from the vehicle and located outside the vehicle. Other external devices include external storage devices 254, such as solid-state drives, pen drives, USB drives, etc. Without departing from the scope of the present disclosure, the external device 150 may communicate with the onboard computing system 109 wirelessly or via a connector. For example, the external device 150 may communicate with the onboard computing system 109 through the external device interface 212 over the network 260, a universal serial bus (USB) connection, a direct wired connection, a direct wireless connection, and / or other communication links.

[0032] The external device interface 212 may provide a communication interface to enable the vehicle computing system to communicate with a mobile device associated with a user's contacts. For example, the external device interface 212 may enable a voice call to be established with a mobile device 128 associated with a user's contacts and / or a text message (e.g., SMS, MMS, etc.) to be sent to the mobile device (e.g., via a cellular communication network). In addition, in some examples, a vehicle user may adjust autonomous vehicle operations via an application on the mobile device 128 associated with the user. The external device interface 212 may additionally or alternatively provide a wireless communication interface to enable the vehicle computing system to synchronize data with one or more devices in the vehicle (e.g., a user's mobile device) via a WIFI direct connection.

[0033] One or more applications 248 may operate on external services 246. For example, external service applications 248 may be operated to aggregate and / or analyze data from multiple data sources. For example, external service applications 248 may aggregate data from one or more social media accounts of a user, data from an onboard computing system (e.g., sensor data, log files, user input, etc.), data from Internet queries (e.g., weather data, POI data), etc. The collected data may be transmitted to another device and / or the collected data may be analyzed by the application to determine the context of the user, vehicle, and environment and perform actions based on the context (e.g., request data from / send data to other devices).

[0034] The onboard computing system 109 may also include an antenna 206. Antenna 206 is shown as a single antenna, but may include one or more antennas in some embodiments. The onboard computing system may obtain broadband wireless Internet access via antenna 206, and may also receive broadcast signals such as radio, television, weather, traffic, etc. The onboard computing system may receive positioning signals such as GPS signals via one or more antennas 206. The onboard computing system may also receive wireless commands via FR, such as via antenna 206 or via infrared or other means through an appropriate receiving device. For example, antenna 206 may receive voice calls (e.g., telephone calls). In addition, antenna 206 may provide AM / FM radio signals to external devices 150 (e.g., to mobile device 128) via external device interface 212.

[0035] The vehicle control system 230 may include controls for controlling various aspects of the various vehicle subsystems 231 involved in different vehicle functions. For example, multiple vehicle subsystems 231 may be controlled in order to operate the vehicle. In some examples, the vehicle subsystems 231 may be at least partially controlled by user input, while in some examples, the vehicle subsystems 231 may be at least partially controlled via the autonomous vehicle perception, path planning, and control system 261. Figure 2 A number of vehicle subsystems 231 are shown, including a braking system 232, an acceleration control system 234, a telecommunications control system 236, a steering control system 238, a climate control system 240, an exterior lighting control system 242, and an audio system 244. In some examples, additional vehicle systems may be included in the vehicle subsystems 231. The additional vehicle systems may include a cruise control system, an object detection system, a parking system, a lighting system, and the like.

[0036] Braking system 232 may be configured to control the amount of braking force applied to the vehicle. For example, during a non-autonomous operating mode, braking system 232 may be controlled by a brake pedal. For example, a user may depress the brake pedal to increase the amount of braking applied to the vehicle. During an autonomous operating mode, braking system 232 may be controlled by autonomous vehicle perception, path planning, and control system 261. For example, autonomous vehicle perception, path planning, and control system 261 may determine that additional braking is requested, and may apply additional braking. In some examples, autonomous vehicle perception, path planning, and control system may depress the brake pedal to apply braking (e.g., reduce vehicle speed and / or stop the vehicle).

[0037] The acceleration control system 234 may be configured to control the amount of acceleration applied to the vehicle. For example, during a non-autonomous operating mode, the acceleration control system 234 may be controlled by an accelerator pedal. For example, a user may depress the accelerator pedal to increase the amount of torque applied to the vehicle's wheels, thereby accelerating the vehicle. During an autonomous operating mode, the acceleration control system 234 may be controlled by the autonomous vehicle perception, path planning, and control system 261. For example, the autonomous vehicle perception, path planning, and control system 261 may determine that additional vehicle speed is requested and may increase the vehicle speed by accelerating. In some examples, the autonomous vehicle perception, path planning, and control system 261 may depress the accelerator pedal to accelerate the vehicle.

[0038] The steering control system 238 may be configured to control the direction of the vehicle. For example, during a non-autonomous operating mode, the steering control system 238 may be controlled by a steering wheel. For example, a user may turn the steering wheel to adjust the direction of the vehicle. During an autonomous operating mode, the steering control system 238 may be controlled by the autonomous vehicle perception, path planning, and control system 261. For example, the steering control system 261 may determine that a change in vehicle direction is requested, and may change the vehicle direction by controlling the steering control system 238. For example, the steering control system 238 may adjust an axle of the vehicle to change the vehicle direction.

[0039] The climate control system 240 may be configured to provide a comfortable environment within the cabin or passenger compartment of the vehicle 102. The climate control system 240 includes components that implement controlled ventilation, such as vents, heaters, air conditioners, integrated heater and air conditioning systems, etc. Other components connected to the heating and air conditioning equipment may include a windshield defrosting and defogging system that can clean the windshield, and a ventilation air filter for cleaning the outside air entering the passenger compartment through the fresh air inlet. In some examples, the components of the climate control system 240 may be controlled by the autonomous vehicle perception, path planning and control system 261. However, the user can adjust the climate control system 240 via user input devices such as the touch screen 108 and other buttons in the vehicle cabin. For example, although the autonomous vehicle perception, path planning and control system 261 can maintain the temperature in the vehicle cabin at a predetermined temperature, the user can adjust the climate control system to adjust the temperature.

[0040] The vehicle control system 230 may also include controls for adjusting settings of various vehicle controls (or vehicle system control elements) associated with the engine and / or auxiliary elements within the vehicle cabin, such as steering wheel controls (e.g., steering wheel mounted audio system controls, cruise controls, windshield wiper controls, headlight controls, turn signal controls, etc.); instrument panel controls; microphone; shifter; door / window controls located in the driver's door or passenger door; seat controls; cabin light controls, etc. The vehicle controls may also include internal engine and vehicle operation controls (e.g., engine controller module, actuators, valves, etc.) configured to receive instructions via the vehicle's CAN bus to change the operation of one or more of the engine, exhaust system, transmission, and / or other vehicle systems. In addition, to autonomously control the vehicle, the object detection system may use sensor data, such as from cameras and motion sensors, to detect obstacles in the vehicle's path and adjust vehicle operation based on the obstacles. The control signals may also control a vehicle audio system (not shown). For example, the control signals may adjust audio output characteristics such as volume, balance, audio image (e.g., an audio signal configuration for producing an audio output that appears to a user to originate from one or more defined locations), audio distribution between multiple speakers, etc. Similarly, the control signals may control vents, air conditioning, and / or heaters of the climate control system 240. For example, the control signal may increase the delivery of cooling air to a particular section of the vehicle cabin. For example, when operating in autonomous mode, the autonomous vehicle perception, path planning, and control system 261 may control some or all of the above-described vehicle controls.

[0041] Control elements located outside the vehicle (e.g., controls for safety systems) may also be connected to the computing system 109, for example, via the communication module 222. The control elements of the vehicle control system may be physically and permanently located on and / or in the vehicle for receiving user input. In addition to receiving control instructions from the onboard computing system 109, the vehicle control system 230 may also receive input from one or more external devices 150 operated by a user, such as from a mobile device 128. This allows aspects of the vehicle subsystems 231 and vehicle controls 261 to be controlled based on user input received from the external devices 150.

[0042] In order to control the vehicle subsystem 231, the autonomous vehicle perception, path planning and control system 261 includes multiple local machine learning models 262. When operating in autonomous mode, the autonomous vehicle perception, path planning and control system 261 can combine various data from various sources (e.g., from the sensor subsystem 210, the navigation subsystem 211, etc.) to visualize the environment and predict future events. For example, the autonomous vehicle perception, path planning and control system 261 includes a perception system that can monitor the vehicle's surrounding environment based on sensor data, a high-level path planning system, a low-level path planning, and a vehicle controller. Multiple local machine learning models 262 can provide models for controlling vehicle systems based on various data. In addition, multiple local machine learning models 262 can be trained during vehicle operation. For example, the autonomous vehicle perception, path planning and control system 261 can use machine learning to improve the accuracy of the control method of each of the vehicle subsystems 231. For example, the local machine learning model 262 can be trained on vehicle data to improve the accuracy of the local machine learning model. The local machine learning model can be updated when operating in autonomous mode and non-autonomous mode. In addition, in order to simplify the learning process, a separate machine learning model can be developed for each vehicle system of the vehicle subsystem 231. For example, a first local machine learning model may be trained on data from braking system 232 , while a second machine learning model may be trained on data from acceleration control system 234 .

[0043] In addition, the external device interface 212 can provide a communication interface with the autonomous vehicle cloud 256. For example, the autonomous vehicle perception, path planning and control system 261 can exchange data with the autonomous vehicle cloud 256 to improve the accuracy and reliability of autonomous operation. For example, the vehicle can participate in collaborative learning, for example, by sending local data and / or local machine learning models to the autonomous vehicle cloud 256, which can provide updated machine learning models based on data from multiple vehicles. In order to maintain safety in such a system, the vehicle 102 may include safety scheme data 250 for authentication, as will also be referenced below Figures 6 to 8For example, the autonomous vehicle cloud 256 may include a plurality of controllers, each controller including a non-volatile memory for storing executable instructions. Figure 7 As shown, autonomous vehicle cloud 256 may include a first controller that acts as a group manager, a second controller that acts as a coordinator, and a third controller that acts as a revocation mechanism.

[0044] In previous machine learning approaches, vehicle data is shared with a cloud server, and the cloud server provides updated machine learning models to the vehicle to improve the accuracy of the local machine learning model. Examples of vehicles using this approach are Figure 3 Specifically, Figure 3 A network 300 is shown including a vehicle 308 providing local vehicle data 304 to a machine learning cloud 302. The vehicle 308 may be an autonomous vehicle (e.g., a fully autonomous vehicle or a partially autonomous vehicle) including a plurality of local machine learning models, each of which is trained on the vehicle data 304 to automatically operate the vehicle 308. The vehicle 308 may periodically and / or continuously upload the vehicle data 304 to the machine learning cloud 302. For example, the machine learning cloud 302 may include a global machine learning model that may be trained based on the vehicle data 304 from the vehicle 308. The machine learning cloud 302 may provide an updated local machine learning model 306 to the vehicle 308 based on the updated global machine learning model to improve the accuracy of the operation of the vehicle 308. This process may continue indefinitely. For example, the vehicle may continue to upload vehicle data to the machine learning cloud 302, the machine learning cloud 302 may continue to update the global machine learning model based on the data from the vehicle, and the machine learning cloud 302 may continue to transmit the updated local machine learning model to the vehicle 308. Thus, the local machine learning model used to control the vehicle 308 can be adjusted over time, thereby improving vehicle accuracy and increasing customer satisfaction.

[0045] However, Figure 3The machine learning system described in may include training data from a single vehicle. In order to further improve the accuracy of autonomous vehicle perception, path planning, and control systems, collaborative machine learning (CML) methods can be applied to autonomous vehicles (e.g., vehicle 102) to continuously update local machine learning models. Using the CML method, multiple local machine learning models from multiple vehicles can be uploaded to a machine learning cloud. For example, the machine learning cloud can be an autonomous vehicle cloud 256. Based on multiple local machine learning models, the machine learning cloud can build a global machine learning model, which can be transmitted to each of the multiple vehicles. For example, since the global machine learning model is based on multiple local machine learning models, it can contain more data and be more accurate. The machine learning cloud can transmit the global machine learning model to each of the multiple vehicles. In some examples, each vehicle can have multiple local machine learning models, each of which is dedicated to a vehicle system. For example, a vehicle may have a local machine learning model for a steering system and a local machine learning model for a braking system. Such a CML method can be applied to each local machine learning model.

[0046] Figure 4 An example system 400 for applying the CML method to an autonomous vehicle is shown. Specifically, Figure 4 Two vehicles are shown participating in the CML method, a first vehicle 410 and a second vehicle 412. Each of the first vehicle 410 and the second vehicle 412 may be associated with Figure 1 and Figure 2 Specifically, each of the first vehicle 410 and the second vehicle 412 may be at least partially autonomous vehicles, and each of the first vehicle 410 and the second vehicle 412 may include an autonomous vehicle perception, path planning, and control system (e.g., autonomous vehicle perception, path planning, and control system 261) having at least one local machine learning model. For example, as described with respect to Figure 2As detailed, the autonomous vehicle perception, path planning and control system may include multiple local machine learning models, each of which controls a vehicle system. For example, a first local machine learning model may control a braking system, and a second local machine learning model may control an acceleration system. As shown, a first vehicle 410 may upload a local machine learning model 404 from the first vehicle 410 to a cloud server 402, and a second vehicle 412 may upload a local machine learning model 408 from the second vehicle 412 to the cloud server 402. For example, each of the first vehicle 410 and the second vehicle 412 may upload coefficients from a trained neural network. Each of the first vehicle 410 and the second vehicle 412 may upload more than one local machine learning model to the cloud server 402. For example, each of the first vehicle 410 and the second vehicle 412 may upload multiple local machine learning models corresponding to multiple vehicle systems (e.g., a steering system, a braking system, etc.). The cloud server 402 may analyze the machine learning models uploaded from the vehicles, and may build updated machine learning models to share via the collaborative learning module 406. For example, based on the coefficients from each of the local machine learning models, the cloud server 402 can determine an updated global machine learning model for each vehicle system, which can incorporate the learning from each of the local machine learning models. In this way, data from each of the first vehicle 410 and the second vehicle 412 can be combined into the global machine learning model. For example, the updated global machine learning model can be shared with each of the first vehicle 410 and the second vehicle 412, such that the updated global machine learning model replaces the local machine learning model. As vehicle operation continues, each of the first vehicle 410 and the second vehicle 412 can continue to train and update their local machine learning models so that the process continues, thereby providing continuous collaborative updates to the machine learning models for each vehicle. For example, applying the CML method to an autonomous vehicle can improve the accuracy of the vehicle's local machine learning model. A system including the cloud server 402 and the vehicles 410 and 412 participating in the CML method Figure 4 The system can be called a CML network.

[0047] However, privacy and security issues may arise when providing vehicle data (e.g., coefficients from local machine learning models). For example, CML networks (e.g. Figure 4Unauthorized participants in a CML network (shown in ) may access vehicle-related data without permission. As another example, unauthorized participants in a CML network may provide false vehicle data to a cloud server, resulting in a decrease in the accuracy of a global machine learning model. For example, by providing inaccurate vehicle data to a cloud server, the accuracy of a global machine learning model may be reduced. In addition, a global machine learning model trained on inaccurate vehicle data may be transmitted to vehicles in the CML network, thereby reducing the accuracy of the autonomous vehicle perception, path planning, and control systems of the vehicles in the CML network. Overall, privacy issues and security issues may reduce customer satisfaction and reduce the use of autonomous vehicle perception, path planning, and control systems.

[0048] To demonstrate the security and privacy issues associated with CML networks, Figure 5 An exemplary CML network 500 including unauthorized vehicles is shown. For example, Figure 5 4, a cloud server 402, a first vehicle 410, and a second vehicle 412 are shown, and the first vehicle 410 and the second vehicle 412 participate in the CML method. Figure 5 As shown, unauthorized vehicles 502 may participate in the network. For example, unauthorized vehicles 502 may provide incorrect or biased data to cloud server 402, which may reduce the accuracy of the global machine learning model. Figure 5 As shown, the unauthorized vehicle 502 provides inaccurate data to the cloud server 402, which affects the updated global model distributed to the first vehicle 410 and the second vehicle. As another example, the unauthorized vehicle 502 can access data about other vehicles in the network (e.g., data from the first vehicle 410 and the second vehicle 412), which may reduce privacy in the system. For example, the reduction in global model accuracy and the reduction in privacy may lead to decreased customer satisfaction. Therefore, there is a need for methods to increase security and privacy in a CML network, such as by identifying unauthorized vehicles and removing them from the CML network.

[0049] The inventors herein have recognized the above-mentioned deficiencies and have advantageously recognized that a group signature scheme can be applied to a CML network for autonomous vehicles in order to identify and remove unauthorized vehicles. A group signature scheme is an advanced form of digital signature that provides enhanced privacy and anonymity capabilities. For example, a group may include multiple vehicles (e.g., group members), a group manager, a collaborator, and a revocation mechanism. The group manager, the collaborator, and the revocation mechanism may each be an independent controller housed in a location remote from each of the multiple vehicles. In some examples, the group manager, the collaborator, and the revocation mechanism may be included in a single remote server. The group signature scheme allows a group member (e.g., a vehicle in a vehicle group) to anonymously sign a message with a signature on behalf of the group using a private key (e.g., data transmission to a collaborator, such as coefficients of a local machine learning model). For example, a signature can be verified relative to a single group public key, but the identity of the message signer may not be disclosed. Therefore, a valid group signature is publicly verifiable, so that anyone with the group public key can verify the group signature without compromising the privacy of the message signer. For example, a collaborator can determine the identity of the group based on a given message, but may not be able to determine the identity of the individual who signed the message. Furthermore, linking digital signatures to group members may be computationally challenging, which may prevent unauthorized participants in the network from identifying group members (e.g., identifying another vehicle in the group). Furthermore, valid group signatures may not be misattributed. In the event of a dispute (e.g., a suspected unauthorized participant in a signature-based network), the group manager may open the message to track the identity of the signer. Thus, the unauthorized participant may be identified by the revocation authority and removed from the group.

[0050] Therefore, the proposed system in which the CML network includes a group signature scheme allows data to be shared between multiple vehicles while maintaining the privacy and security of multiple vehicles. The proposed method can be used to share models between ADAS, digital cockpits, and V2X applications. For example, the proposed system implements anonymous sharing of local machine learning models while ensuring model correctness. The group signature scheme described herein may be a Boneh, Boyen, and Shacham (BBS) group signature scheme. However, Atenese, Camenisch, Joye, and Tsudik (ACJT) group signature scheme, Camenisch and Groth (CG) group signature scheme, and Boneh Shacham (BS) group signature scheme may be used without departing from the scope of the present disclosure. The BBS group signature scheme may not support dynamic registration of vehicles. However, other group security schemes may support dynamic registration of vehicles.

[0051] Figure 6A diagram 600 is shown of components of a group signature scheme used in a CML network and labels and symbols used to represent the components. As shown in row 602, the symbol GV can correspond to a group of vehicles, such as a group including multiple autonomous vehicles. For example, GV refers to vehicles that use the group signature scheme when participating in the CML method. For example, each vehicle can be represented by Vi (row 612), each vehicle belongs to GV and includes autonomous vehicle perception, path planning and control systems (e.g., Figure 1 and Figure 2 102 of the autonomous vehicle perception, path planning and control system 261). For example, a first vehicle of the GV may be denoted as V1, and a second vehicle of the GV may be denoted as V2. In addition, as shown in row 604, a group manager may be denoted by the symbol GM. The GM may establish a group signature scheme by receiving detailed information of the vehicles as parameters, generating and distributing private and public keys, and revealing the identity of the vehicles in case of a dispute. For example, the GM may be a cloud server (e.g., Figure 2 A component of the autonomous vehicle cloud 256 of the present invention, such as a controller having instructions stored in a non-transitory memory for establishing and managing a group signature scheme. For example, when establishing a group signature scheme, the GM can generate a group manager secret key gmsk (line 610).

[0052] In addition to the GM, other entities involved in the group signature scheme include a collaborator CM (line 606) and a revocation authority RA (line 624). For example, the CM and the RA may also be components of a cloud server. In other examples, one or both of the CM and the RA may be hosted in a different server separate from the server hosting the GM. The CM may be a controller having instructions stored in a non-transitory manner for aggregating local machine learning models from vehicles into a global machine learning model and verifying digital signatures on messages from the vehicles. Furthermore, the RA may be a controller having instructions stored in a non-transitory memory for managing and publishing a revocation list of revoked vehicles. Furthermore, the GM may generate a group public key denoted by gpk (line 608), and may provide a different vehicle secret key gsk for each vehicle (line 614). For example, the vehicle secret key gsk may be stored in a non-transitory memory according to the vehicle secret key gsk. Figure 8 The method of assigning gpk and gsk when establishing a group signature scheme. For example, each participant in the group signature scheme can attach a digital signature σ to a message, such as a message sent to the collaborator CM containing a local machine learning model for collaborative learning. Each vehicle can generate a digital signature σ based on gsk (line 622). In addition, the gpk of the GV can be used to verify this signature σ by the CM.

[0053] Specifically, each vehicle Vi may include a local model LMi (line 616) corresponding to a model for controlling a vehicle system. For example, a second vehicle V2 in the GV may include a local model LM2. For example, LM2 may be a model for a single vehicle system such as a braking system or a steering system. For example, each vehicle may transmit coefficients and parameters from their local machine learning models to the CM via a message M (line 620). In addition, the message M may be signed with a digital signature σ, which is generated by the vehicle based on gsk. Thus, each vehicle may generate a different private digital signature. The CM may be a controller having instructions stored in a non-volatile memory for verifying digital signatures, analyzing local machine learning models from vehicles in the GV, and building updated global machine learning models. If the signature of the message M is accepted by the CM, the global model (GB, line 618) may be updated based on the message M and transmitted to each of the multiple vehicles. If the digital signature σ of the message M is verified by the CM using gpk, the CM may accept the vehicle to participate in the CML network. Figure 7 A CML network with a group security scheme is shown and the Figure 8 Method 800 describes in more detail a method for establishing and executing a group signature scheme.

[0054] In other examples, each vehicle (e.g., a first vehicle and a second vehicle) may include multiple local machine learning models, each associated with a vehicle system. Each different local machine learning model of the vehicle may be updated via a different CML network with a group signature scheme. For example, to update a local machine learning model for a braking system of a first vehicle (e.g., a braking system model), the first vehicle may sign a message including coefficients and parameters of the braking system model using a digital signature, the digital signature being generated in part based on a group public key of a first group signature scheme. The first vehicle may then transmit the message to a cloud server (e.g., a cloud server hosting a collaborator, a group manager, and a revocation mechanism), which may compile coefficients generated from a braking system model of a first group of vehicles having a braking system similar to that of the first vehicle. In addition, to update a local machine learning model for a steering system of a first vehicle (e.g., a steering system model), the first vehicle may sign a message including coefficients of the steering system model using a digital signature, the digital signature being generated in part based on a group public key of a second group signature scheme. The first vehicle may then transmit the message to a cloud server, which may compile coefficients of a steering system model of a second group of vehicles having a steering system similar to that of the first vehicle. In other words, the vehicle may participate in multiple group signature schemes with different participants such that different group signature schemes increase the privacy of each local machine learning model of the vehicle's multiple machine learning models as updated by the CML method.

[0055] The group signature scheme provides several benefits for user security. First, the identity of each vehicle can be protected, making it impossible for each vehicle participating in the group signature scheme to identify other vehicles in the group. For example, the collaborator (CM) may not be able to know the identity of the vehicle that sent the signature to the collaborator. Therefore, the group manager (GM) knows the identity of each vehicle. In addition, it is impossible for the CM to link different messages from the same vehicle. In addition, no member of the group (or GM or CM) can generate a valid signature on behalf of other participants because the gsk of a particular participant is secret and known to the vehicle. For example, the GM can detect an unauthorized participant who signs a message using a digital signature generated by the gsk of another vehicle. In addition, each vehicle that signs a message (e.g., a message including a local machine mode) using its private key (e.g., gsk) may not reject this activity in the future. For example, because the gsk is known to the vehicle, not to other entities in the network, no entity other than the vehicle can generate a legitimate signature. In addition, the group signature scheme can maintain message correctness. For example, a message with a valid signature may be accepted by the collaborator. For example, an unauthorized participant may try to inject incorrect data into the global model. To this end, an unauthorized participant may try to generate a signature. However, without the secret key (e.g., gsk), the signature may be determined to be invalid. Therefore, any invalid signature from an unauthorized participant can be detected by the collaborator during the verification phase and will be discarded. In addition, the event can be reported to the GM for further action, such as revoking the unauthorized participant from the group via a revocation authority (RA).

[0056] Next, Figure 7A CML network 700 with a group security scheme is shown. For example, a group of vehicles (GV) 718 can participate in a CML method for collaborative learning and can also participate in a group security scheme for providing data security. GV 718 includes multiple vehicles, including a first vehicle 714 and a second vehicle 716. As shown, a group manager (GM) 704 can establish a group signature scheme, for example, by distributing a group public key (gpk) and a vehicle secret key (gsk) to each vehicle in the group, such as a first vehicle 714 and a second vehicle 716. In some examples, GM 704 can allow non-member vehicles to join the group. In addition, in the event of a dispute, GM 704 can track signed messages to reveal the identity of the vehicles in GV 718. For example, each vehicle (e.g., a first vehicle 714 and a second vehicle 716) can share a local machine learning model with a collaborator (CM) 702 via a message, which can be signed using a vehicle secret key. As shown, the first vehicle 714 transmits the local machine learning model 708 to the CM 702 via a message signed with a digital signature, and the second vehicle 716 transmits the local machine learning model 710 to the CM 702 via a message signed with a digital signature. For example, each vehicle may periodically send a message with a local machine learning model to the CM 702 so that the global machine learning model may be updated to improve accuracy. Specifically, as Figure 7 As shown, CM 702 includes a global machine learning model (GB) 722. CM 702 may verify each digital signature using a group public key (gpk) and aggregate local models to update GB 722. After updating GB 722 based on multiple local machine learning models (e.g., local machine learning model 710 and local machine learning model 708), CM 702 may distribute the updated GB 722 via collaborative learning module 712.

[0057] In the event of a dispute, GM 704 can determine the identity of the vehicle. For example, if CM 702 determines that the signature on the message is invalid, CM 702 can request GM 704 to reveal the vehicle identity to determine whether an unauthorized vehicle is participating in the group signature scheme. In addition, revocation authority (RA) 706 can manage and publish a revocation list of revoked vehicles (e.g., vehicles determined by GM 704 to be unauthorized). For example, GM 704, CM 702, and RA 706 can be hosted by remote server 720.

[0058] As described above, each vehicle may include multiple local machine learning models for controlling different vehicle systems and may therefore participate in multiple CML networks with multiple groups participating in group safety schemes. In addition, each vehicle may participate in different groups of CML networks based on the local machine learning models included in each vehicle's autonomous vehicle perception, path planning, and control systems.

[0059] Next, Figure 8 A method for managing a group signature scheme to provide updated machine learning models to multiple vehicles is shown. Figure 7 The group signature scheme described is described in detail and includes a CML network of multiple vehicles. Each of the multiple vehicles may be related to Figure 1 and Figure 2 The vehicle 102 described herein. For example, each of the plurality of vehicles includes a controller and an autonomous vehicle perception, path planning and control system, the autonomous vehicle perception, path planning and control system including at least one local machine learning model for controlling the vehicle system. In addition, the CML network includes a remote server hosting a plurality of entities, the entities including a collaborator (CM), a group manager (GM) and a revocation mechanism (RA). For example, each of the CM, GM and RA can be a controller of the remote server. The remote server can be coupled to each of the plurality of vehicles in a communicative manner via a wireless connection, such as a WiFi connection, an ultra-wideband (UWB) connection, etc. Instructions for executing method 800 and the remaining methods included herein can be executed by the controller based on instructions stored on the memory of the controller and in combination with signals received from sensors of the vehicle system. For example, a portion of method 800 can be executed by a vehicle controller including executable instructions stored in a non-transitory memory. In addition, a portion of method 800 can be executed by at least one of a CM controller including executable instructions stored in a non-transitory memory, a GM controller including executable instructions stored in a non-transitory memory and an RA controller including executable instructions stored in a non-transitory memory.

[0060] At 802, method 800 includes the GM designing a group signature scheme system based on detailed information of the vehicle group, such as the number of vehicles in the group, the type of machine learning model to be updated via the CML method, and the type of vehicle. Specifically, the GM may take the detailed information of the vehicles in the group as independent variables and may proceed to perform the following mathematical process to generate gpk, gmsk, and secret keys for each user. In this process, G 1 and G 2 are two (multiplicative) cyclic groups of prime order p, g 1 It's G 1 The generator of g 2 It's G 2 In addition, ψ is a generator of G 2 To G 1 The computable isomorphism of ψ(g 2 ) = g 1The group signature scheme uses a hash function H:{0,1}*→Zp, which can be regarded as a random oracle in the security proof. The public value is g1,u,v,h∈G 1 and g 2 ,w∈G 2 Here, u, v and h are in G 1 is random, g 2 It's G 2 The random generator, g 1 Equal to ψ(g 2 ), and for some (secret) γ∈Z p , w is equal to g γ 2 The protocol proves that we have a pair (A,x) where A ∈ G 1 And x∈Z p , so that A x+γ =g 1 This pair satisfies e(A,wg x 2 )=e(g 1 ,g 2 ). To generate the key, Gm can first 2 The generator g is selected uniformly and randomly from 2 Next, the GM can set g 1 =ψ(g 2 ), and select a random number h = G 1 \{1 G1} and ξ 1 ,ξ 2 ∈ R Z* p Next, GM can set u,v∈G 1 , so that uξ 1 =vξ 2 =he. In addition, GM can choose γ∈ R Z* p And set w = g 2 γ Next, use γ to generate a tuple (A i ,x i ), where x i ∈ R Z* p , and set A i =g 1 1 / γ+xi ∈G 1 g. The group public key is gpk = (g 1 ,g 2 ,h,u,v,w), the group private key is gmsk=(ξ 1 ,ξ 2), and the secret key of each user is their tuple gsk[i] = (A i ,x i ). Thus, through this process, the GM is equipped with gpk and gmsk. In addition, the GM transmits each vehicle secret key gsk[i] to the corresponding vehicle in the group GV. Thus, each vehicle receives the gpk of the group, and each vehicle receives a different gsk that identifies the vehicle.

[0061] At 804, method 800 includes each vehicle in the vehicle group sharing a local model with the CM. For example, for a group signature scheme, each vehicle (V i ) can sign a message (M) containing the details of the vehicle's local machine learning model (e.g., the coefficients of the local machine learning model) before sharing it with the collaborator. Thus, each vehicle can follow a mathematical process to sign the message with a different digital signature σ. First, a vehicle (e.g., vehicle V i ) can select message M∈{0,1}. In addition, vehicle V i The following values ​​can be calculated: T 1 =u α , T 2 =v β , T 3 =Ah α+β , with exponents α,β∈ R Z p . The vehicle can choose r α ,r β ,r x ,r δ1 ,r δ2 ∈ R Z p , and calculate R 1 =u r α , R 2 =u r β , R 3 =e(T 3 ,g 2 ) r x .e(h,w) -r α -r β .e(h,g 2 ) -r δ1 -r δ2 , R 4 =T 1 r x .u -rδ1 , R 5 =T 2 r x .v -r δ2 Next, vehicle Vi can use the hash function to calculate the challenge c as: c = H(M,T 1 ,T 2 ,T 3 ,R 1 ,R 2 ,R 3 ,R 4 ,R 5 )∈Z p In addition, the vehicle V i The value c can be used to construct the following value: α =r α +cα、s β =r β +cβ、s x =r x +cx,s δ1 =r δ1 +cδ1、s δ2 =r δ2 +cδ2. Finally, the signature of the message is calculated as follows: σ=(T 1 ,T 2 ,T 3 ,c,s α ,s β ,s x ,s δ1 ,s δ2 ), with auxiliary values ​​δ1←xα and δ2←xβ∈Zp. Therefore, each vehicle signs its message M with a different digital signature σ (e.g., generated with gsk) and transmits the signed message to CM. Fig. 9 A method for operating an individual vehicle to send a local machine learning model to a collaborator is described in more detail.

[0062] At 806, method 800 includes the collaborator calculating and verifying the vehicle signature. For example, the collaborator may calculate the following value to verify the digital signature σ: R' 1 =u s α . T 1 –c , R' 2 =v s β . T 2 –c , R' 3 =e(T 3 ,g 2 )s x.e(h,w) -s α -s β .e(h,g 2 ) -s δ1 -s δ2 .(e(T 3 ,w) / e(g 1 ,g 2 ))c、R' 4 =T 1 s x .u -s δ1 , R' 5 =T 2 s x .v -s δ2 In addition, the collaborator can calculate the challenge c' = H(M,T 1 ,T 2 ,T 3 ,R' 1 ,R' 2 ,R' 3 ,R' 4 ,R' 5 ), and can accept a signature if and only if c is equal to c'. Therefore, the collaborator verifies each digital signature σ using the group public key (gpk) before accepting the message M.

[0063] At 808, method 800 includes determining whether a dispute is detected. For example, if the collaborator determines that the vehicle signature is invalid, for example, if c is not equal to c', a dispute may be detected. As another example, if the first vehicle attempts to identify another participant based on the transmitted message, a dispute may be detected.

[0064] If method 800 determines that no dispute is detected at 808, method 800 continues to 810 and includes determining the identity of the vehicle. For example, the collaborator may not request the group manager to determine the identity of the vehicle with an invalid signature, and all vehicles may remain in the group. In addition, since message M is accepted, the collaborator may merge the contents of M into an updated global machine learning model, which in turn may be distributed to each vehicle in the GV. For example, the updated machine learning model may incorporate detailed information from each vehicle in the GV, so that the updated machine learning model is more accurate relative to the local machine learning model of each vehicle. Therefore, the accuracy of the autonomous vehicle perception, path planning, and control systems of each vehicle may be improved. Method 800 may then end.

[0065] If method 800 determines that a dispute is detected at 808, method 800 continues to 812 and includes the GM opening the signature of the disputed vehicle to determine the identity of the disputed vehicle. For example, if the collaborator determines that c from the vehicle's signature is not equal to c', the collaborator can request the GM to open the signature and determine the vehicle's identity. This step can provide traceability so that unauthorized vehicles can be tracked and identified. To open the signature to reveal the identity of the corresponding vehicle, the GM can take as input: gpk = (g 1 ,g 2 ,h,u,v,w), the group manager's private key gmsk=(ξ 1 ,ξ 2 ), message M and signature σ=(T 1 ,T 2 ,T 3 ,c,s α ,s β ,s x ,s δ1 ,s δ2 ). In addition, the GM can verify that σ is valid on the message M using the steps given in 806. If the signature σ is valid, then the vehicle's element A can be restored to A=T by 3 / (T ξ1 1 ,T ξ2 2 )d to track the identity of the vehicle. The GM can traverse the user index corresponding to the element A of the calculated vehicle private key. Therefore, the GM can determine the identity of the corresponding vehicle.

[0066] At 814, method 800 includes determining whether an unauthorized vehicle is detected. For example, based on the identity of the corresponding vehicle, the GM can determine whether the vehicle is not authorized to join the CML network. For example, if the identity of the corresponding vehicle does not correspond to an authorized vehicle, the GM can determine that an unauthorized vehicle is detected.

[0067] If the method 800 determines that an unauthorized vehicle is not detected at 814, the method 800 continues to 816 and includes not revoking the vehicle from the group. For example, the GM may not report the unauthorized vehicle to a revocation authority (RA), and the RA may not remove the unauthorized vehicle from the group. For example, the GM may determine that the corresponding vehicle may continue to participate in the CML system.

[0068] If method 800 determines that an unauthorized vehicle is detected at 814, method 800 continues to 818 and includes revoking the unauthorized vehicle from the group. For example, the GM may notify the revocation authority that a subset of vehicles {1, 2, 3, ... r} are unauthorized and may be revoked from the group. First, the revocation authority (RA) publishes a revocation list (RL) containing the private keys gsk of all revoked vehicles. Next, the RL is provided to all vehicles and collaborators in the system to obtain an updated group public key c. Collaborators and each vehicle in the group can calculate y = Π r i =1(γ+x i )∈Z* p , to calculate g' 1 =g 1 1 / y 、g' 2 =g 2 1 / y 、w'=(g' 2 ) γ Therefore, the new public key becomes gpk = (g' 1 ,g' 2 ,h,u,v,w')h. For example, a non-revoked member whose old private key is gsk=(A,x) can calculate an updated private key (A',x), where A'=(g' 1 ) 1 / (γ+x) . Thus, an unauthorized vehicle may not compute a new public key and therefore may not send a message to the collaborator. Additionally, in some examples, the RA may add the vehicle to a revocation list when the vehicle requests to leave the group security scheme. For example, due to a change in geographic location, the vehicle may request to leave the group security scheme and the RA may add the vehicle to the revocation list. Method 800 may then end.

[0069] In this way, a group manager can establish a group security scheme for a group of vehicles, a collaborator can verify digital signatures and compile local machine learning models to update the global machine learning model, a group manager can resolve disputes, a revocation authority can remove unauthorized vehicles, and a collaborator can distribute updated global machine learning models via the CML network. Overall, the system privacy and security of the CML network is increased.

[0070] Next, Fig. 9 A flow chart of a method 900 is included for operating an individual vehicle to participate in a CML network using a group safety scheme to send a local machine learning model signed with a digital signature, and receive an updated machine learning model to improve the accuracy of autonomous operation. For example, the updated machine learning model can be updated based on data from the vehicle and other cooperating vehicles participating in the CML network. Specifically, with respect to Figure 1 Vehicle 102 and Figure 2 Method 900 is described for an onboard computing system 109 including an autonomous vehicle perception, path planning and control system 261. In addition, a vehicle may join a CML network with a group safety scheme, such as Figure 7 Instructions for executing method 900 may be stored in a non-transitory memory of the vehicle-mounted computing system (eg, Figure 2 Thus, the processor (e.g., Figure 2 The operating system processor 214) based on the stored instructions and in combination with the sensors from the vehicle system, such as the above reference Figure 2 The method 900 is performed based on the signals received by the described sensors.

[0071] At 902, method 900 includes receiving a gsk and a gpk from the GM of the group security scheme. For example, as described with respect to 802 of method 800, the GM can establish a group signature scheme based on detailed information of the vehicles in the group, including generating a gpk for the group and a gsk for the vehicle. The GM can provide the gsk and the gpk to the vehicle so that the vehicle can generate different digital signatures for the message.

[0072] At 904, method 900 includes training a local machine learning model based on sensor data. For example, the vehicle includes an autonomous vehicle perception, path planning, and control system that can control at least one vehicle system based on the local machine learning model. Using machine learning techniques known in the art, the machine learning model is updated during vehicle operation based on vehicle sensor data to increase the accuracy of the machine learning model. For example, data from the vehicle sensors can be compared to the training data to update one or more coefficients of the local machine learning model.

[0073] At 906, method 900 includes determining whether to request a CML network update. For example, based on vehicle settings and CML network settings, the vehicle may periodically provide coefficients and parameters of the local machine learning model to the CM of the vehicle group. At 906, the method determines whether to request such periodic updates.

[0074] If the method 900 determines that a CML network update is not requested at 906, the method 900 continues to 908 and includes not sending the updated model signed with the digital signature to the CM. For example, the method may not generate a digital signature and send a message with the coefficients and parameters of the local machine learning model to the collaborator. The method 900 may then return to 904 and continue to train the local machine learning model based on sensor data, connected autonomous vehicle subsystem outputs, etc.

[0075] If the method 900 determines at 906 that a CML network update is requested, the method 900 continues to 910 and includes generating a digital signature based on gpk and gsk. For example, using the mathematical process described with respect to 804 of the method 800, the vehicle computing system calculates a different digital signature σ based on gsk and gpk. As described with respect to 804 of the method 800, the mathematical process takes gsk and gpk as inputs and calculates a different digital signature as output, which can be used to sign messages to the collaborator. For example, the digital signature can be verified by the collaborator and can be tracked by the group manager.

[0076] At 912, method 900 includes signing the message with a digital signature and transmitting the message to the orchestrator via a wireless connection. For example, the vehicle-mounted computing system may include an external device interface (e.g., Figure 2 ) for wirelessly transmitting messages to an autonomous vehicle cloud (e.g., Figure 2 Autonomous vehicle cloud 256). For example, the autonomous vehicle cloud is a remote server that hosts the collaborator, the group manager, and the revocation agency. For example, the on-board computing system can sign a message including the coefficients and parameters of the local machine learning model using a digital signature and transmit the signed message to the autonomous vehicle cloud via a wireless connection. Therefore, the collaborator can receive the message including the coefficients and parameters of the local machine learning model. In addition, the collaborator can verify the signature and update the global machine learning model based on the messages from multiple vehicles in the vehicle group to increase the accuracy of the global machine learning model.

[0077] At 914, method 900 includes receiving coefficients and parameters for an updated machine learning model from the collaborator. For example, the collaborator may send an updated machine learning model to the vehicle, the updated machine learning model being based on a global machine learning model that is updated based on a shared local machine learning model from vehicles in the vehicle group. The vehicle may receive the updated machine learning model via a wireless connection to an external device interface.

[0078] At 916, method 900 includes adjusting vehicle operation based on the updated machine learning model received from the collaborator. For example, the local machine learning model can be updated based on the updated machine learning model, thereby updating the coefficients and parameters of the local machine learning model. Therefore, the local machine learning model can control the vehicle system in different ways during autonomous operation. Therefore, the autonomous vehicle perception, path planning, and control system can be updated by updating the local machine learning model using the CML method. Updating the local machine learning model can improve the accuracy of the autonomous vehicle perception, path planning, and control system, thereby improving customer satisfaction. Method 900 may then end.

[0079] In this way, autonomous vehicles can participate in the CML method to improve the accuracy of the vehicle's local machine learning model. For example, the vehicle can send the coefficients and parameters of the local machine learning model to the cloud server, and the cloud server can compile the coefficients generated from multiple local machine learning models from multiple vehicles in order to update the global machine learning model. The cloud server can transmit the updated global machine learning model to the vehicle to increase the accuracy of the local machine learning model. In addition, a group signature scheme can be formulated to verify each vehicle in the CML network and identify unauthorized vehicles. For example, the vehicle can be equipped with gpk and gsk, and different digital signatures can be generated to sign messages including the coefficients and parameters of the local machine learning model. By formulating a group signature scheme, network security can be enhanced while maintaining vehicle privacy. For example, under the group signature scheme, each vehicle may be anonymous unless a dispute is detected, and unauthorized vehicles may be detected. In addition, by adopting the CML method, the accuracy of the local machine learning model of each vehicle can be improved, which can improve customer satisfaction.

[0080] The technical effect of transmitting a message signed with a digital signature is to improve the accuracy of the local machine learning model and improve the security of the system. The message includes coefficients and parameters of the local machine learning model and the digital signature is based on a group signature scheme.

[0081] For example, a method includes: generating a digital signature in a vehicle based on a first key and a second key, both of which are received from a group manager; and transmitting a message signed with the digital signature to a collaborator, the message including coefficients of a local machine learning model of the vehicle. In the foregoing examples, the method additionally or optionally also includes: receiving an updated machine learning model from the collaborator, and adjusting the local machine learning model based on the updated machine learning model. In one or both of the foregoing examples, additionally or optionally, adjusting the local machine learning model based on the updated machine learning model includes adjusting at least one coefficient of the local machine learning model. In any or all of the foregoing examples, additionally or optionally, the updated machine learning model is updated by the collaborator based on data from the vehicle and a plurality of collaborating vehicles. In any or all of the foregoing examples, additionally or optionally, the first key is a group public key, which is distributed by the group manager to a plurality of collaborating vehicles and each of the collaborators. In any or all of the foregoing examples, additionally or optionally, the second key is a vehicle secret key, which is unique to the vehicle and distributed to the vehicle by the group manager. In any or all of the foregoing examples, additionally or alternatively, the group manager is a computing system communicatively coupled to the vehicle, the plurality of cooperating vehicles, and each of the cooperators. In any or all of the foregoing examples, additionally or alternatively, the local machine learning model of the vehicle is configured to at least partially control operation of a vehicle system of the vehicle without input from a user of the vehicle. In any or all of the foregoing examples, additionally or alternatively, the vehicle system is one of a steering system, a braking system, an acceleration system, a transmission system, an object detection system, a cruise control system, and a climate control system.

[0082] As another example, the method includes: receiving a vehicle secret key and a group public key from a group manager at a vehicle; adjusting coefficients of a local machine learning model of the vehicle based on sensor data from at least one sensor of the vehicle at the vehicle; generating a digital signature at the vehicle based on the vehicle secret key and the group public key; transmitting the coefficients of the local machine learning model to a collaborator via a message signed with the digital signature from the vehicle; receiving the coefficients of the updated machine learning model from the collaborator at the vehicle; and adjusting a vehicle system of the vehicle based on the updated machine learning model. In the foregoing examples, additionally or alternatively, each of the group manager and the collaborator is a processor external to the vehicle, each of the group manager and the collaborator is communicatively coupled to the vehicle, and the group manager is communicatively coupled to the collaborator. In one or both of the foregoing examples, additionally or alternatively, the updated machine learning model from the collaborator is based in part on the coefficients of the local machine learning model. In any or all of the foregoing examples, additionally or alternatively, the local machine learning model is a machine learning algorithm trained to autonomously control the vehicle system of the vehicle. In any or all of the foregoing examples, the method additionally or optionally also includes: receiving a second vehicle secret key and a second set of public keys from the group manager at the vehicle; adjusting coefficients of a second local machine learning model of the vehicle at the vehicle based on sensor data from at least one sensor of the vehicle; generating a second digital signature at the vehicle based on the second vehicle secret key and the second set of public keys; transmitting the coefficients of the second local machine learning model to the collaborator via a message from the vehicle signed with the second digital signature; receiving coefficients of a second updated machine learning model from the collaborator at the vehicle; and adjusting a second vehicle system of the vehicle based on the second updated machine learning model.

[0083] As another example, a system includes: a vehicle system of a vehicle; an autonomous vehicle perception, path planning and control system, which includes a local machine learning model, the local machine learning model corresponding to the vehicle system and configured to at least partially control the vehicle system; and a controller, which stores executable instructions in a non-transitory memory, the instructions when executed causing the controller to: receive a group public key and a vehicle secret key from a remote server via a wireless connection; update coefficients of the local machine learning model based on data from at least one sensor of the vehicle; generate a digital signature based on the group public key and the vehicle secret key; transmit a message signed with the digital signature to the remote server, the message including the coefficients of the local machine learning model; and receive an updated machine learning model from the remote server. In the foregoing example, additionally or alternatively, the remote server includes a group manager, a collaborator, and a revocation mechanism. In one or both of the foregoing examples, additionally or alternatively, the vehicle system is one of a braking system, an acceleration system, a steering system, a cruise control system, a climate control system, an object detection system, a lighting system, and a transmission system. In any or all of the foregoing examples, additionally or alternatively, the controller includes additional instructions stored in a non-transitory memory, which when executed cause the controller to: adjust the local machine learning model based on the updated machine learning model from the remote server. In any or all of the foregoing examples, additionally or alternatively, a digital signature is generated based on one of a BBS group signature scheme, an ACJT group signature scheme, a CG group signature scheme, and a BS group signature scheme. In any or all of the foregoing examples, additionally or alternatively, the autonomous vehicle perception, path planning, and control system uses input from a user to at least partially control vehicle operation.

[0084] In another representation, a collaborative machine learning system includes: a first vehicle, the first vehicle including a first vehicle controller and a first local machine learning model; a second vehicle, the second vehicle including a second vehicle controller and a second local machine learning model; an autonomous vehicle cloud, the autonomous vehicle cloud including a group manager controller, a collaborator controller and a revocation mechanism controller, each of the group manager controller, the collaborator controller and the revocation mechanism controller being communicatively coupled to each of the first vehicle controller and the second vehicle controller via a wireless connection, the group manager controller being communicatively coupled to each of the collaborator controller and the revocation mechanism controller, and the group manager controller including instructions stored in a non-volatile memory that, when executed, cause the group manager controller to: generate a group manager secret key and a group public key; assign a vehicle secret key to each of the characteristics of the first vehicle and the second vehicle; and in response to detecting a dispute, determine the identity of one of the first vehicle and the second vehicle based on a digital signature. In the foregoing example, additionally or optionally, the first vehicle controller includes instructions stored in a non-volatile memory, which, when executed, causes the first vehicle controller to: receive a vehicle secret key and the group public key from the group manager controller; adjust coefficients of a local machine learning model of the first vehicle based on sensor data from at least one sensor of the first vehicle; generate a digital signature for the first vehicle based on the vehicle secret key and the group public key; transmit the coefficients of the local machine learning model to the collaborator controller via a message signed with the digital signature of the first vehicle; receive updated coefficients of the machine learning model from the collaborator controller; and adjust a vehicle system of the first vehicle based on the updated machine learning model. In one or both of the foregoing examples, additionally or optionally, the second vehicle controller includes instructions stored in a non-volatile memory, which, when executed, causes the second vehicle controller to: receive a vehicle secret key and the group public key from the group manager controller; adjust coefficients of a local machine learning model of the second vehicle based on sensor data from at least one sensor of the second vehicle; generate a digital signature for the second vehicle based on the vehicle secret key and the group public key; transmit the coefficients of the local machine learning model to the collaborator controller via a message signed with the digital signature of the second vehicle; receive updated coefficients of the machine learning model from the collaborator controller; and adjust a vehicle system of the second vehicle based on the updated machine learning model.In any or all of the foregoing examples, additionally or optionally, the collaborator controller includes instructions stored in a non-transitory memory, which when executed cause the collaborator controller to: receive a local machine learning model of the first vehicle and a local machine learning model of the second vehicle; update a global machine learning model based on the local machine learning model of the first vehicle and the local machine learning model of the second vehicle; and transmit the global machine learning model to each of the first vehicle and the second vehicle. In any or all of the foregoing examples, additionally or optionally, the revocation mechanism controller includes instructions stored in a non-transitory memory, which when executed cause the revocation mechanism controller to: in response to an unauthorized vehicle, add the unauthorized vehicle to a revocation list; determine a new group public key; and distribute the new group public key to each of the first vehicle, the second vehicle, the group manager, and the controller.

[0085] In another representation, a method includes: generating a group manager secret key and a group public key at a group manager of a remote server; assigning a first vehicle secret key to the first vehicle and assigning a second vehicle secret key to the second vehicle at the group manager of the remote server; transmitting the first vehicle secret key and the group public key to the first vehicle at the group manager of the remote server; transmitting the second vehicle secret key and the group public key to the second vehicle at the group manager of the remote server; and in response to detecting a dispute, determining the identity of one of the first vehicle and the second vehicle at the group manager of the remote server based on a digital signature and the group manager secret key.

[0086] In another representation, a method includes: receiving a local machine learning model of a first vehicle and a local machine learning model of a second vehicle at a collaborator of a remote server; verifying a digital signature from the first vehicle and a digital signature from the second vehicle at the collaborator of the remote server; updating a global machine learning model based on the local machine learning model of the first vehicle and the local machine learning model of the second vehicle at the collaborator of the remote server; and transmitting the global machine learning model to each of the first vehicle and the second vehicle at the collaborator of the remote server.

[0087] In another representation, a method includes: in response to an unauthorized vehicle, adding the unauthorized vehicle to a revocation list; distributing the revocation list to each of a group manager, a collaborator, and each of a group of vehicles; determining a new group public key; and distributing the new group public key to each of the first vehicle, the second vehicle, the group manager, and the controller. In the foregoing examples, additionally or alternatively, each of the group of vehicles includes a local machine learning model configured to autonomously control a vehicle system. In one or both of the foregoing examples, additionally or alternatively, each of the revocation mechanism, the collaborator, and the group manager is a component of a remote server that is communicatively coupled to each of the first vehicle and the second vehicle. In any or all of the foregoing examples, additionally or alternatively, the new group public key is not transmitted to the unauthorized vehicle.

[0088] The description of the embodiments has been presented for the purpose of illustration and description. Suitable modifications and changes to the embodiments may be performed in light of the above description or may be acquired by practicing the methods. For example, unless otherwise indicated, one or more of the methods described may be performed by suitable apparatus and / or combinations of apparatuses, such as reference Figure 1 The described telematics unit 30. The method may be performed by executing stored instructions using one or more logic devices (e.g., processors) in combination with one or more additional hardware elements, such as storage devices, memories, hardware network interfaces / antennas, switches, actuators, clock circuits, etc. The described methods and associated actions may also be performed in various orders other than the order described in this application, in parallel, and / or simultaneously. The described systems are exemplary in nature and may include additional elements and / or omit elements. The subject matter of the present disclosure includes all novel and non-obvious combinations and sub-combinations of the various systems and configurations disclosed and other features, functions, and / or properties.

[0089] As used in this application, an element or step described in the singular and preceded by the word "one" or "a kind of" should be understood as not excluding a plurality of said elements or steps, unless such exclusion is specified. In addition, reference to "one embodiment" or "an example" of the present disclosure is not intended to be interpreted as excluding the existence of additional embodiments that also incorporate the described features. The terms "first", "second", and "third", etc. are used only as labels and are not intended to impose numerical requirements or specific positional order on their objects. The attached claims specifically point out the subject matter that is considered novel and non-obvious from the above disclosure.

Claims

1. A method for collaborative machine learning for a vehicle, the method include: generating, in a vehicle in a collaborative machine learning network, a digital signature based on a first key and a second key, both of which are received from a group manager; transmitting a message signed with the digital signature to a collaborator, the message including parameters of a local machine learning model of the vehicle; as well as The digital signature generated by the vehicle is identified, wherein the vehicle detected as invalid by the collaborator will be removed from the collaborative machine learning network.

2. The method according to claim 1, further comprising: include: Receive an updated machine learning model from the collaborator and adjust the local machine learning model based on the updated machine learning model.

3. A method as claimed in claim 2, wherein adjusting the local machine learning model based on the updated machine learning model includes adjusting at least one coefficient of the local machine learning model.

4. The method of claim 3, wherein the updated machine learning model is updated by the collaborator based on data from the vehicle and a plurality of collaborating vehicles. 5 . The method of claim 1 , wherein the first key is a group public key distributed by the group manager to each of a plurality of cooperating vehicles and the cooperator. 6 . The method of claim 5 , wherein the second key is a vehicle secret key, the vehicle secret key being unique to the vehicle and distributed to the vehicle by the group manager. 7 . The method of claim 6 , wherein the group manager is a computing system communicatively coupled to each of the vehicle, the plurality of cooperating vehicles, and the cooperator.

8. The method of claim 1, wherein the local machine learning model of the vehicle is configured to at least partially control operation of a vehicle system of the vehicle without input from a user of the vehicle.

9. The method of claim 8, wherein the vehicle system is one of a steering system, a braking system, an acceleration system, a transmission system, an object detection system, a cruise control system, and a climate control system.

10. A method for collaborative machine learning for a vehicle, the method include: receiving, at a vehicle in the collaborative machine learning network, a vehicle secret key and a group public key from a group manager; adjusting, at the vehicle, coefficients of a local machine learning model of the vehicle based on sensor data from at least one sensor of the vehicle and connected autonomous vehicle subsystem outputs; generating, at the vehicle, a digital signature based on the vehicle secret key and the group public key; transmitting the coefficients of the local machine learning model to a collaborator via a message from the vehicle signed with the digital signature; identifying the digital signature generated by the vehicle, wherein the vehicle detected as invalid by the collaborator is removed from the collaborative machine learning network; receiving, at the vehicle, updated coefficients of the machine learning model from the collaborator; and Adjusting vehicle systems of the vehicle based on the updated machine learning model.

11. The method of claim 10, wherein each of the group manager and the collaborator is a processor external to the vehicle, each of the group manager and the collaborator is communicatively coupled to the vehicle, and the group manager is communicatively coupled to the collaborator.

12. A method as claimed in claim 10, wherein the updated machine learning model from the collaborator is based in part on the coefficients of the local machine learning model.

13. The method of claim 10, wherein the local machine learning model is a machine learning algorithm trained to autonomously control the vehicle system of the vehicle.

14. The method of claim 10, further comprising: include: receiving, at the vehicle, from the group manager, a second vehicle secret key and a second group public key; adjusting, at the vehicle, coefficients of a second local machine learning model of the vehicle based on sensor data from at least one sensor of the vehicle; generating, at the vehicle, a second digital signature based on the second secret vehicle key and the second set of public keys; transmitting the coefficients of the second local machine learning model to the orchestrator via a message from the vehicle signed with the second digital signature; receiving, at the vehicle, coefficients of a second updated machine learning model from the collaborator; and Adjust a second vehicle system of the vehicle based on the second updated machine learning model.

15. A system for collaborative machine learning in a vehicle, the system include: Collaborative machine learning for vehicle-to-vehicle systems in a network; an autonomous vehicle perception, path planning, and control system including a local machine learning model corresponding to the vehicle system and configured to at least partially control the vehicle system; as well as A controller storing executable instructions in a non-transitory memory, the instructions, when executed, causing the controller to: receiving a group public key and a vehicle secret key from a remote server via a wireless connection; updating coefficients of the local machine learning model based on data from at least one sensor of the vehicle; generating a digital signature based on the group public key and the vehicle secret key; transmitting a message signed with the digital signature to the remote server, the message including the coefficients of the local machine learning model; identifying the digital signature generated by the vehicle, wherein the vehicle detected as invalid by the collaborator is removed from the collaborative machine learning network; as well as Receive an updated machine learning model from the remote server.

16. The system of claim 15, wherein the remote server comprises a group manager, a coordinator, and a revocation mechanism.

17. The system of claim 15, wherein the vehicle system is one of a braking system, an acceleration system, a steering system, a cruise control system, a climate control system, an object detection system, a lighting system, and a transmission system.

18. The system of claim 15, wherein the controller includes further instructions stored in the non-transitory memory that, when executed, cause the controller to: Adjust the local machine learning model based on the updated machine learning model from the remote server.

19. The system of claim 15, wherein the digital signature is generated based on one of a BBS group signature scheme, an ACJT group signature scheme, a CG group signature scheme, a BS group signature scheme, and a group signature scheme with dynamic registration.

20. The system of claim 15, wherein the autonomous vehicle perception, path planning, and control system uses input from a user to at least partially control vehicle operation.

Citation Information

Patent Citations

  • Securely exchanging vehicular sensor information

    CN107004091A

  • Method for improving network safety in motor vehicle, equipment and storage medium

    CN109286614A

  • Methods and arrangements for vehicle-to-vehicle communications

    CN109951823A

  • Controlling in-vehicle computing system based on contextual data

    US20150160019A1

  • Vehicle-data analytics

    US20200017117A1