DTG-Based Dynamic Scheduling Hybrid Vehicle Control System Using AI and Method for Controlling the Same
The hybrid vehicle control system addresses high communication load and latency by reallocating resources and using adaptive transmission and federated learning to prevent secondary collisions and improve accident detection in hazardous materials transport.
Patent Information
- Application Number
- KR1020260094933
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-05-26
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2046-05-26
AI Technical Summary
Existing vehicle terminals for hazardous materials transport face issues with existing technologies have not effectively addressed the challenges of high communication and latency, specifically with existing vehicle terminals for hazardous materials transport, which are overloaded by navigation updates and the submission of transport plans, leading to false alarms and delayed responses in detecting accidents, and the inability to prevent chain collisions, and the inability to distinguish between simple communication disconnection and accidents, resulting in high operating expenses and secondary accidents.
A hybrid vehicle control system with a vehicle-mounted edge terminal and driver's second communication terminal, utilizing resource isolation, adaptive transmission control, multi-stage cross-verification, federated learning, and time-series data embedding to dynamically reallocate resources, filter false positives, and propagate emergency messages without dedicated hardware, using a commercial wireless network.
This system reduces communication latency, lowers operational expenses, prevents secondary collisions, and provides precise accident prediction and response, all while being applicable to legacy terminals and existing communication networks.
Smart Images

Figure 112026063648997-PAT00016_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to an Intelligent Transportation System (ITS) and distributed data processing. Specifically, it relates to an integrated architecture system that offloads a complex administrative task UI to a second communication terminal (such as a mobile device) to reallocate secured resources to distributed DTG computation and federated learning of an edge terminal cluster, confirms accidents through adaptive data transmission control, time-series data embedding-based similarity analysis AI, and multiple cross-validation, and propagates customized emergency avoidance messages to a trailing cluster on the same link via a commercial wireless communication network. Background Technology
[0002] Currently, vehicles transporting hazardous materials are equipped with onboard terminals for location tracking and digital tachograph (DTG) transmission. These terminals are constantly overloaded by simultaneously performing heavy applications such as closed-network navigation updates and the submission of transport plans, which causes false alarms.
[0003] In addition, when a terminal was completely destroyed in a long tunnel or similar location, the existing system could not distinguish between a simple communication disconnection and an accident, resulting in massive operating expenses (OPEX) as it relied on 24-hour manual verification by control personnel.
[0004] Above all, there was a structural limitation in that it took several minutes for the central control center to recognize the initial accident and transmit the danger to surrounding vehicles, making it impossible to prevent chain collisions (secondary accidents) involving following vehicles along the same path within the golden time. Recently, adaptive V2X (Vehicle to Everything) systems that monitor the vehicle's surrounding environment to provide early warnings of danger have been proposed, but since this presupposes an environment where dedicated communication hardware is established, it was impossible to apply them to existing legacy cargo trucks.
[0005] Moreover, legally mandated transmission systems such as the e-TAS (e-TAS) adopt a method of collecting 1-second data from terminals in 10-second intervals and transmitting it in batches to reduce communication load. However, while this preserves data resolution, it causes a physical communication latency of up to 10 seconds until the data reaches the server, which has the limitation of making real-time accident defense essentially impossible. The problem to be solved
[0007] First, while maintaining the terminal's navigation function, resource isolation techniques are applied, and administrative tasks such as the submission of transportation plans are offloaded to the driver's second communication terminal to dynamically reallocate deactivated rendering resources to DTG preprocessing.
[0008] Second, adaptive transmission control is performed to reduce communication load by maximizing local computation of the vehicle-mounted edge terminal and buffering 1-second data for 10 seconds for batch transmission during normal operation, but dynamically switching to an immediate 1-second real-time transmission mode to eliminate transmission delay when signs of an accident are suspected occur.
[0009] Third, to distinguish between communication blind spots such as long tunnels and major accidents caused by the complete destruction of the terminal, a multi-stage cross-verification logic is provided that combines navigation route masking and the sensor (impact amount) of the second communication terminal.
[0010] Fourth, when a primary risk is detected at the terminal, false positives are filtered through 'primary active verification,' which requires confirmation from the technician via an audiovisual notification (popup, etc.) before transmission to the control server.
[0011] Fifth, to prevent redundant loading of raw data into the proprietary cloud, excluding essential transmission to the national control network, we introduce Federated Learning. This involves retraining a lightweight anomaly detection AI model locally on edge terminals using raw data and transmitting only the weights to the proprietary server, thereby fundamentally eliminating the cost of loading raw data.
[0012] Sixth, when an accident of the highest risk level, such as total destruction, is confirmed, an emergency avoidance message is propagated to the network address of a cluster of trailing edge terminals on the same road link by conditionally bypassing central control based on a commercial wireless communication network without dependency on dedicated hardware (V2X), thereby preventing secondary collision accidents at the source.
[0013] Seventh, to overcome the limitations of simple threshold-based risk detection and prevent competitors from bypassing it, we provide a driving pattern embedding AI module that directly maps multidimensional time-series data collected in real-time during driving into feature vectors in a multidimensional feature space and mathematically compares and predicts the similarity with past major accident patterns in real-time. means of solving the problem
[0015] A DTG-based hybrid vehicle control system according to the present invention for solving the above problem comprises a master server, a vehicle-mounted edge terminal, and a driver's second communication terminal; wherein the vehicle-mounted edge terminal is configured in the form of a black box with a User Interface (UI) excluded, or performs mobile off-loading to transfer the UI process to the external second communication terminal to provide computing resources for edge learning, characterized by a hybrid architecture.
[0016] The present invention includes a resource isolation module that preprocesses vehicle data and DTG time-series data in a sandbox isolated from the app environment, an offloading module that transfers administrative task UI operations to a second communication terminal and returns computation resources, an edge distribution scheduler that dynamically controls resource distribution and computation processes between the vehicle-mounted edge terminal and the master server, and an adaptive transmission control logic that dynamically switches the data transmission cycle (buffering and delay time) between normal and emergency situations.
[0017] In addition, the present invention provides a cross-verification module that confirms a large total destruction accident by combining a 2D path masking table and an impact vector of a second terminal when communication is interrupted, a primary active verification (response confirmation) module, a federated learning module that updates the weights of a lightweight AI model, a time series embedding-based similarity analysis AI module that searches and analyzes distance similarity with past accident information by encoding real-time time series data into a multidimensional embedding feature space vector, and a spatial cluster propagation module that multicasts an emergency avoidance message to an adjacent edge cluster using a commercial wireless communication network, so that the above components can be organically combined according to the system purpose or operated independently as needed.
[0018] Here, the targets determined by the edge distributed scheduler and the time-series embedding-based similarity analysis AI module are not limited to specific legal standards and include abnormal driving patterns, signs of vehicle status anomalies, or preset driving risk events defined based on at least one of vehicle driving status information, driving behavior information, vehicle internal sensor data, and available external spatial information; the signs and events are configured as a higher-level concept encompassing driver drowsy driving patterns, signs of deterioration or failure of the vehicle powertrain and battery, determination of eco-driving indices for fuel efficiency optimization, detection of cargo compartment temperature and cargo status anomalies, and detection of variable vehicle behavior anomalies according to road surface conditions.
[0019] In addition, the DTG time-series data collected from the vehicle-mounted edge terminal is subjected to hybrid routing control to be transmitted to a public control server (e-TAS) through at least one of the vehicle-mounted edge terminal and the master server; when the DTG time-series data is relayed to the public control server via the master server, the master server drops the raw data of the DTG time-series data from its data storage when validity verification conditions are satisfied, such as receiving a transmission success (ACK) event from the public control server or when a preset temporary buffering period expires, thereby minimizing the infrastructure computation and load of the entire control operation system. Effects of the invention
[0021] Zero-latency incident response and traffic optimization: By performing local analysis on edge terminals rather than server-level analysis, communication load is reduced by buffering 1-second data for 10 seconds and transmitting it in batches during normal operation, while securing the golden time by switching to an immediate transmission mode at 1-second intervals without buffering waiting as soon as signs of an incident are detected.
[0022] Dramatic reduction in cloud costs (traffic) based on federated learning: Local preprocessing of DTG dangerous driving is performed by maximizing idle resources of edge terminals through user UI offloading. In particular, by completely stopping the loading of raw data into the company's cloud except for the legally required transmission to the e-TAS, and by retraining a local lightweight AI model with raw data and then encrypting and loading only the model's updated weights and event metadata, traffic and data loading costs of the analysis infrastructure can be reduced by more than 30% on a measured basis.
[0023] Reduction of control operating expenses (OPEX) and elimination of blind spots in total destruction detection: By combining navigation route-based 2D masking and multi-axis impact sensors of a second communication terminal, 'false detection in tunnels' is resolved, and operating costs are significantly reduced by lowering the reliance on manual monitoring by control personnel.
[0024] Intelligent accident cause prediction and precise response based on pattern embedding AI: Moving beyond simple numerical threshold comparison methods, this system performs feature embedding on collected raw time-series data itself and infers similarity in real-time within a stored time-series accident feature space. This allows for the precise identification of causes such as black ice slipping, drowsy driving, and tire failure, and provides customized avoidance guides for following vehicles to enable actual response.
[0025] Applicability to legacy terminals and securing golden time: It utilizes existing truck terminals and commercial wireless communication networks without expensive V2X hardware. By explicitly bypassing the control room only in the event of the highest warning level accident, such as total destruction, and immediately multicasting to network identifiers of edge clusters in the trailing direction, it achieves both broadcast storm prevention and defense against secondary chain collisions. Brief explanation of the drawing
[0027] FIG. 1 is the overall architecture of a DTG-based dynamic scheduling hybrid vehicle control system using artificial intelligence (AI) according to one embodiment of the present invention. Figure 2 is a flowchart of distributed processing for dynamic resource reallocation based on resource isolation and second communication terminal offloading. Figure 3 is a flowchart for confirming a total destruction accident based on the second communication terminal cross-verification. Figure 4 is a flowchart of time series embedding-based similarity analysis AI, federated learning, and spatial clustering secondary thinking active propagation. Specific details for implementing the invention
[0028] The definitions of key terms used throughout the specification below are as follows.
[0030] Transportation Administration UI: A data input user interface (UI) for ancillary tasks not directly related to the vehicle's actual driving or navigation route guidance, such as submitting a pre-transportation plan for hazardous materials, verifying cargo receipts, and writing driving logs, which must be processed by the driver.
[0032] Federated Learning: A distributed machine learning technique that does not transmit raw DTG data generated within an edge terminal to an internal analysis cloud except for essential transmission to an external public control network (e.g., e-TAS), but instead prioritizes using it to retrain a lightweight anomaly detection AI model installed locally on the terminal, and transmits only the weight parameters of the updated model to the cloud to update the global model.
[0034] Communication blind spots and 2D masking table: Registered through prior positioning data and navigation maps, and the masking table in the present invention is maintained in the form of a 2D table mapping 'Road Link ID - Estimated Time of Passage - Estimated Communication Quality Level of Passage', thereby masking communication disconnection events during the estimated time of passing through communication blind spots such as tunnels.
[0036] Destruction: A state of the highest risk level satisfying multiple events in which communication disconnection of both the edge terminal and the second communication terminal persists outside the grace period of the masking table, and the multi-axis composite impulse vector of the second communication terminal continuously exceeds a set time (e.g., 100ms to 500ms).
[0038] Spatial cluster: A set of dynamic IP networks of vehicle terminals located on the same road link ID relative to a specific accident location, having a dot product of the direction vectors toward the accident point greater than or equal to a reference value, located within a rear radius (e.g., 2 km), and having an estimated time of arrival (ETA) within a reference time (e.g., within 90 seconds).
[0040] The Digital Tachograph (DTG)-based hybrid vehicle control system according to the present invention is configured to include a vehicle-mounted edge terminal installed in a vehicle and a master server implemented on the cloud, and is characterized by providing a hybrid terminal and computational architecture to maximize flexibility in the system environment and hardware configuration.
[0042] The above vehicle-mounted edge terminal can be configured as a 'Blackbox-type' that is mounted inside the vehicle without a separate display panel or high-load UI (User Interface) process and is optimized solely for data collection and edge learning / computation.
[0044] Alternatively, in the case of an existing display-driven terminal, by performing 'Mobile Off-loading' to fully transfer screen output and administrative task UI processes to an external mobile terminal (smartphone, tablet, etc.) carried by the driver, the terminal is equipped with 'edge learning computing resources' that allow the physical computing processor (CPU / GPU) resources inside the terminal to be dedicated solely to edge learning and preprocessing computations.
[0046] The above master server performs 'Dynamic Scheduling' of resource distribution and computation processes, which organically determines whether to perform a specific computation process at the vehicle-mounted edge terminal or on the master server side through an Edge Distributed Scheduler, based on the computational resources provided in hardware or software of the vehicle-mounted edge terminal.
[0048] At this time, the determination criteria for the edge distributed scheduler to dynamically schedule the computation process is determined by a combination of parameters including at least one of the current CPU occupancy and memory availability of the vehicle-mounted edge terminal, the signal strength (RSSI) and available bandwidth of the wireless communication network (LTE / 5G / 6G) in the area where the vehicle is located, the current real-time session capacity of the master server, and the risk score of the road link on which the vehicle is traveling.
[0050] For example, when the wireless network status is poor due to entry into a communication dead zone or when there is sufficient spare computing resources on the terminal side (such as when a black box type terminal or offloading is complete), the weight of local terminal computation and model training is dynamically scheduled upward; conversely, when entering an urban area where communication status is excellent and the master server's resources are idle, dynamic switching control is applied to a server-centric, high-volume analysis process, thereby maximizing the network and computing efficiency of the entire monitoring infrastructure.
[0052] In the present invention, the target of the preprocessing operation performed by the vehicle-mounted edge terminal or master server is not limited to specific legal standards, and it extensively determines abnormal driving patterns, signs of abnormal vehicle status, or preset driving risk events defined based on at least one of vehicle driving status information, driving behavior information, vehicle internal sensor data (CAN data and impact sensors, etc.), and available external spatial information (GIS, tunnel / shaded area masking table, etc.).
[0054] The aforementioned determination targets include all types of intelligent analysis event patterns associated with vehicles and operation, such as the Korea Transportation Safety Authority's standard dangerous driving behaviors, driver drowsy driving patterns, detection of signs of degradation or failure in vehicle powertrains and electric vehicle batteries, determination of eco-driving indices for fuel efficiency optimization, detection of abnormal cargo temperature and cargo condition, and signs of variable vehicle behavior abnormalities due to road surface conditions (black ice, potholes, etc.).
[0056] The above vehicle-mounted edge terminal continuously identifies and learns the target patterns through a locally mounted lightweight anomaly detection AI model, and selectively transmits only the weight parameters of the AI model and the identified event metadata—which are the results of the learning—to the master server to load them into data storage.
[0058] In the case of DTG time-series data (raw data), which is legally mandated data for submission to government agencies, it is controlled in a hybrid form so that it is either directly wirelessly transmitted from the vehicle-mounted edge terminal to the public control server (e-TAS) or relayed to the public control server via a master server, depending on the system's network architecture settings.
[0060] If the above DTG time series data is relayed to the public control server (e-TAS) via the master server, the master server drops the raw data of the DTG time series data within its data storage if at least one validation condition is satisfied, such as receiving a transmission success event from the public control server or the expiration of a preset temporary buffering period.
[0062] Through this, the wireless communication session on the terminal side is maintained as a single session to prevent an increase in overall communication costs, while simultaneously drastically reducing the burden of accumulated loading of large volume raw data entering the central server and storage maintenance costs (OPEX) by more than 70% compared to conventional technology.
[0064] Furthermore, the present invention goes beyond simple threshold-based risk detection and integrates a Time-series Embedding-based Similarity Analysis AI architecture. Specifically, multidimensional DTG time-series data for a preset period immediately prior to the occurrence of an accident. (Here, driving information for each time period) A D-dimensional driving variable vector (including vehicle speed, RPM, braking pressure, steering angle, acceleration sensor values, etc.) passes through a driving pattern embedding encoder having a multi-layer neural network structure such as 1D-CNN or Transformer to become a unique feature vector in a high-dimensional feature space. It is mapped and encoded into an (M-dimensional continuous embedding vector).
[0066] The master server is a set of reference feature vectors in the feature space representing representative major incident patterns that have been pre-collected and labeled. ( It is managed by pre-constructing it as ). When the driving time series embedding vector z transmitted from the edge terminal is received by the master server, the similarity analysis AI module calculates the cosine similarity in the feature space Calculate at high speed or:
[0067]
[0068] Or, perform K-Nearest Neighbor search (K-NN) to find a predefined threshold for similarity Determines specific past accident cause patterns that are both abnormal and maximum (e.g., black ice road slipping).
[0070] Once the accident is finally confirmed, the master server, in order to overcome the limitations of legacy terminals, transmits customized emergency avoidance messages and avoidance guidelines (e.g., "Entering section suspected of black ice slipping, please apply engine brake immediately") in real time as emergency signals via multicast propagation directly to the terminal network address (IP) session of the 'spatial cluster' mapped to the aforementioned conditions, based on a commercial wireless communication network and without going through a separate V2X module.
[0072] Meanwhile, regarding the operation and condition determination of each module according to the present invention, numerical setting examples are specified to support the technical feasibility of the system and provide specific implementation possibilities. It is specified that the numerical values and threshold values disclosed below are merely examples assuming a specific operating environment, and that the technical scope of the present invention is not limited by these specific numerical values.
[0074] First, the DTG data transmission cycle of the vehicle-mounted edge terminal is dynamically switched to optimize network load and maximize the utility of edge computing. The edge terminal collects high-resolution time-series data in real time at intervals of one second (or less), but for transmission to the server, it operates in a normal mode in which approximately 10 seconds of data during normal driving is accumulated in a local buffer and transmitted in batches at 10-second intervals. This results in no loss of resolution upon arrival at the server but entails a communication delay (latency) of up to 10 seconds. However, if a preprocessing algorithm or lightweight AI model at the edge terminal's local location directly detects signs of an accident, or if a suspected risk point is detected where the multi-axis composite impact force index exceeds a preliminary threshold, the device switches to a real-time emergency mode that skips buffer waiting and immediately transmits data at 1-second intervals, thereby supplying data to the AI of the master server so that it can infer the cause of the accident without delay.
[0076] Second, the multi-axis composite impulse index that determines the confirmation of a total destruction accident at the second communication terminal The detection logic is based on the following formula.
[0077]
[0078] Here , , is the acceleration value per axis measured by the acceleration sensor of the second communication terminal, and represents the magnitude of the composite acceleration vector. To filter out one-time impact errors, such as a simple drop of a device, the magnitude of the composite acceleration vector A total destruction shock event is recognized as valid only when the preset threshold shock value (preferably 10G) within the range of 8G to 12G is satisfied for a specific continuous time interval of 100ms to 500ms (preferably 200ms or more).
[0080] Third, the 2D masking table is activated to dynamically delay the reception of communication disconnection events when entering a registered tunnel or an area with unstable GPS reception on a road link scheduled for driving. The masking delay time is calculated by correlating the actual distance of the tunnel with the real-time speed of the driving vehicle, and is maintained for a quantified weighted delay filter time of at least 5 seconds to a maximum of 20 seconds (preferably 10 seconds) relative to the time of completion of passing through the tunnel, taking into account the transition margin until the vehicle completely exits the tunnel and the commercial network connection stabilizes. The masking table is dynamically triggered in an environment where the threshold level of communication quality drops to -105dBm to -115dBm (preferably -110dBm) or lower based on the Received Signal Strength Indicator (RSSI).
[0082] Fourth, the setting of the risk reach range in the spatial clustering mapping module is physically calculated based on the braking limit distance of the heavy cargo vehicles following a rollover of a hazardous material transport vehicle on a highway. The rear threshold radius is limited to a range of at least 1 km and up to 3 km (preferably within 2 km), and the estimated time of arrival (ETA) limit for selecting accident-reaching vehicles is dynamically differentially set to a maximum of 60 seconds to 120 seconds (preferably within 90 seconds) by tracking the variable driving speed of the vehicles. Industrial applicability
[0084] The present invention is immediately and substantially applicable to intelligent control systems, B2G hazardous materials transport safety management systems, highway C-ITS projects, ultra-large AI-based traffic risk prediction platforms, and cloud data migration architecture infrastructure.
[0085] No content Explanation of the symbols
[0086] 10: Vehicle involved in the accident (Edge Node) 100: Master Server (Cloud Monitoring Center) 110: Edge Distributed Scheduler 120: Heterogeneous device cross-validation module 130: Time Series Embedding Similarity AI (Feature Space) 140: Spatial Cluster Active Propagation Module 200: Vehicle-mounted edge terminal 210: DTG computation agent 220: Lightweight AI learning models 300: Second communication terminal (driver mobile device) 310: Transportation Administration UI 320: Multi-axis impact sensor 400: Lagging spatial clustering 410, 420: Trailing edge terminals A, B 500: Public Control Server (e-TAS) S210~S250: Each step of distributed processing for dynamic resource reallocation S310~S370: Confirmed total destruction accident, each stage S410~S470: Each stage of similarity search and active propagation
Claims
Claim 1 A vehicle control system based on a digital tachograph (DTG), comprising a vehicle-mounted edge terminal mounted in a vehicle and a master server in the cloud, wherein the vehicle-mounted edge terminal is configured in the form of a black box excluding a User Interface (UI) or performs mobile offloading to transfer the UI process to an external second communication terminal to provide computational resources for edge learning, and the vehicle control system based on the digital tachograph (DTG) further comprises an edge distributed scheduler that controls the computational load in conjunction with the vehicle-mounted edge terminal and the master server, wherein the edge distributed scheduler monitors the real-time available computing resources of the vehicle-mounted edge terminal and dynamically schedules resource distribution and computation processes so that preprocessing computations and local learning of an Artificial Intelligence (AI) model are executed at least one of the vehicle-mounted edge terminal and the master server, and the preprocessing computation involves an abnormal driving pattern, a sign of an abnormal vehicle state, or a preset driving risk event defined based on at least one of vehicle driving state information, driving behavior information, vehicle internal sensor data, and available external spatial information. A digital tachograph (DTG)-based vehicle control system characterized by determining, and controlling the DTG time series data collected from the vehicle-mounted edge terminal to be transmitted to a public control server (e-TAS) through at least one of the vehicle-mounted edge terminal and the master server. Claim 2 A digital tachograph (DTG)-based vehicle control system according to claim 1, wherein the vehicle-mounted edge terminal operates in a normal mode in which it accumulates the DTG time series data collected at a preset resolution in a local buffer during a first time interval and transmits it in batches during a normal driving state, and further includes an adaptive transmission control logic that dynamically switches to a real-time emergency mode in which, when a risk suspicion point is detected through process monitoring within the edge terminal, such that a sign of an accident is locally detected or an impact indicator exceeds a preliminary threshold, the collected DTG time series data is immediately transmitted in a second time interval period shorter than the first time interval to eliminate communication latency caused by buffering during the first time interval. Claim 3 The digital tachograph (DTG)-based vehicle control system according to claim 1 comprises a plurality of vehicle-mounted edge terminals mounted on each of a plurality of vehicles, and the edge distributed scheduler continuously monitors changes in the availability of computational resources provided in each of the plurality of vehicle-mounted edge terminals and dynamically reallocates DTG dangerous driving event preprocessing tasks among the plurality of vehicle-mounted edge terminals, wherein, when reallocating tasks, the system comprehensively reflects not only the network delay and CPU load of individual terminals but also the accident risk level per road link, and prioritizes reallocating preprocessing computational resources and tasks to vehicle terminals located on road links with high warning risk, and wherein the monitoring cycle for changes in the availability of computational resources is controlled based on a pre-set threshold cycle condition. Claim 4 A digital tachograph (DTG)-based vehicle control system according to claim 1, wherein the vehicle-mounted edge terminal identifies a plurality of target patterns associated with vehicle driving and status using a locally mounted lightweight anomaly detection AI model; the master server receives the weights of the lightweight anomaly detection AI model and the identified event metadata from the vehicle-mounted edge terminal and loads them into a data storage; and when the DTG time series data is relayed to the public control server (e-TAS) via the master server, the master server drops the raw data of the DTG time series data within the data storage of the master server when at least one validity verification condition is satisfied, such as receiving a transmission success event from the public control server or the expiration of a preset temporary buffering period. Claim 5 delete Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 delete Claim 10 delete
Citation Information
Patent Citations
Vehicle accident notification system and method of providing vehicle accident notification service
KR1020230053815A
System and Method for Road Traffic Hazard Guidance and Automatic Safety Control Responsive to Driver Unresponsiveness Using Trajectory-Based Accident Recovery Information Dissemination
KR1020260070053A
Apparatus for vehicle emergency accident notification and method thereof
KR102948433B1
Systems and methods for improving scheduling of task offloading within a vehicle
US20210406065A1
Detection of irregularity in linked cloud-edge vehicle
JP2018144800A
