Abstract: Technologies related to compressing a large language model (LLM) that has been fine-tuned using low rank adaption are described. Minimally removable structures in the LLM are identified, and node groups are constructed based upon the identified minimally removable structures. Progressive structured pruning is employed to remove prunable variables corresponding to prunable node groups of the LLM. The pruned LLM is then fine-tuned to recover knowledge lost during pruning.

Abstract: The technology described herein provides systems and methods for an Autonomous Vehicle Cloud Control System (AVCCS) with a World Model. The AVCCS comprises a cloud-based platform, a communication module, and/or an onboard unit (OBU). The AVCCS leverages generative models, predictive models, and reinforcement learning methods to generate and synthesize comprehensive information at real-time, short-term, and long-term scales for sensing, transportation behavior prediction and management, planning and decision-making, and/or vehicle control. The comprehensive information generated from the World Model comprises vehicle surrounding information, weather information, vehicle attribute data, traffic state information, road information, and incident information.

Abstract: The technology described herein provides systems and methods for an Automated Driving Cloud System (ADCS) for long-tail corner cases. The ADCS for long-tail corner cases comprises a cloud-based platform, a communication module, and/or an onboard unit (OBU). The ADCS leverages world models to provide automated driving functions including sensing, prediction, planning, decision making, and control at microscopic, mesoscopic, and/or macroscopic levels. The system is specifically designed to address long-tail corner cases, which include work zones, special events, reduced speed zones, incident detection, buffer spaces, and adverse weather conditions. Additionally, the ADCS is configured to provide safety and efficiency measures for vehicle operations and control at various special scales that require additional system coverage, including construction zones, special event zones, and special weather conditions. The ADCS enables adaptive and reliable automated driving in highly uncertain and dynamic environments.

Abstract: The technology described herein provides a cloud-based learning system (CLS) for end to end and/or sequential models for autonomous driving. The CLS provides high-performance computation capability that allocates computation power for sensing, prediction, planning and decision making, and control at a microscopic level, a mesoscopic level, and/or a macroscopic level. The CLS can acquire computation resources from a cloud system and from one or more of a roadside unit network, a network of vehicles comprising onboard units, a traffic control center/traffic control unit, or a traffic operations center. Additionally, the CLS is configured to optimize and generate detailed customized information and time-sensitive control instructions for vehicles by processing data through learning models to fulfill driving tasks and provide operations and maintenance services for vehicles.

Abstract: The technology described herein provides an improved framework for a face editing task performed by a machine-learning model. The technology provides a self-training strategy aimed at achieving more robust and generalizable face video editing. The self-training strategy helps overcome a shortage of training data relevant to the face editing task. The technology also provides a semantically disentangled architecture capable of catering to a diverse range of editing requirements. The technology also provides sparse learning to avoid over editing. The sparse learning technology partitions the model being trained according to facial regions being edited. This strategy teaches the model to transform only the most pertinent facial areas for a specific task. For example, when changing the eyebrows on a face the eye area will change, but the mouth area should remain unchanged.

Abstract: Methods, systems, and computer program products are provided for signature verification. Signature verification may be provided for target signatures using genuine signatures. A signature verification model pipeline may extract features from a target signature and a genuine signature, encode and submit both to a neural network to generate a similarity score, which may be repeated for each genuine signature. A target signature may be classified as genuine, for example, when one or more similarity scores exceed a genuine threshold. A signature verification model may be updated or calibrated at any time with new genuine signatures. A signature verification model may be implemented with multiple trainable neural networks (e.g., for feature extraction, transformation, encoding, and/or classification).

Abstract: This application describes a proactive sensing system for an autonomous vehicle. This system fuses vehicle sensing data and sensing data from a roadside unit, a Traffic Control Unit, and/or a cloud to provide full 360-degree coverage and birds-eye view of the driving environment. This proactive sensing system cooperatively uses vehicle-based sensor data and sensor data from external sources to provide better and more efficient sensing of longtail or corner cases, such as blind spots and blockage by surrounding objects. Specifically, this proactive sensing system effectively identifies major sensing points where vulnerable road users, such as pedestrians and bicycles, are major challenges for autonomous vehicles at intersections, roundabouts, or work zones. Accordingly, the technology significantly improves the safety of autonomous vehicles.

Abstract: The technology described herein provides an improved framework for novel view synthesis utilizing scene-level features and pixel-level features. In particular, the technology provides semantic representations corresponding to the scene, along with semantic representations corresponding to the each pixel, so that inherent interconnections within objects in the scene can be determined by transformer encoders that would not otherwise be determined by the pixel-level feature representations alone. In this regard, the technology described herein improves the generalizability of Neural Radiance Fields (NeRF) based techniques to novel scenes to avoid the need for retraining for specific scenes and the few-shot capability of NeRF-based techniques to render novel views using a limited number of reference images.

Abstract: The technology relates to a systematic intelligent system (SIS) configured to share data and allocate computing resources among automated driving systems (ADS), e.g., using unified data specifications and interfaces. The SIS comprises systematic intelligent units (SIU) that serve components of ADS, including vehicle intelligent units (VIU) and roadside intelligent units (RIU), and is configured to perform methods to serve an entire trip.

Abstract: The invention provides a vehicle AI computing system (VACS) that supports autonomous driving through an Onboard Unit (OBU) for vehicle-based computing and distributed computing based on vehicle-road-cloud. The vehicle-based computing can effectively complete various computational tasks by using onboard computing resources. The distributed computing allows the vehicle to work in collaboration with roadside units (RSUs) and/or the cloud to effectively complete various computational tasks. The VACS features an OBU with a sensing module, a communication module, and a data processing module that integrates data from vehicle sensors, RSUs, and the cloud. The OBU also includes a vehicle control module that helps control the vehicle based on the data of RSU and cloud. The VACS leverages high-performance computation resources to implement end-to-end driving tasks including sensing, prediction, planning and decision-making, and control.

Abstract: This invention presents a vehicle safety system (VSS) for autonomous vehicles (AVs) utilizing an onboard unit (OBU) capable of communicating with various entities, comprising roadside units (RSUs), other OBUs, and cloud platforms. The OBU incorporates a safety subsystem designed to implement proactive, active, and passive safety measures. The proactive safety measures comprise incident prediction and warnings. The active safety measures comprise emergency braking, and the passive safety measures comprise incident response and dynamic routing. This VSS can be further enhanced by incorporating sensing and processing capabilities of the RSU network or the cloud platform, enabling distributed and integrated safety functions and improved environmental awareness. The system aims to enhance AV safety by addressing incidents before, during, and after their occurrence through comprehensive safety measures at a system level, comprising a vehicle centric system, a vehicle-road system, or a vehicle cloud system.

Abstract: This technology provides systems and methods for a vehicle computing system (VCS) for autonomous driving. This VCS furnishes End-to-End models that provide sensing, prediction, planning, decision-making, and control functions. The VCS executes vehicle control algorithms, trains general AI models, and makes inferences from those AI models. Specifically, a computing subsystem of the VCS performs computation methods that train a tensor-centered model and/or make inferences from a tensor-centered model. Additionally, the VCS gathers data from a roadside unit network, an onboard unit, a cloud platform, a traffic control center/traffic control unit, and a traffic operations center (TOC), thereby enhancing the safety and efficiency of autonomous driving.

Abstract: The invention provides systems and methods for a vehicle-based cloud computing system (VCCS) for autonomous driving. This VCCS builds world models based on a series of complex scenario data to optimize sensing, prediction, planning, decision making, and control for autonomous driving. The VCCS can execute vehicle control algorithms, train general AI models, and make inferences to optimize autonomous driving. Specifically, it dynamically adjusts driving strategies based on long tail scenarios including but not limited to weather, work zone information, and traffic status, ensuring safe and efficient vehicle operation. Additionally, the VCCS can gather supplementary data from (a) a roadside unit (RSU) network, (b) another OBU, (c) a cloud platform, (d) a traffic control center/traffic control unit (TCC/TCU), and (e) a traffic operations center (TOC), thereby further improving control and efficiency in complex driving environments.

Abstract: Provided herein is technology relating to aspects of a Distributed Driving System (DDS) for managing Connected and Automated Vehicles (CAV) and particularly, but not exclusively, to systems, designs, and methods for a Device Allocation System (DAS) configured to allocate and distribute resources to devices of a Distributed Driving Systems (DDS).

Abstract: Provided herein is a technology for an Autonomous Vehicle Cloud System (AVCS). This AVCS provides sensing, data fusion, prediction, decision-making, and/or control instructions for specific vehicles at a microscopic level based on data from one or more of other vehicles, roadside unit (RSU), cloud-based platform, and traffic control center/traffic control unit (TCC/TCU). Specifically, the AVs can be effectively and efficiently operated and controlled by the AVCS. The AVCS provides individual vehicles with detailed time-sensitive control instructions for fulfilling driving tasks, including car following, lane changing, route guidance, and other related information. The AVCS is configured to predict individual vehicle behavior and provide planning and decision-making at a microscopic level. In addition, the AVCS is configured to provide one or more of virtual traffic light management, travel demand assignment, traffic state estimation, and platoon control.

Abstract: Provided herein is a technology for an Autonomous Vehicle Intelligent System (AVIS), which facilitates vehicle operations and control for autonomous driving. The AVIS and related methods provide vehicles with vehicle-specific information for a vehicle to perform driving tasks such as car following, lane changing, and route guidance. The AVIS comprises an onboard unit (OBU), wherein the OBU comprises a communication module communicating with one or more of other autonomous vehicles (AV), a roadside unit (RSU), a cloud platform, and a traffic control center/traffic control unit (TCC/TCU). The AVIS implements one or more of the following functions: sensing, prediction, decision-making, and vehicle control using onboard information and vehicle-specific information received from other AVs, the RSU, the cloud platform, and/or the TCC/TCU.

Abstract: The invention provides systems and methods for a computing power allocation system for autonomous driving (CPAS-AD), which is a component of an Intelligent Road Infrastructure System (IRIS). The CPAS-AD incorporates advanced computing capabilities that effectively allocate computational power for sensing, prediction, planning, decision-making, and control functions to enable end-to-end driving functions. In addition to the vehicle, the CPAS-AD can acquire additional computation resources from one or more of: (a) a roadside unit (RSU) network, (b) a cloud platform, (c) a traffic control center/traffic control unit (TCC/TCU), and (d) a traffic operations center (TOC). Additionally, tailored to different traffic scenarios, the CPAS-AD can allocate data and computation resources (including but not limited to CPU and GPU) for vehicle sensing, prediction, planning, decision-making, and control functions, thereby enabling safe and efficient autonomous driving.

Abstract: The invention provides systems and methods for a function-based computing power allocation system (FCPAS), which is a component of an Intelligent Road Infrastructure System (IRIS). The FCPAS incorporates advanced computing capabilities that effectively allocate computational power for prediction, planning, and decision making functions. Specifically, through the FCPAS, an AV can acquire additional computational resources for vehicle prediction, planning, and decision-making functions, thereby enabling safe and efficient autonomous driving. Additionally, tailored to different traffic scenarios, the FCPAS can allocate data and computational resources (including but not limited to CPU and GPU) for vehicle automation.

Abstract: The technology provided herein relates to a roadside infrastructure sensing system for Intelligent Road Infrastructure Systems (IRIS) and, in particular, to devices, systems, and methods for data fusion and communication that provide proactive sensing support to connected and automated vehicle highway (CAVH) systems.

Abstract: This technology provides designs and methods for the vehicle on-board unit (OBU), which facilitates vehicle operations and control for connected automated vehicle highway (CAVH) systems. OBU systems provide vehicles with individually customized information and real-time control instructions for vehicle to fulfill the driving tasks such as car following, lane changing, route guidance. OBU systems also realize transportation operations and management services for both freeways and urban arterials. The OBU composed of the following devices: 1) a vehicle motion state parameter and environment parameter collection unit; 2) a multi-mode communication unit; 3) a location unit; 4) an intelligent gateway unit, and 5) a vehicle motion control unit. The OBU systems realize one or more of the following function categories: sensing, transportation behavior prediction and management, planning and decision making, and vehicle control.