top of page

Development of a Generalized Integrity Monitoring Framework FOR CAV APPLICATION

​

Project Description: CAV applications can be broadly classified into three major categories: safety, mobility, and environmental. Mobility and environmental applications require a coarse positioning accuracy (5-10 m) and lane-level positioning accuracy (< 1 m), while safety-critical applications require a where-in-lane level accuracy (< 0.2 m). Uncertainty in positioning information can sabotage driving functionalities and cause a safety concern. Position uncertainty of a vehicle can be attributed to the sensor suite available on the vehicle, along with any additional external sensor information that may aid the in-vehicle sensors. Furthermore, a fully connected and Automated vehicle (CAV) may turn itself into a degraded CAV, AV-only, CV-only, or Human-driven vehicle (HDV) while experiencing communication or control loss. The type, quantity, placement, and measurement uncertainty of sensors play a great role in determining the navigation performance of a vehicle. Additionally, in a mixed traffic scenario, there exist various types of positioning solutions, vehicles, sensor modalities, communications capabilities, and applications. It is unlikely that two vehicles having the same set of sensors and computation hardware will output similar navigation performance. Therefore, it is important to study and analyze the integrity of positioning systems under various conditions. 

Integrity monitoring (IM) methods have been extensively studied and developed for in-vehicle sensor systems primarily consisting of GNSS receivers (e.g. RAIM) often integrated with IMUs, vehicle odometry, and perception sensors such as radar, camera, and LiDAR. The localization performance and safety of a vehicle have mainly been studied from an ego vehicle's perspective focusing on enabling automated driving functions through onboard sensors and compute platform. Further, there is a surge in V2V/V2I/V2P/V2X research focusing on information sharing between road agents for improved positioning, navigation, and control. Given the number of sensor sources and the amount of data shared between road agents, it is imperative to develop integrity monitoring frameworks for cooperative scenarios. Based on our initial literature survey, despite the growing interest in this area, there is a noticeable gap in the current literature that addresses the "cooperative-IM (integrity monitoring)" framework, indicating the need for further research to propose new Required Navigation Performance (RNP) parameters that may support CAV applications.  

As a part of our smart intersection projects (City of Riverside & City of Rialto), we have conducted various Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) based cooperative positioning experiments, where perception data from both onboard and roadside sensors i.e. camera, LiDAR, is shared among the vehicles and infrastructure through C-V2X communication. We have successfully demonstrated improved vehicle detection and positioning accuracy (0.3 meters with Roadside LiDAR). Furthermore, we are currently experimenting with Cohda wireless MK6 modules to transmit basic safety messages (BSMs) and other information among vehicles in a V2V setting. However, the question remains whether the empirically determined Required Navigation Performance (RNP) values hold consistent across different driving conditions, positioning hardware, and communication topologies. Inconsistencies in RNP parameters would lead to the failure of CAV applications. To address these issues, the following tasks are proposed: 

Task 1: V2X Dataset Review for Cooperative Perception and Positioning in Traffic Environments 

This task involves a comprehensive review of publicly available V2X datasets, such as TUMTraf-V2X and V2V4Real, to assess their suitability for cooperative object detection and tracking. The datasets, containing data from onboard and roadside sensors like multi-band RTK GNSS, IMU, camera, and LiDAR, will be examined for challenging traffic scenarios including traffic violations, near misses, and occlusions. Additionally, sensor data quality and coverage will be assessed to understand its applicability for ego-vehicle positioning enhancements through techniques like cooperative perception and SLAM. This review will also consider the impact of V2V & V2I communication constraints, such as packet dropouts and latency, on positioning accuracy to establish a baseline for developing a CAV-centric real-time IM framework. 

Task 2: Track-to-track fusion for tracking surrounding road agents in challenging traffic scenarios 

This task will explore various track-to-track fusion algorithms, such as Covariance Intersection (CI) and Inverse Covariance Intersection (ICI), to estimate real-time vehicle states for the Ego vehicle and surrounding road agents. Using the TUMTRAF-V2X dataset, the intersection node will run a local tracker that leverages multi-modal sensor fusion techniques like BEVFusion, combining data from cameras and LiDAR. Meanwhile, the vehicle node, equipped with onboard GNSS, IMU, cameras, and LiDAR, will run a separate local tracker that integrates these sensor inputs. The fusion process will involve combining the local tracks from both the vehicle and the intersection to produce a global track of all agents at the intersection. To simulate real-world conditions, we will introduce communication constraints, such as latency and packet loss in exchanging local track information, to assess the impact on the quality of the global track. Given the time-critical nature of certain maneuvers and safety-related Connected and Autonomous Vehicle (CAV) applications, we will also investigate optimal sensor selection methods. Techniques such as Dempster-Shafer evidence theory could be applied to strike a balance between tracking speed and uncertainty, optimizing the overall performance in estimating road agents’ states. 

Task 3: Cooperative tracking of Vulnerable Road Users (VRUs): Leveraging data from smart devices 

In this task, we aim to enhance the cooperative tracking of Vulnerable Road Users (VRUs), such as pedestrians and cyclists, by integrating data from smart devices (e.g., smartphones, wearables) with traditional sensors from vehicles and infrastructure. This multi-sensor fusion approach leverages GPS, accelerometers, and gyroscopes from VRUs' smart devices, combined with data from vehicle-mounted cameras, radar, and LiDAR, as well as infrastructure-based sensors (e.g., roadside cameras. We will fuse these diverse data streams to generate accurate real-time tracking of VRUs, even in challenging environments where direct line-of-sight might be obstructed. By leveraging cooperative localization through Vehicle-to-Everything (V2X) communication, the system will share positional data between vehicles, infrastructure, and VRUs’ smart devices. We have collected data from an intersection in Rialto, CA, where two Ouster OS1-128 LiDAR sensors were mounted on traffic poles to monitor the surroundings. In this dataset, we also gathered smartphone data from controlled participants simulating various Vulnerable Road Users (VRUs), such as pedestrians and scooter riders. Moving forward, we plan to expand our data collection efforts at the smart intersection of Iowa and University Ave. in Riverside, CA, which is equipped with cameras and LiDAR sensors. Additionally, we will leverage a sensor suite-equipped test vehicle to enhance our data collection capabilities and further improve VRU tracking and analysis. 

Task 4:  Resiliency in vehicle positioning in real-world GNSS-compromised environment: 

With STARNAV’s support, a GNSS-compromised region will be created which will include spoofing and jamming of GNSS satellite signals affecting the positioning performance of the test vehicles equipped with a GNSS receiver. The goal is to equip the test vehicles with a sensor platform containing perception sensors such as a camera, LiDAR, and Cohda MK6 wireless modules. The spoofed GNSS measurements, perception data, and information shared between the test vehicles through the MK6 modules will be logged for post-processing and analysis purposes. Existing V2V-based cooperative datasets such as V2V4Real, DAIR-V2X, etc. do not contain spoofed or jammed GNSS measurements. Thus, this experiment is an attempt to build a V2V dataset where GNSS measurements are spoofed with a GNSS spoofer. Furthermore, a Fault Detection and Isolation (FDI) technique such as a cross-consistency check among different sensor information will be implemented to identify faulty GNSS measurements and isolate them. This experiment is set to be conducted in late September 2024 in the state of New Mexico, USA. 

US DOT Priorities: This research project aligns with USDOT’s goals of enhancing road safety by developing safety management systems and driver assistance technologies. We will investigate the impact of positioning uncertainty on the feasibility of CAV (Connected and Autonomous Vehicle) applications and work towards developing a CAV-centric Integrity Monitoring (IM) framework. Failure to detect and report sensor failures, measurement errors, and cybersecurity risks can jeopardize a CAV application and compromise passenger safety. Our research will involve testing and analyzing Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication-based sharing of sensor information, and redefining existing Required Navigation Performance (RNP) parameters to better meet the needs of CAV applications. This approach will enable us to dynamically adjust various parameters, making CAV applications more robust and resilient against measurement uncertainties and external threats. 

Outputs:  In this project: 

  • A new Required Navigation Performance (RNP) framework will be developed to cater to the needs of CAV applications. In addition to the existing parameters (accuracy, continuity, integrity risk) provided by the classical RNP frameworks, new parameters such as “Timeliness”, and “Interoperability ” will be explored, monitored, and quantified for different CAV applications. 

  • Sensor malfunctions, faults, misinformation in communication, etc. will be studied and Fault Detection and Isolation (FDI) strategies shall be developed to address these issues. 

  • Sensitivity analysis will be performed in simulation to evaluate the feasibility of CAV application for an ego vehicle under different sensor modalities, information uncertainties, driving, and environment conditions, which would be otherwise difficult to test in real-world situations.  

  • Automotive functional safety standards will be reviewed to understand the positioning requirement definitions in the automotive industry and use them as a baseline to draft the CAV-centric Integrity Monitoring (IM) framework. 

Outputs/Impacts:  

  • CAV-centric Integrity Monitoring (CAV-IM) framework will be developed and implemented for key CAV applications such as Intersection Movement Assist (IMA) to relay warnings to the drivers or engage in evasive maneuvers to ensure safety. 

Final Research Report: Upon completion of the project we will provide a link to the final report. 

Image by Annie Spratt

SUBSCRIBE

Sign up to receive CARNATIONS news and updates.

Thanks for submitting!

© 2023 by CARNATIONS. Powered and secured by Wix.

  • Youtube
  • LinkedIn
  • Instagram
bottom of page