127 - Cross-Domain Nonlinear State Estimation for Autonomous Systems Using Unscented Kalman Filtering

Autonomous systems, whether operating on land, in the air, at sea, or in structured indoor environments, require robust, real-time state estimation to navigate complex, dynamic, and GNSS-challenged conditions. Traditional estimation techniques such as the Extended Kalman Filter (EKF) often rely on local linearization of the system dynamics and measurement models, which can result in degraded performance in highly nonlinear and uncertain environments. This limitation is particularly acute when combining heterogeneous sensors (e.g., GNSS, IMU, LiDAR, and vision), where unmodeled nonlinearities and noise can accumulate rapidly and compromise safety, resilience, and autonomy.

This study proposes the development of a modular Unscented Kalman Filter (UKF)-based state estimation engine designed to operate reliably in GNSS-challenged and sensor- degraded scenarios. The UKF, which propagates a distribution of sigma points through nonlinear dynamics without requiring Jacobian calculations, has demonstrated superior accuracy over EKF in simulated and real-world applications. Despite its theoretical advantages, the UKF remains underutilized in real-time autonomous platforms due to computational concerns and a lack of integrated cross-domain frameworks.

The objective of this activity is to design, prototype, and validate a real-time, UKF-based estimation framework optimized for autonomous systems, enabling robust multi-sensor fusion and accurate vehicle state estimation under challenging conditions.

By leveraging advanced unscented transformations and adaptive noise modelling techniques, the framework will be designed to preserve accuracy and integrity even in scenarios involving sensor degradation, GNSS outages, or highly dynamic motion.

The activity will target, as the primary domain, either ground vehicles, integrating data from automotive-relevant sensors (GNSS, IMU, odometry, LiDAR), or aerial systems (e.g. drones), at the discretion of the bidder, to demonstrate feasibility via an experimental campaign.

Key innovations include:

•  Nonlinear estimation without linearization, preserving model fidelity and accuracy;
•  Adaptive noise modelling to enhance robustness in variable conditions;
•  Modular, domain-specific architecture suitable for integration into ADAS, autonomy stacks, or vehicle control systems;
•  Platform validation, including optional demonstration on ground vehicle and / or UAVs to illustrate domain portability.

The tasks to be performed shall include:

•  Use Case Description & Requirements Capture;
•  System Modelling & Sensor Abstraction;
•  UKF Framework Design;
•  Prototype Deployment and Hardware-in-the-Loop (HiL) Testing;
•  Field test Validation, performing comparative testing with EKF and other benchmarks;
•  Impact Assessment & Roadmap for commercialization or further development

The main outputs of the activity will consist of:

•  A validated UKF prototype for real-time multi-sensor fusion;
•  A performance benchmark against EKF and other conventional filters;
•  A reference dataset for testing in automotive scenarios, including urban, rural, and off- road use cases, or in aerial systems scenarios;
•  An integration roadmap for commercial automotive or aerial systems applications, including, for example, potential follow-up work on compliance to ISO-26262 (Road vehicles – Functional safety standard), automotive-grade hardware deployment, and open-source module release.

It is noted that no Participating State expressed their opt-out for this activity (EL1-127).