105 - Hybrid black-white-modelling estimation and machine learning algorithms for PNT engines

The application of machine learning (based on “black-box” modelling) is of interest in problems that are difficult to solve based on traditional (optimal) estimators for simple models (“white-box” modelling). Black-box modelling is difficult to be explained and understood (i.e., difficult to understand what to expect in unknown or new situations), involve high computation complexity, and is not necessarily deriving the optimal solution.

In many problems traditional methods are optimal and require much lower number of operations than machine learning techniques. Moreover, their behaviour can be better understood. In the context of PNT engines, traditional solutions can provide optimal solutions in controlled environments, while require more advanced solutions in difficult environments. A hybrid black-white estimation approach targets to exploit the best of both approaches, enhancing and refining the solution of the traditional solution with machine learning only when needed. One of many other possible examples of hybrid processing is KalmanNet, in this case focusing on the learning of complex dynamics based on EKF (as “white-box” modelling).

The objective of this activity is to study, design, implement and demonstrate the application of hybrid black-white-modelling estimation and machine learning algorithms for PNT engines, exploiting the best of each approach to achieve optimum solutions with minimum resources used. 

Note that EL1-087 (verifiable AI) does not cover the achievement of optimum performance with hybrid solutions while minimizing the required resources. Its results will be included in the solicitation package, as far as possible, being relevant for the verification point of view.  

The tasks to be performed shall include:

• improve the achieved accuracy in harsh propagations, while being also optimum in mild propagation conditions 
• minimize the number of resources required in the receiver
• demonstrate that the hybrid approach works in a controlled and accurate way when operating in new changing environments
• assessment of the derived solutions, to be performed based on field tests, considering code and carrier observables from commercial GNSS receivers, as well as inertial sensors.

The main outputs of the activity will consist of:
 

• Definition and design of innovative hybrid black-white PNT solutions
• Hybrid black-white PNT engine processing tool and test report
• Roadmap for commercialization, including potential Navisp EL2 activities.