02 Mar 2021

Maintaining separation minima is crucial to ensure the security of UAS operation. Preventing collisions has become a common challenge for industry and academia around the world. Many studies have been undertaken to explore the integration of the airspace by UAS and to model the separation safety, some of them have been described in the document.

The main objective of WP4 was to develop and to devise algorithms for assigning separation methods and computing separation minima. The main contributions of this WP have been: 1) the definition of a novel risk model applicable to small UAS, 2) the definition of algorithms to compute separation minima and methods for a given use case to guarantee a Target Level of Safety (TLS). Implementation of such models and algorithms will be performed and evaluated in WP5 and WP6.

Deliverable ** D4.1 Algorithm for separation minima determination** describes a novel methodology to compute the risk model and to define separation minima and methods for UAS. Different approaches are possible: “qualitative models” (SORA methodology) or “quantitative models” (e.g., Reich model) or even a mix of both. The BUBBLES model assumes that strategic mitigations are used to reduce the frequency of conflicts, while tactical mitigations are used to reduce the probability that a conflict degenerates into a collision.

The BUBBLES risk model is based on dividing the collision rate in two components:

*Freq(collision) = Freq(conflict) × Prob(collision *|*conflict)*

where *Freq(conflict) *depends on use-case representative trajectories, while *Prob(collision *|*conflict)* is estimated with pairwise analysis. More specifically, *Freq(conflict) *evaluates the impact of strategic mitigations, while *Prob(collision *|*conflict) *evaluates the impact of tactical mitigations.

To evaluate the risk of strategic mitigations, it is necessary to build a model estimating the conflict frequency . Several types of models can be used to evaluate this frequency:

*Analytical models*, including probabilistic models like the Gas Model or the Reich-Marks model, used by ICAO in Doc. 9689 for the determination of existing separation minima in manned aviation.*Simulation models*, generating trajectories of aircrafts and estimating conflict rates through simulations.

Since BUBBLES risk model is based on use-case representative trajectories, we have used simulation models to generate specific techniques. Methods for trajectory generation include:

*Bayesian models*. These models are based on Bayesian Networks. They use real radar tracks to calculate the weights of the network and to determine the optimal structure of the network.*Artificial Intelligence (AI) methods*. AI models need data for the learning process. These data can come from real radar tracks or from simulations. A Monte Carlo simulation is adopted to estimate the conflict rate. The idea is to simulate behaviours of the UAS in an operational scenario and to estimate the conflict rate. The advantage with respect to analytical models is the flexibility of modelling particular scenarios, dropping out many assumptions made in the analytical models. To achieve that, a set of random trajectories representative of a set of missions has been generated; the trajectory, as the basis of distributing flight information, is the core information used by the system.

Some examples of generated trajectories are illustrated in the Deliverable.

Probability of collisions given a conflict can be estimated instead by considering only a pairwise analysis. When two aircraft are involved in a conflict, the probability of collision depends on the deadline for solving the conflict: the lesser is the deadline for solving a conflict, the higher is the probability that the conflict ends up in a collision.

Several situations are analysed and tables reporting separation minima (also denoted as Remain Well Clear parameters) are provided.

A full example is provided to illustrate the estimation of conflict frequency and probability of collisions obtained by the BUBBLES risk model.

Deliverable *D4.2***Guidelines to implement separation minima and methods **describes the methodology for automatic computation of separation minima and methods for UAS. The methodology can be applied to a given use case and compute results that guarantee a predefined Target Level of Safety (TLS), exploiting the BUBBLES collision risk model described in Deliverable D4.1.

More specifically, the methodology is based on the use of Artificial Intelligence (AI) decision-making techniques for automatic generation of trajectories in some training use cases and Machine Learning (ML) techniques to train models and use their predictions to compute separation values in new (similar) use cases.

Summarizing, the overall methodology is based on four steps:

- extract relevant information from a use case to generate representative trajectories
- use such trajectories and the BUBBLES collision risk model to create a data set for separation values
- train ML models with such data
- use ML predictions to determine separation minima and methods for a new use case and a desired TLS.

The overall output of the proposed methodology is a set of ML models and functions to combine their predictions that are able to estimate complex strategic mitigations and the corresponding separation minima.

For example, we can deal with use cases in which many UAS are assigned to different kinds of missions, in which each mission is composed by a set of mission phases, and each phase is associated with a specific trajectory type. BUBBLES methodology will assign to each mission phase of each UAS involved in the use case suitable separation minima and methods that guarantee a given TLS.

A simple example is described in Deliverable D4.2 to show the application of the proposed methodology.

Further developments include implementation, testing, and validation of this methodology (WP5) and definition of safety and performance requirements (WP6).

Deliverables D4.1 and D4.2 are available for download here: