1. DART. Data-driven AiRcraft Trajectory
José Manuel Cordero García
Data Driven ATM: Going Digital!
8th March 2018
2. DART Consortium
SESAR 2020 - Exploratory Research 2
University of Piraeus Research Center - UPRC
Reference Centre for Research, Development and
Innovation in ATM (CRIDA)
Fraunhofer Gesellschaft zur Förderung der
Angewandten Forschung e.V. (IAIS)
Boeing Research & Technology Europe
3. DART Objectives and Impact
SESAR 2020 - Exploratory Research 3
Challenge of exploring the applicability of data science and
complexity science techniques to the ATM domain with the aim of
improving aircraft trajectory prediction capabilities of the ATM system
It is expected that data-driven techniques help to improve the
accuracy of predictions by complementing classical model-based
4. Problem addressed by DART
SESAR 2020 - Exploratory Research 4
Minimizing the cost throught
maximizing the adherence to the
airlines preferred FPs.
Decide which flights to modify to
resolve sector imbalances and
Minimizing the sector
imbalances and potential
DART aims to develop a collaborative decision making process that would support
multi-objective optimization taking into account stakeholders’ expectations.
Collaborative Trajectory Prediction
Select the flights
to modify (NM)
from an individual
STP predict new single
trajectories for these flights
CTP detection of imbalances
and conflicts, selection of
flight to modify and
proposition of new FPs
new FP for
Network Manager Air Navigation
Solution proposed: Single and
Collaborative Trajectory Prediction
SESAR 2020 - Exploratory Research 5
6. DART Concept
SESAR 2020 - Exploratory Research 6
for single trajectory
- Hidden Markov Models (HMM)
- Similarity-based retrieval
- Reinforcement Learning
learning algorithms for
agent based modelling
Visual Interfaces for
of solutions and
DART explores the applicability of a collection of machine learning and agent- based
models and algorithms to derive a data-driven trajectory prediction capability.
Advanced visual analytics techniques will be used to facilitate data exploration,
quality assessment, and algorithms parameters and features selection.
7. DART Data Sources
SESAR 2020 - Exploratory Research 7
DDR -2 Spanish
The datasets used for training and testing the algorithms are a key component
of the project.
8. Visually supported Trajectory
SESAR 2020 - Exploratory Research 8
Visually supported detection of clusters of trajectories and flight plans from Madrid to Barcelona
Historical data provide recurrent patterns
of trajectories (enriched with contextual
information – e.g. weather) that data-
driven methods learn.
Prediction of a single trajectory
involves choosing the pattern
that fits better a flight plan
w.r.t. contextual data.
Expected Achievements & Conclusions
Expected Achievements (Maturity: TRL-1)
• Data-driven algorithms for single trajectory predictions;
Implementation and evaluation of the following algorithms:
(a) Hidden Markov Models (HMM),
(b) Similarity-based retrieval (HMM/clustering),
(c) Reinforcement Learning Models applied on Aircraft Intent.
• Collaborative reinforcement learning algorithms for agent based modelling of
ATM; Agents correspond to aircraft in specific trajectories.
Implementation and evaluation of collaborative reinforcement learning algorithms
• Visual Interfaces for Interactive exploration of solutions and decision making,
providing an overview of modeling results in space and time.
SESAR 2020 - Exploratory Research 9
10. This project has received
funding from the SESAR
Joint Undertaking under
the European Union’s
Horizon 2020 research
programme under grant
agreement No [number]
The opinions expressed herein reflect the author’s view only.
Under no circumstances shall the SESAR Joint Undertaking be responsible for any use that may be made of the information
DART. Data-driven AiRcraft Trajectory prediction research
11. Collaborative Trajectory Prediction
• Reinforced Learning approach considered (work in progress):
1. Independent Reinforcement Learners approach
Each agent (flight) is self-interested and
learns by itself to resolve the DCB problem,
by measuring its own reward after each
2. Edge-Based Collaborative Reinforcement Learners approach
Reward received from global state and joint action of all agents (global
3. Agent-based Collaborative Reinforcement Learners approach
Variant of the previous approach. Agents are not flights, but
flights+sectors. Would allow rerouting extension
11International Workshop on Uncertainty and Air Traffic Management
13. Collaborative Trajectory Prediction
• Early results (Method 2, Sparse Collaborative Q-Learning – Agent-
based decomposition – Edge based update):
International Workshop on Uncertainty and Air Traffic Management
Understanding what can be achieved today by data-driven trajectory prediction models, also accounting for ATM network effects.
Improved network management and airport operations due to reduction of uncertainty factors, focusing on delays
Advanced collaborative decision making processes at the pre-tactical stage leading to more efficient ATM procedures
The aim is to develop a collaborative decision making process that would support multi-objective optimization taking the requirements of the different actors in the ATM system into account at the planning phase (i.e. few days before operation):
Aircraft Operators (AOs) : Minimizing the cost thought maximizing the adherence to the airlines preferred FPs.
Network Manager (NM): Decide which flights to modify to resolve sector imbalances and potential conflicts.
Air Navigation Service Providers (ANSPs): Minimizing the sector imbalances and potential conflicts.
Major research issues:
• What are the supporting data required for robust and reliable trajectory predictions?
• What is the potential of data-driven machine learning algorithms to support high-fidelity aircraft trajectory prediction?
• How the complex nature of the ATM system impacts the trajectory predictions?