Metron Aviation is in the process of developing
an intent inference algorithm, which is an algorithm
that infers the intent of the pilot of an aircraft
that is being tracked by a surveillance system.
Data describing the environment around the aircraft,
for instance, the location of nearby aircraft,
weather, Navaids, alternate airports, turbulence,
and operational data are used to determine plausible
routes for travel. Operational data and domain
knowledge from pilot and air traffic controller
interviews are used to identify how pilots react
to these elements in the National Airspace System
(NAS). The algorithm imbeds operational data and
domain knowledge into human decision-making computer
models; these models are then used to predict
the future trajectory of the vehicle and to identify
intent. The outputs of the algorithm are an inferred
intent, a level of confidence in the intent, and
a continuous predicted path.
 |
|
|
|
- Refine Knowledge Base and
Intent Models
- Implement a real-time version of the intent algorithm in C++
- Demonstrate accuracy of results and benefits using real-world ETMS data
- Demonstrate accuracy of results and benefits using synthesized ADS-B data
- Demonstrate algorithm to the airlines and air traffic service providers
|
|
 |
|
A novel correlation measure is used to
correlate the observed state data of an aircraft
being tracked with a set of intent models. Furthermore,
the intent models are used to predict the future
motion of the aircraft in a way that standard
prediction techniques, including Kalman filters,
currently do not.
Applied to Enhanced Traffic Management
System (ETMS) data in our preliminary study, the
intent inference algorithm was able to discriminate
between aircraft flying their filed flight plans,
direct-to routes, holding patterns, and air traffic
controller path stretching maneuvers. We demonstrated
how the algorithm predicts the future flight path
given the inferred intent (see Figure below).
Our preliminary study accomplished the following:
Established a set of
domain knowledge based on pilot and controller
interviews
Developed mathematical
intent models for several human decision-making
pilot behaviors, including: follow flight
plan, return to flight plan, fly direct-to
routing, hold altitude, hold heading, and
execute a holding pattern
Demonstrated the utility
of the intent inference algorithm on real-world
ETMS data
