Trajectory Projection | Short-Horizon Aircraft Path Estimation

How it works

Forward kinematics from the aircraft's own state

Every ADS-B report carries ground speed, track angle and vertical rate. AeroScope integrates those forward to estimate where the aircraft would be a few seconds to a minute from now if it simply kept doing what it is doing, the same dead-reckoning that keeps the live map gliding between fixes, projected ahead of "now."

➡️

Constant-velocity core

A great-circle extrapolation along the current track at the current ground speed, with altitude carried forward by the reported vertical rate.

🌀

Turn awareness

A sustained turn rate observed across recent reports bends the projection, so an aircraft mid-orbit isn't projected straight off its arc.

📐

Growing uncertainty

Confidence narrows close in and widens with time, the further ahead you look, the wider the cone of plausible positions. Short horizons are useful; long ones are not claimed.

What it's for

Anticipation, not prophecy

Geofence pre-warning, flag aircraft on course to enter a zone you've drawn, seconds before they do.
Proximity anticipation, surface traffic closing on your observer point ahead of closest approach.
Smoother display, the same forward integration removes jitter between server fixes.

Honest scope. This is short-horizon physical extrapolation, heading, speed, vertical rate and observed turn. It does not model pilot intent, ATC instructions, weather or destination, and it makes no use of reinforcement learning or trained trajectory networks. Beyond a short horizon, only the widening uncertainty cone is meaningful.

Method

Kinematic projection, and why the cone widens

Projection here is deliberately simple physics rather than a learned predictor. Current position, ground speed, track and vertical rate are integrated forward in time. Because none of those inputs is perfectly known and any of them can change without warning, the projection is drawn as a widening cone rather than a line.

The cone widens for three compounding reasons: measurement uncertainty in the reported values, the aircraft's freedom to manoeuvre, and time. Ten seconds ahead the cone is barely wider than the aircraft. Ten minutes ahead it covers a large area, which is an honest depiction of how little a straight-line assumption is worth at that horizon.

There is no reinforcement learning and no neural network here. An earlier RL prototype was evaluated and retired. Dead reckoning has the advantage that its failure mode is obvious: the moment an aircraft turns, the projection is visibly wrong, rather than confidently wrong.

Reading it

What a projection is good for

Short horizons only. Useful over seconds to a couple of minutes. Beyond that, treat it as a rough area of possibility.
Steady flight only. It assumes the aircraft continues as it is. Aircraft in circuits, orbits or approach will diverge immediately, which is why pattern classification matters alongside it.
Geometry, not intent. A cone intersecting an area means the current vector points that way, nothing more.
Not a conflict-detection system. It must not be used for separation or collision avoidance. See limitations.

FAQ

Frequently asked questions

How far ahead can AeroScope project an aircraft?

Usefully, on the order of seconds to about a minute. The projection is pure forward kinematics, so its uncertainty cone widens with time; short horizons are reliable for anticipating geofence and proximity events, longer ones are not claimed to be accurate.

Does trajectory projection use machine learning or RL?

No. It is deterministic dead-reckoning from the aircraft's reported ground speed, track, vertical rate and observed turn rate. There is no reinforcement learning and no trained trajectory model, an earlier RL prototype was retired across the platform.

Why does the predicted path widen into a cone?

Because small uncertainties in speed and heading compound over time. AeroScope shows confidence narrowing near the aircraft and widening ahead, which is an honest representation of how much a constant-velocity assumption can be trusted.

How does AeroScope predict where an aircraft is going?

By kinematic dead reckoning: current position, ground speed, track and vertical rate are integrated forward in time. The result is drawn as a widening cone rather than a line, because measurement uncertainty and the aircraft's freedom to manoeuvre compound as the horizon extends.

Why is the projection a cone instead of a line?

Because a line would imply a confidence the data does not support. The cone widens with measurement uncertainty, the aircraft's ability to turn at any moment, and elapsed time. Ten seconds ahead it is narrow; ten minutes ahead it honestly covers a large area.

Does trajectory projection use machine learning?

No. It is straightforward physics. An earlier reinforcement-learning prototype was evaluated and retired. Dead reckoning has a useful property: when an aircraft turns, the projection is visibly wrong rather than confidently wrong.

Can the projection be used for collision avoidance?

No. It is a short-horizon situational aid only and must never be used for separation, conflict detection or any safety-of-life purpose. Certified systems exist for that role.