How has the missing information problem been fixed temporarely?

The AI layer provides decisions as maneuvers according to the events the model sends. In this process, the pilot know-how skills are necessarily used. Unfortunately these skills are missing. Nevertheless, to achieve the prototype it is crucial to enable the AI layer to communicate with the model even if it is necessary to set some data randomly. That is the point of what follows.

Figure 1: AI Layer and Model implemented

At this step, the AI layer has been implemented and a model Layer has been implemented too.

Ai Layer:

• Input : events ( angleOFff., aspectAngle, enemySpeed, mySpeed, range, enemyTurnRatem, myTurnRate)
• Ouput : maneuvers ( speed, angleOff, aspect Angle, turnRate)

Model:

• Input: basic actions (turnRate, Acceleration)

• Ouput: basic data (x, y , z, TurnRate, Speed, heading)

To fix this missing information issue, the following idea has been implemented: as long as it is possible, it has been avoided dealing with missing knowledge. The idea was to make the AI output as close as possible to the model input. That is making the maneuvers precise enough to be almost directly passed to the model. The intermediate layer represented by a question mark would only be in charge with converting data by setting some skills randomly if necessary. Its functionalities are the following:

NOTE: ‘t0’ is the current date. ‘t1’ is the date until when the AI process can foresee thanks to the data provided by ‘event(t0)’.

• From Speed(to) and Speed (t1) compute an ‘acceleration’.
• From angleOff (to), angleOff (t1) make a turnRate.
• Converting basicData into Events.

The question should be: how the intermediate layer compute a acceleration from these data?

The initial speed is known (t0), the target speed also (that should be reached at t1). To compute the acceleration(t0) that should be performed and immediately sent to the model the following approximation could be made.

Figure 2: Approximation to compute acceleration

This choice is arbitrary. The problem now is the value of t1 – t0. It represents the amount of time until when the Ai module could foresee. Of course, the value of the acceleration directly depends on it. The shorter this delay is the higher the acceleration will be. To illustrate this point, let us consider the following diagram:

Figure 3: influence of t1 on acceleration

Let assume that t1’ > t1. The acceleration is much higher with speed foreseen at t1 than at t1’. Nevertheless, knowing exactly when an F16 can reach a given speed thanks to the current data of the aircraft (altitude, speed, throttle.) deals with knowledge a pilot learnt through experience. Thus, we do not have these skills. That is why this value has been set randomly. This choice will influence a lot the behavior of the bot. The consequences will be discussed later.