# Cooked Control¶

Note

It is possible for more than one processor part to control the cooked steering methods at any given time. Whichever processor is the last to update the value of the given control will take priority.

Changed in version v1.1.3: kOS no longer throws an error if multiple parts attempt to set the value of the same control.

For more information, check out the documentation for the SteeringManager structure.

In this style of controlling the craft, you do not steer the craft directly, but instead select a goal direction and let kOS pick the way to steer toward that goal. This method of controlling the craft consists primarily of the following two commands:

## The special LOCK variables for cooked steering¶

LOCK THROTTLE TO expression. // value range [0.0 .. 1.0]

This sets the main throttle of the ship to expression. Where expression is a floating point number between 0.0 and 1.0. A value of 0.0 means the throttle is idle, and a value of 1.0 means the throttle is at maximum. A value of 0.5 means the throttle is at the halfway point, and so on.

The expression used in this statement can be any formula and can call your own user functions. Just make sure it returns a value in the range [0..1].

Warning

Do not use both SAS and lock steering at the same time. See the warning below.

Warning

It’s a very bad idea toWAIT during the execution of the expression in a LOCK THROTTLE. See the note in the next section below.

LOCK STEERING TO expression.

This sets the direction kOS should point the ship where expression is a Vector or a Direction created from a Rotation or Heading:

A Rotation expressed as R(pitch,yaw,roll). Note that pitch, yaw and roll are not based on the horizon, but based on an internal coordinate system used by KSP that is hard to use. Thankfully, you can force the rotation into a sensible frame of reference by adding a rotation to a known direction first.

To select a direction that is 20 degrees off from straight up:

LOCK STEERING TO Up + R(20,0,0).


To select a direction that is due east, aimed at the horizon:

LOCK STEERING TO North + R(0,90,0).


UP and NORTH are the only two predefined rotations.

A heading expressed as HEADING(compass, pitch). This will aim 30 degrees above the horizon, due south:

LOCK STEERING TO HEADING(180, 30).


Vector

Any vector can also be used to lock steering:

LOCK STEERING TO V(100,50,10).


Note that the internal coordinate system for (X,Y,Z) is quite complex to explain. To aim in the opposite of the surface velocity direction:

LOCK STEERING TO (-1) * SHIP:VELOCITY:SURFACE.


The following aims at a vector which is the cross product of velocity and direction down to the SOI planet - in other words, it aims at the “normal” direction to the orbit:

LOCK STEERING TO VCRS(SHIP:VELOCITY:ORBIT, BODY:POSITION).


"kill" string

Steering may also be locked to the special string value of "kill" which tells the steering manager to attempt to stop any vessel rotation, much like the stock SAS’s stability assist mode.

Like all LOCK expressions, the steering and throttle continually update on their own when using this style of control. If you lock your steering to velocity, then as your velocity changes, your steering will change to match it. Unlike with other LOCK expressions, the steering and throttle are special in that the lock expression gets executed automatically all the time in the background, while other LOCK expressions only get executed when you try to read the value of the variable. The reason is that the kOS computer is constantly querying the lock expression multiple times per second as it adjusts the steering and throttle in the background.

Warning

About lock steering and SAS : While kOS had previously supported enabling SAS at the same time as locking steering, this functionality broke when the underlying KSP method was changed in a version upgrade. It is our hope to evenentually restore this functionality. Please check github issue #2117 for updates.

LOCK WHEELTHROTTLE TO expression. // value range [-1.0 .. 1.0]

(For Rovers) This is used to control the throttle that is used when driving a wheeled vehicle on the ground. It is an entirely independent control from the flight throttle used with LOCK THROTTLE above. It is analogous to holding the ‘W’ (value of +1) or ‘S’ (value of -1) key when driving a rover manually under default keybindings.

WHEELTHROTTLE allows you to set a negative value, up to -1.0, while THROTTLE can’t go below zero. A negative value means you are trying to accelerate in reverse.

Unlike trying to drive manually, using WHEELTHROTTLE in kOS does not cause the torque wheels to engage as well. In stock KSP using the ‘W’ or ‘S’ keys on a rover engages both the wheel driving AND the torque wheel rotational power. In kOS those two features are done independently.

The expression used in this statement can be any formula and can call your own user functions. Just make sure it returns a value in the range [0..1].

Warning

It’s a very bad idea to WAIT during the execution of the expression in a LOCK WHEELTHROTTLE. See the note in the next section below.

LOCK WHEELSTEERING TO expression.

(For Rovers) This is used to tell the rover’s cooked steering where to go. The rover’s cooked steering doesn’t use nearly as sophisticated a PID control system as the flight cooked steering does, but it does usually get the job done, as driving has more physical effects that help dampen the steering down automatically.

There are 3 kinds of value understood by WHEELSTEERING:

• GeoCoordinates - If you lock wheelsteering to a GetCoordinates, that will mean the rover will try to steer in whichever compass direction will aim at that location.
• Vessel - If you try to lock wheelsteering to a vessel, that will mean the rover will try to steer in whichever compass direction will aim at that vessel. The vessel being aimed at does not need to be landed. If it is in the sky, the rover will attempt to aim at a location directly underneath it on the ground.
• Scalar Number - If you try to lock wheelsteering to just a plain scalar number, that will mean the rover will try to aim at that compass heading. For example lock wheelsteering to 45. will try to drive the rover northeast.

For more precise control over steering, you can use raw steering to just directly tell the rover to yaw left and right as it drives and that will translate into wheel steering provided the vessel is landed and you have a probe core aiming the right way.

A warning about WHEELSTEERING and vertically mounted probe cores:

If you built your rover in such a way that the probe core controlling it is stack-mounted facing up at the sky when the rover is driving, that will confuse the lock WHEELSTEERING cooked control mechanism. This is a common building pattern for KSP players and it seems to work okay when driving manually, but when driving by a kOS script, the fact that the vessel’s facing is officially pointing up at the sky causes it to get confused. If you notice that your rover tends to drive in the correct direction only when on a flat or slight downslope, but then turns around and around in circles when driving toward the target requires going up a slope, then this may be exactly what’s happening. When it tilted back, the ‘forward’ vector aiming up at the sky started pointing behind it, and the cooked steering thought the rover was aimed in the opposite direction to the way it was really going. To fix this problem, either mount your rover probe core facing the front of the rover, or perform a “control from here” on some forward facing docking port or something like that to get it to stop thinking of the sky as “forward”.

Warning

It’s a very bad idea to WAIT during the execution of the expression in a LOCK WHEELSTEERING. See the note in the next section below.

## Don’t ‘WAIT’ or run slow script code during cooked control calculation¶

Be aware that because LOCK THROTTLE, LOCK STEERING, LOCK WHEELTHROTTLE, and LOCK WHEELSTEERING are actually the highest priority types of triggers that exist in kOS, they cause your expression to be calculated every single physics update tick behind the scenes. So you should not execute a WAIT command in the code that performs the evaluation of the value used in them, as that will effectively cheat the entire script out of the full execution speed it deserves.

For example, if you attempt this:

function get_throttle {
wait 0.001.  // this line is a bad idea.
return 0.5.
}
lock throttle to get_throttle().


Then kOS will attempt to call the WAIT command every single update, as the kOS system keeps trying to re-run the lock throttle expression to learn what you want the new throttle value to be. This will starve your script of the CPU time it deserves, having the effect of running the lock function every-other-tick, and the rest of your code every-other-tick on the ticks in-between. (When the system hits the wait inside the throttle expression, it will stop there, not resuming until the next update, effectively meaning it doesn’t get around to running any of your main-line code until the next tick.)

Again, note that the cooked steering LOCKS mentioned here are the highest priority triggers there are in kOS. That means they can even interrupt other triggers like WHEN/THEN or GUI callbacks. Do not make them call complex functions that take a lot of instructions to return a value, or else you might find that there’s not enough instructions per update left to run the rest of your program effectively.

Normally when you use a LOCK command, the expression is only evaluated when it needs to be by some other part of the script that is trying to read the value. But with these special cooked control locks, remember that the kOS system itself will query the value repeatedly in the background so it knows how to adjust the piloting. Unlike normal LOCKs, these LOCKs will be executed again and again even when you’re not explicitly trying to get their values.

## Unlocking controls¶

If you LOCK the THROTTLE or STEERING, be aware that this prevents the user from manually controlling them. Until they unlock, the manual controls are prevented from working. You can free up the controls by issuing these two commands:

UNLOCK STEERING.
UNLOCK THROTTLE.


When the program ends, these automatically unlock as well, which means that to control a craft you must make sure the program doesn’t end. The moment it ends it lets go of the controls.

## Tuning cooked steering¶

While cooked steering tries to come balanced to perform decently without user interaction, there are some instances where you may need to help tune the behavior. There are a number of settings you can adjust to tweak the behavior of the cooked steering if it’s not performing exactly as you’d like. It may be the case that making your own control mechanism from scratch, while entirely possible with kOS, might be unnecessary if all you really want to do is just make the cooked steering behave slightly differently.

The adjustments described below all come from the SteeringManager structure, which has its own detailed documentation page.

### Some simple suggestions to try fixing common problems¶

If you don’t want to understand the intricate details of the cooked steering system, here’s some quick suggestions for changes to the settings that might help solve some problems, in the list below:

• problem: A large vessel with low torque doesn’t seem to be even trying to rotate very quickly. The controls may be fluctuating around the zero point, but it doesn’t seem to want to even try to turn faster.
• solution: Increase STEERINGMANAGER:MAXSTOPPINGTIME to about 5 or 10 seconds or so. Also, slightly increase STEERINGMANAGER:PITCHPID:KD and STEERINGMANAGER:YAWPID:KD to about 1 or 2 as well to go with it.
• explanation: Once the steering manager gets such a ship rotating at a tiny rate, it stops trying to make it rotate any faster than that because it’s “afraid” of allowing it to obtain a larger momentum than it thinks it could quickly stop. It needs to be told that in this case it’s okay to build up more “seconds worth” of rotational velocity. The reason for increasing the Kd term as well is to tell it to anticipate the need to starting slowing down rotation sooner than it normally would.
• problem: A vessel seems to reasonably come to the desired direction sensibly, but once it’s there the ship vibrates back and forth by about 1 degree or less excessively around the setpoint.
• solution: Increase STEERINGMANAGER:PITCHTS and STEERINGMANAGER:YAWTS to several seconds.
• explanation: Once it’s at the desired orientation and it has mostly zeroed the rotational velocity, it’s trying to hold it there with microadjustments to the controls, and those microadjustments are “too tight”.
• problem: The vessel’s nose seems to be waving slowly back and forth across the set direction, taking too long to center on it, and you notice the control indicators are pushing all the way to the extremes as it does so.
• solution: Increase STEERINGMANAGER:PITCHPID:KD and STEERINGMANGER:YAWPID:KD.
• explanation: The ship is trying to push its rotation rate too high when almost at the setpoint. It needs to anticipate the fact that it is going to reach the desired direction and start slowing down BEFORE it gets there.
• problem: The vessel’s nose seems to be waving slowly back and forth across the set direction, taking too long to center on it, but you notice that the control indicators are NOT pushing all the way to the extremes as it does so. Instead they seem to be staying low in magnitude, wavering around zero and may be getting smaller over time.
• solution: Decrease STEERINGMANAGER:PITCHTS and/or STEERINGMANAGER:YAWTS
• explanation: While larger values for the settling time on the Torque PID controller will help to smooth out spikes in the controls, it also results in a longer time period before the steering comes to a rest at the setpoint (also knows as settling). If you had previously increased the settling time to reduce oscillations, try picking a value half way between the default and the new value you previously selected.

But to understand how to tune the cooked steering in a more complex way than just with that simple list, you first have to understand what a PID controller is, at least a little bit, so you know what the settings you can tweak actually do.

If you don’t know what a PID controller is and want to learn more, you can read numerous descriptions of the concept on the internet that can be found in moments by a web search. If you just want to know a two minute explanation for the sake of tuning the cooked steering a bit, read on.

### Quick and Dirty description of a PID controller¶

You can think of a PID controller as a magic mathematical black box that can learn where to set a control lever in order to achieve a given goal. A good example of this is cruise control on a car. You tell the cruise control what speed you’d like it to maintain, and it attempts to move the accelerator pedal to the necessary position that will maintain that constant speed.

That, in a nutshell is the goal of a PID controller - to perform tasks like that. You have control over a lever or dial of some sort, and it indirectly affects a phenomenon you can measure, and you feed the mathematical black box of the PID controller the measurement of the phenomenon, and obey its instructions of where to set the control lever. Over time, the PID controller, under the assumption that you are obeying its instructions of where to set the control lever, learns how to fine tune its commands about how to set the lever to get the measurement to settle on the value you asked for.

A more complex discussion of PID controllers than that is outside the scope of this document, but you can check out the PID Loop tutorial.

### Cooked Steering’s use of PID controllers¶

kOS’s cooked steering uses two nested PID controllers per axis of rotation:

Seek direction   Current Direction Measurement
|                |
|                |
\|/              \|/
+-seek me---------cur val---+
|                           |
|  Rotational Velocity PID  |
|                           |
+-output--------------------+
desired
rotational
velocity
(i.e. "I'd like to be rotating at 3 degrees per second downward")
|
|
|           Current Rotational Velocity measurement
|                |
|                |
\|/              \|/
+-seek me---------cur val---+
|                           |
|       Torque PID          |
|                           |
+-output--------------------+
desired
control
setting
(i.e. "ship:control:pitch should be -0.2")
|
|
|
|
\|/
Feed this control value to KSP.  (This is the value you can see
on the control indicator meters in the lower-left of the screen).


#### The Rotational Velocity PID¶

The first PID controller looks at the current direction the ship is pointed, versus the direction the ship is meant to be pointed, and uses the offset between the two to decide how to set the desired rotational velocity (rate at which the angle is changing).

The suffixes to SteeringManager allow direct manipulation of the rotational velocity’s PID tuning parameters.

#### The Torque PID¶

But there is no such thing as a lever that directly controls the rotational velocity. What there is, is a lever that directly controls the rotational acceleration. When you pull on the yoke (i.e. hold down the “S” key), you are telling the ship to either rotate faster or slower than it already is.

So given a result from the Rotational Velocity PID, with a desired rotational velocity to seek, the second PID controller takes over, the Torque PID, which uses that information to choose how to set the actual controls themselves (i.e. the WASDQE controls) to accelerate toward that goal rotational velocity.

The suffixes to SteeringManager don’t quite allow direct manipulation of the torque PID tuning parameters Kp, Ki, and Kd, because they are calculated indirectly from the ship’s own attributes. However, there are several suffixes to SteeringManager that allow you to make indirect adjustments to them that are used in calculating the values it uses for Kp, Ki, and Kd.

This technique of using two different PID controllers, the first one telling the second one which seek value to use, and the second one actually being connected to the control “lever”, is one of many ways of dealing with a phenomenon with two levels of indirection from the control.

Keeping the above two things separate, the rotational velocity PID versus the Torque PID, is important in knowing which setting you need to tweak in order to achieve the desired effect.

#### One pair of PID’s per axis of rotation¶

The above pair of controllers is replicated per each of the 3 axes of rotation, for a total of 6 altogether. Some of the settings you can adjust affect all 3 axes together, while others are specific to just one. See the descriptions of each setting carefully to know which is which.

#### Corrects 2 axes first, then the 3rd¶

The cooked steering tries to correct first the pitch and yaw, to aim the rocket at the desired pointing vector, then only after it’s very close to finishing that task does it allow the 3rd axis, the roll axis, to correct itself. This is because if you try correcting all three at the same time, it causes the cooked steering to describe a curved arc toward its destination orientation, rather than rotating straight towards it.

This behavior is correct for rockets with radial symmetry, but is probably a bit wrong for trying to steer an airplane to a new heading while in atmosphere. For flying an airplane to a new heading, it’s still best to make your own control scheme from scratch with raw steering.

#### The settings to change¶

First, you can modify how kOS decides how fast the ship should turn:

// MAXSTOPPINGTIME tells kOS how to calculate the maximum allowable
// angular velocity the Rotational Velocity PID is allowed to output.
// Increasing the value will result in the ship turning
// faster, but it may introduce more overshoot.
// Adjust this setting if you have a small amount of torque on a large mass,
// or if your ship appears to oscillate back and forth rapidly without
// moving towards the target direction.
SET STEERINGMANAGER:MAXSTOPPINGTIME TO 10.

// You can also modify the PID constants that calculate desired angular
// velocity based on angular error, in the angular velocity PID controller.
// Note that changes made directly to the PIDLoop's MINIMUM and MAXIMUM
// suffixes will be overwritten based on the value MAXSTOPPINGTIME, the
// ship's torque and moment of inertia.
// These values will require precision and testing to ensure consistent
// performance.
// Beware of large KD values: Due to the way angular velocity and part
// facing directions are calculated in KSP, it is normal to have small rapid
// fluctuations which may introduce instability in the derivative component.
SET STEERINGMANAGER:PITCHPID:KP TO 0.85.
SET STEERINGMANAGER:PITCHPID:KI TO 0.5.
SET STEERINGMANAGER:PITCHPID:KD TO 0.1.


Second, you can change how the controls are manipulated to achieve the desired angular velocity. This is for the Torque PID mentioned above. Internally, kOS uses the ship’s available torque and moment of inertial to dynamically calculate the PID constants. Then the desired torque is calculated based on the desired angular velocity. The steering controls are then set based on the the percentage the desired torque is of the available torque. You can change the settling time for the torque calculation along each axis:

// Increase the settling time to slow down control reaction time and
// reduce control spikes.  This is helpful in vessels that wobble enough to
// cause fluctuations in the measured angular velocity.
// This is recommended if your ship turns towards the target direction well
// but then oscillates when close to the target direction.
SET STEERINGMANAGER:PITCHTS TO 10.
SET STEERINGMANAGER:ROLLTS TO 5.


If you find that kOS is regularly miscalculating the available torque, you can also define an adjust bias, or factor. Check out these SteeringManager suffixes for more details: PITCHTORQUEADJUST, YAWTORQUEADJUST, ROLLTORQUEADJUST, PITCHTORQUEFACTOR, YAWTORQUEFACTOR, ROLLTORQUEFACTOR