Understanding Betaflight feedforward

Feedforward is the part of the controller that pushes your quad into a move while you are still moving the stick, instead of waiting until it lags behind. Here is how it works, why your rates have a say in it, which settings exist, with their default and range from the source code, how to tell too much from too little and where to find feedforward in a blackbox log. Checked against the docs and source code of Betaflight 2026.6.2. Recommended values are left out on purpose.

Short answer

  • What: feedforward is calculated from the stick movement, more precisely from how fast the setpoint changes. P works from the error, D from the motion of the quad.
  • Effect: the quad follows fast stick moves more directly and with less delay. Hold the stick still and feedforward goes towards zero.
  • Rates: the setpoint comes out of your rates curve. Where the curve is steep, the same stick move makes more feedforward, see the worked example.
  • Setting it: in the App with the “Stick Response” slider in the PID Tuning tab, the details further down under Feedforward. The table lists every setting with its default.
  • Too much, too little: too much brings overshoot, jerks on fast inputs and restless HD video, too little a sluggish response. Details under symptoms.

What feedforward is

The docs page Feed Forward 2.0 puts it like this: “FeedForward (FF) is a drive factor that increases stick responsiveness. It is proportional to the instantaneous derivative, or 'rate of change' of stick movement.” The page dates from Betaflight 4.1 and still uses old setting names like ff_boost; this guide uses the names from 2026.6.2.

How feedforward differs from P and D is in the source code. The setpoint is the turn rate you ask for with the stick, gyro is the one the quad is actually flying. P works from the error, the gap between the two (pid.c). D works from the change of the gyro signal and damps every motion (pid.c). Feedforward works from the change of the setpoint (pid.c, rc.c). In the end all terms are added up (pid.c). The difference in flight: P has to wait until an error builds up, feedforward does not. The PID Tuning Guide in the docs, a freestyle pilot’s notes, says: “And it doesn't wait for any error to develop, the response is instantaneous.”

Back to Feed Forward 2.0: “FeedForward helps P drive the quad into turns. Unlike P, FeedForward cannot cause oscillation, no matter how much FF is added.” The same page says feedforward also cuts the delay between input and response, and less delay means less error. No oscillation does not mean more is always better, though. What too much feedforward does is covered further down.

Feedforward and your rates

Feedforward is tied directly to your rates. In the source code it goes like this (rc.c): the stick position runs through the rates formula, and out comes the setpoint in degrees per second. With every radio packet, Betaflight works out how much the setpoint has changed since the last packet and divides by the time in between (rc.c). That is the basis of feedforward. Move the stick at the same speed, and the slope of your rates curve at that point decides how much feedforward you get.

The table works it out with the Betaflight default and the Actual Rates formula from rc.c: the same 10% stick move three times, at center, halfway and at the edge.

Same stick move, different jump: Betaflight default 70/670/0.00 (Actual Rates)
Stick move (same speed)Setpoint before → afterChangeCompared with center
0% → 10%0 → 13 °/s13 °/s1.0×
45% → 55%153 → 220 °/s67 °/s5.2×
90% → 100%549 → 670 °/s121 °/s9.3×

At the edge the setpoint changes about nine times as much as at center for the same move, and feedforward is bigger there accordingly. The docs describe exactly this: “With Expo and SuperRate applied to your Rates, the rate of increase in Setpoint is greatest just before the sticks stop.” That is what Max Rate Limit is for (see the table): it pulls feedforward back on roll and pitch as the setpoint runs towards the max rate of your rates (rc.c). A higher center value makes the jumps around center bigger, a higher max value mostly the ones at the edge.

Jitter reduction, on the other hand, does not look at the setpoint but at the stick movement itself (rc.c). So it depends on your thumb and your radio, not on your rates.

In practice: new rates also change how feedforward feels. Change both at once and you won’t know afterwards what made the difference. More on the rates themselves: Setting up Betaflight rates and Actual Rates explained.

The settings: default, range, effect

The table lists every feedforward setting in Betaflight 2026.6.2: the CLI name, the field in the App, the default and range from the source code (settings.c, pid.c) and the effect according to the docs. The default is the out-of-the-box value, not a recommendation. What suits your quad depends on the quad and on your radio.

Feedforward settings in Betaflight 2026.6.2: firmware default (not a recommendation) and range from the source code, effect according to the docs
Setting (CLI)In the AppDefaultRangeWhat it does
Strength
simplified_feedforward_gainSource: PID Tuning Tab, simplified_tuning.c“Stick Response” slider, PID Tuning tab100 (App: 1.00)0–200 (App: 0–2.00)More response to fast stick moves. Calculates the F values of all axes from the firmware default while the sliders are on (default)
f_roll, f_pitch, f_yawSource: settings.c, pid.h“Feedforward” column; locked while the sliders control it120 / 125 / 1200–1000 (App field up to 2000)Strength of the feedforward term per axis. According to the App help, low or zero gives a smoother but more delayed response
Fine-tuning (App: Feedforward in the PID controller settings)
feedforward_jitter_factorSource: PID Tuning Tab, 4.3 Tuning Notes“Jitter Reduction”70–20Weakens feedforward when the stick moves slowly, wherever it is; on fast moves it comes through in full and without delay. Higher value: higher threshold, more is weakened
feedforward_smooth_factorSource: PID Tuning Tab, pid_init.c“Smoothness”650–95Smooths the feedforward trace. The docs say the smallest value that gives a smooth trace is best. Since 2025.12 a fixed time, whatever the link rate
feedforward_averagingSource: PID Tuning Tab“Averaging”2_POINT (App: “2 Point”)OFF, 2_POINT, 3_POINT, 4_POINTAverages the last feedforward steps: smoother, but some delay. The docs say it works best on up/down jitter, which is common with faster links
feedforward_boostSource: PID Tuning Tab, rc.c“Boost”150–50Extra push when the stick accelerates or slows down; against the delay of motors spinning up. Roll and pitch only
feedforward_max_rate_limitSource: PID Tuning Tab, rc.c“Max Rate Limit”900–200 (App field up to 150)Pulls feedforward back as the stick heads for full deflection, against overshoot at the start of a flip; does nothing at the end. Lower values start earlier. Works off the max rate of your rates. Roll and pitch only
feedforward_transitionSource: PID Tuning Tab“Transition”0 (App: 0.00)0–100 (App: 0.00–1.00)Weakens feedforward linearly near center stick; 0 turns it off. According to the docs not recommended since 4.3, jitter reduction replaced it
CLI only
feedforward_yaw_hold_gainSource: rc.c, pid.hno field, CLI only150–100Yaw only: an extra term that lingers for a while after a yaw move; 0 turns it off. Replaces boost on yaw since 2025.12. From the source code, no docs page
feedforward_yaw_hold_timeSource: pid.hno field, CLI only10010–250Time constant of that yaw term in milliseconds; higher values decay more slowly (source code)
angle_feedforwardSource: Release Notes 4.5, pid.cno field, CLI only500–200Feedforward in angle mode, where the normal F term on roll and pitch is off. According to the 4.5 release notes it makes angle a lot snappier
angle_feedforward_smoothing_msSource: pid.hno field, CLI only8010–250Smoothing of the angle feedforward as a time constant in milliseconds (source code)

Two things are easy to miss in the App. First: the sliders are on by default, for all three axes (pid.c). The “Stick Response” slider then calculates the F value of each axis from the firmware default (simplified_tuning.c), and the fields in the Feedforward column are locked (PidSubTab.vue). The 4.3 Tuning Notes say the same: numeric entry is blocked while sliders control a value.

Second: smoothing and averaging work differently since Betaflight 2025.12 than in 4.3 to 4.5. Averaging is set to 2 point by default (commit 98934a2). Smoothing is set to 65 and acts as a fixed time, whatever rate your radio link sends at (pid_init.c, commit 67abb0d). The numbers per packet rate in the 4.3 Tuning Notes and in the App’s help text date from before, when the value acted per radio packet.

Too much, too little: the symptoms

According to Feed Forward 2.0, too much feedforward can cause:

  • overshoot at the start of a flip, particularly when the stick hits full travel,
  • an exaggerated response to the steps in the radio signal,
  • amplified shaking when your fingers are cold or you are nervous,
  • spikes in the motor traces and brief shakes with big steps,
  • gyro getting ahead of setpoint.

The PID Tuning tab docs describe the “Stick Response” slider from both sides: too much of it can cause overshoot at the end of a flip or roll. Too little increases the delay with which the quad follows your commands and may result in slow bounceback at the end of a flip or roll. So the two docs pages differ on whether the overshoot shows at the start or at the end.

Restless straight-line flight is another known one. The 4.3 Tuning Notes on the time before jitter reduction: “However, it accentuates even the tiniest of RC inputs, and can make for twitchy flight and jerkiness in HD video.” Since 4.3, the same page says: “Jitter reduction attenuates feedforward strongly when the sticks are moving slowly, regardless of their absolute position.” If the quad still wobbles in HD footage for no obvious reason, the 4.3 Tuning Notes give a simple test: “To test if feed-forward is making it worse, do a test flight with feedforward at zero. If there is no change, feed-forward isn't the problem, and can be used at normal settings.”

Other causes of a twitchy quad, from rates to bent props, are covered in Twitchy quad: causes and fixes. If it shakes after a throttle chop, propwash is the more likely cause.

How well do your rates fit your stick? The free check measures it on your blackbox log. No account, the file stays with you.

Check my log for free

No log yet? See a real example · Turn on logging, with pictures

For your next flying day
  • Betaflight App, Blackbox tab: logging device onboard flash or SD card, logging rate 1 kHz
  • RC Commands, Setpoint and Gyro on, then “Save and reboot”
  • Fly 3 to 5 packs of normal acro
  • Then get the log off the quad and drop it into the free check

Feedforward relies on even stick data. Feed Forward 2.0: “Since RC data comes into the flight controller in 'packets', we actually get a series of sudden steps coming in.” The receiver docs page is more direct: “Feedforward is calculated from the packet-to-packet position difference, and absolutely relies on smooth and regular updates in measured gimbal position.” The same page names the ADC filter of OpenTX and EdgeTX as a source of trouble: among other things it causes spikes and noise in feedforward and must be turned off with Betaflight. You find it as the “ADC filter” checkbox in the radio’s hardware settings.

Fast links have a problem of their own. According to the 4.3 Tuning Notes, at 250 Hz and above there are only a few steps and little time to detect a change. Feedforward then takes small steps of uneven size for heavy shaking. Jitter reduction, smoothing and averaging work against that. The same page says all 500 Hz links and nearly all 250 Hz links benefit from 2 point averaging; since 2025.12 that is the firmware default.

On top of that comes RC smoothing, which smooths your stick data before the controller. The 4.3 Tuning Notes: “We very strongly recommend keeping rc smoothing 'on' at all times”. In auto mode it detects your link’s packet rate on arming and sets the smoothing from it; according to the same page it acts equally on setpoint and feedforward. The source code agrees: with RC smoothing on, the PID controller gets the smoothed feedforward (rc.c), filtered at the same cutoff as the setpoint, which in auto mode follows the measured packet rate (rc.c). By default rc_smoothing is on and rc_smoothing_auto_factor is 30, with a range of 0 to 250 (rx.c). The setting rc_smoothing_feedforward_cutoff that the 4.3 Tuning Notes still mention no longer exists in 2026.6.2 (settings.c).

Angle and horizon mode

Since Betaflight 4.3, feedforward also works in the self-levelling modes. The 4.3 Tuning Notes have a tip for beginners: “If you're a beginner and you find Level too twitchy, reduce the feedforward value in the PIDs.” In 2026.6.2, angle mode works differently: on roll and pitch the code sets the normal F term to zero there (pid.c) and uses angle_feedforward instead, heavily smoothed (pid.c). The 4.5 release notes: “Angle mode is a lot snappier, due to angle_feedforward.” The App 2026.6.2 has no field for angle_feedforward; you set it in the CLI only. In horizon mode the normal F term stays active, because the condition in the code only checks for angle mode.

Feedforward in the blackbox log

Betaflight logs the F term per axis (blackbox.c) and writes your feedforward settings into the log header, the F values as the line ff_weight (blackbox.c). In the Betaflight App’s Blackbox Viewer tab you find them like this:

  1. Pick the traces: under “Graph setup”, put PID Feedforward [roll], Setpoint [roll] and Gyro [roll] into one graph (flightlog_fields_presenter.js). The ready-made example graph “Gyro + PID roll” shows P, I, D, feedforward and gyro together (graph_config.js). The viewer shows the PID terms in percent. How the timeline and zoom work is in How to use Blackbox Explorer.
  2. Read it: the feedforward trace only moves when the setpoint changes, that is when you move the stick, and the faster you move, the bigger the swing. RC Command [roll], your stick before the rates, helps for comparison.
  3. Test flight per the docs: Feed Forward 2.0 says: “Advanced users who can log should do a test flight with fast twitchy inputs, quick left/right stabs, and look for the timing and amount of their FF signal. Ideally P should have little work left to do when the boost/FF components are correct.” For overshoot, the same page says to look at the start of a hard flip.
  4. Judge the smoothing: the PID Tuning tab docs: “Feedforward smoothing is essential to attenuate noise in the feedforward trace. The smallest value that gives a smooth trace is best.” So the trace in the log is the yardstick.
  5. Look closer: with debug_mode = FEEDFORWARD, Betaflight logs intermediate values for one axis (roll by default, gyro_filter_debug_axis). The viewer shows them as “Setpoint Speed”, “Feedforward Boost”, “RC Command Delta”, “Jitter Attenuator” and “Is Duplicate Packet”. debug_mode = FEEDFORWARD_LIMIT shows, among others, “Jitter Attenuation %”, “Feedforward (unsmoothed)” and “Smoothed RX Rate (Hz)” (rc.c, debugModes.js).

How to record a log in the first place: Blackbox logging: setup and downloading your log. Which tools analyze logs besides the viewer: Blackbox analysis: which tool for what.

What fpvrates measures and what it doesn’t

fpvrates does not set feedforward, not even as a suggestion. fpvrates measures your rates from your blackbox log: how much stick travel you really use, where your normal flying ends and how steep your curve gets towards the edge. That very curve helps decide how much feedforward a stick move triggers. The free check shows you this on your own log; the tuning pass adds a rates suggestion. How to set rates: Setting up Betaflight rates.

Sources

As of 13 November 2026, checked against Betaflight 2026.6.2 (firmware and App from 16 September 2026, the latest versions on that day):

Last updated: Nov 14, 2026 · checked with Betaflight 2026.6.2