POINTER_GAZE_MOUSE_MOVE=held, the default (the Gaze page's Mouse movement
switch, with why it's on): while the gaze has the pointer, moving the mouse
does nothing; it moves the pointer only while a button is held, as a
correction. A bumped or drifting mouse can't pull the pointer off what you're
looking at, and every mouse move is a correction, so lessons aren't polluted by
mouse moves to somewhere else. With the gaze stale for a second, in a game, or
with the headset off, the mouse moves the pointer as usual. "free" is the old
behaviour.
Pressing the right button while the left one's press is held back right-clicks
where the pointer is instead (correct with the left, then right-click); both
releases are then nothing. Meta+J then Meta+K stays a drag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Drops the quick check's second step (3629681): the user didn't want it, and it
re-entered itself every tick, re-placing and redrawing the panel, which
flickered in the headset.
POINTER_GAZE_NUDGE_MAX is the one learning limit, now 55 degrees by default:
half of the 109 the Frame shows across. The helper measures the correction
itself (raw gaze to click), not the mouse's path; past the limit it isn't
learned and the helper sends "recheck" for the quick check, for the mouse and
keyboard alike. ft-gazed's own limits (8 and 25) follow the setting.
Test, our tracker's 409 clicks since its Sep 29 calibration: a one-dot check
set from any one of them puts the next 2 minutes' clicks within 15 degrees (99%
within 4.2) and the next 10 minutes' within 25, so it gets well under 55.
The panel's dot is still and its capture ring fills in quarters, so it's drawn
again only a few times per dot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
After the quick check's capture, in gaze mode, the dot stays where it is in the
room (the panel's new "lock") and the gaze has the pointer again. You put the
pointer on the dot as you'd correct a click, with the mouse or Meta+J and the
head. That click clicks nothing: the helper sends it as "calverify", learned
whatever its size. Over POINTER_GAZE_NUDGE_MAX, the capture runs again, up to
3 times. A right click or Meta+K skips it.
POINTER_GAZE_NUDGE_MAX is now the one limit: 15 degrees by default (was 8),
and ft-gazed's own limits (8 for SteamVR's tracker, 25 for ours) follow it. A
keyboard correction past it isn't learned but opens a quick check ("recheck"),
in place of the 30 degree keyboard limit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ft-gazepanel (gaze/panel) is a SteamVR overlay that stays put in your view and
shows dots at known head-relative directions; the gaze service runs it and
drives it (gaze/gazecheck.py):
- a one-dot quick check 3 s after the headset goes on, when our tracker asks
for a click (reseat), at most every 2 minutes, and from Quick check on the
Gaze page; five dots follow if the next 3 lessons are still over 2 degrees off
- the full calibration (the probe's three rounds, dark to bright) when gaze mode
comes on without one, or from Calibrate; quitting it with still no calibration
turns gaze mode off
Each dot takes the gaze once it has held still for 0.6 s; a left click or Meta+J
takes it at once, a right click or Meta+K closes the panel (the pointer helper
hides its dot and passes those presses on while "calpanel" lasts). Our tracker
gets clicks and its own calibration; SteamVR's gets lessons per eye, or a new
calibration.json.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Steam reads the Frame controllers itself, outside SteamVR's bindings, and every
press and release it sees takes SteamVR out of laser mode, so controller clicks
at the gaze can't be done cleanly (docs/gaze-controllers.md, from the gaze-first
branch's tests).
- Gaze precision and Gaze drag take a mouse button or a key combination only;
the controller source, its aim steering, and POINTER_PRECISION_GAIN,
POINTER_PRECISION_DEADZONE and POINTER_GAZE_DRAG_GAIN are gone.
- The gaze actions (including Gaze pointer on/off) can't be mapped to controller
buttons: the relay ignores them there and doesn't ask the helper for those
buttons, and Input Settings no longer offers them.
- Last used wins in gaze mode too: picking up a controller hands it the laser,
as docs/design.md already said.
- The gaze dot shows all the time (POINTER_GAZE_DOT=always, the default;
moving brings back the old behaviour), with a switch on the Gaze page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A GTK 4 / libadwaita app (remote/ft-remote-settings, on the host's own
Python like the gaze probe): remote access on and off (REMOTE, applied at
once when the desktop allows it), the tailnet name and address to connect
to, and the VNC password shown, copied, or replaced. The password stays
random and made on the Frame, in ~/.config/frametop-remote; none is in
the code.
session/remote-ctl.sh starts, stops, and reports remote access; the
session uses it and leaves a remote-capable marker, since KWin allows the
capture only in a desktop that started with REMOTE=1. The password
between krdp and the VNC bridge (local only, but on krdp's command line)
is now new at every start. The installer adds the app to the menu.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Our own eye tracker joins the gaze folder: ft-eyegrab (the root frame
grabber, frametop-eyegrab.service) and ft-eyes, run by the gaze service
when GAZE_TRACKER=own, with its lab tools.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The hand role the pointer takes isn't something to choose by hand. The
helper still reads POINTER_ROLE (right by default) for experiments.
Tried and reverted (not committed): in gaze mode, taking the stylus role
by reconnecting. SteamVR gave the device no role at all (hint 5, role
none) and kept the dashboard laser on the held controllers, so the mouse
couldn't click either. The dashboard laser needs a hand role, and a held
Frame controller takes its hand's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two new actions for mouse buttons, controller buttons, and key
combinations: gaze_precision (hold: the pointer stops where you look and
the button's device steers it, a controller by where it points at
POINTER_PRECISION_GAIN, the mouse by its moves; release: click there) and
gaze_drag (the same with a real press at once, dragging until the
release). The relay sends "precision|gazedrag <source> 1|0" to the
helper, which runs them through the same holds as pinches and grips.
- Key combinations on any keyboard ("key_bindings" in the rules, e.g.
Ctrl+Alt+G for gaze on/off): the last key isn't typed.
- POINTER_GAZE_MOUSE: in gaze mode the left button is a precision button
(the default, as before) or clicks right away (direct).
- In gaze mode a moving controller no longer takes the pointer away.
- POINTER_ROLE (right, left, stylus): the driver takes a "role" command,
so the pointer can stay off the hand holding the precision controller,
which would otherwise take the role back. Needs the rebuilt driver.
- Input Settings: the actions on the Controllers and Buttons pages; on the
Gaze page the mouse choice, the role, the precision sliders, and key
combinations (captured from any keyboard).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hand tracking no longer starts with SteamVR: hands/run.sh install leaves
the units disabled and links hands/ft-handsctl into ~/.local/bin, which
turns it on and off (on | off | status | log | cutouts on|off | gestures).
With it on, hands show through the screens; pinches and grips move the
pointer only with POINTER_HANDS=1, now off by default. ft-camd's service
runs the mono cameras only: while the headset is worn, the colour module
writes just a half-size image into the top-left quarter of its buffers,
which ft-camd can't use yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The palm-down filter is off by default: the user's deliberate pinches,
hand raised, read 0.90-0.99, like typing.
- Typing is told apart by the keyboard instead: the input relay sends the
pointer helper "typing" on key presses, and it takes no pinch within
POINTER_PINCH_TYPING (1 s) of one.
- Grips are still held to hands raised (POINTER_GRIP_BELOW, 0.35 m below
the eyes); pinches aren't, since the user's own sat 0.35-0.45 m below,
elbow resting.
- A grip doesn't begin with the thumb on the index tip: that's a pinch
with the other fingers curled, which was taken for a grip.
- A pinch's point is the index and middle knuckles: the tips' midpoint
moved 1-2 cm as the pinch opened, dragging every release off its press.
- ft-hands --gesture-log prints what the detectors measure, 10 times a
second.
Pinches still aren't reliable enough to use; hand tracking is parked for
now in favour of the controllers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ft-hands tracks with the mono IR cameras in dim light and with every
camera (or the colour pair, HANDS_BRIGHT) in bright light, going by the
colour frames' mean brightness with hysteresis and a 2 s hold
(HANDS_CAMERAS=auto, the default; mono, color and all fix it). Colour
frames are placed on the mono cameras' clock by their dequeue time, and a
view in a camera a step lacks waits for that camera's next frame.
A grip (a closed hand) is a second gesture next to the pinch, in version 2
of the gestures file: every fingertip curled toward the wrist, beginning
only on a hand seen open within a second and held up in front. On the
2026-09-30 lit recording that leaves 6 false grips of 14, all with the
hands on the desk; pinch counts are unchanged. ft-handreplay logs grips
and finger curl, watch_gestures.py shows them, and tools/cut_sets.py
copies a few sets out of a recording.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hand tracking (ft-camd, ft-hands, and their tools) joins the desktop. The
hands file and ring move to /run/user/UID/frametop-hands/, which ft-screens'
hand cutouts now read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Floating windows join the keyboard and the hand cutouts. Floating panels
don't get cutouts yet: their panel and popups show crops of the client
buffer (texture bounds), which the side-by-side cutout buffer doesn't
match. KWin gets both the spare outputs and our input method, and the
pointer helper's frametop. prefix already covers the float panels.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
frame-eyes, a separate project until now, becomes gaze/tracker:
- ft-eyegrab (was fe-bufprobe) copies the eye-camera frames, read-only, out of
SteamVR's eyetracking process. It runs as the system service
frametop-eyegrab.service, which gaze/tracker/install.sh installs to
/etc/frametop with sudo. It keeps only CAP_SYS_PTRACE, CAP_DAC_READ_SEARCH, and
CAP_CHOWN, and copies frames only while /dev/shm/frametop-eyes-want is fresh,
holding none of the tracker's buffers otherwise.
- ft-eyes (was fe-trackd) runs under ft-gazed in the dev container, with
build/venv's pinned numpy and OpenCV: while Eye tracker is Own tracker, or on
the probe's lease ("eyes SECONDS"). No sudo password or fe-live script at
run time any more.
- lab/ holds the research tools (ft-eyes-score, -e2e, -record, -replay,
-session) and findings.md. Recordings live outside the repo, in
~/.local/share/frametop/eyes/captures; .gitignore catches stray frame dumps.
Its socket is now @ft_eyes, its output /dev/shm/frametop-eyes-gaze, and its state
~/.local/state/frametop/gaze/eyes. On practice1 -> practice2 the whole live path
(ft-eyes-e2e) gives 1.30 deg median and 3.18 for the worst tenth, as before the
move (1.30, 3.21).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Programs: ft-camd (the camera broker), ft-hands (the tracker), and
ft-handreplay and ft-ringplay for recordings, built by hands/build.sh
into hands/build/ with one Makefile. The first build fetches ncnn at
frame-hands' pinned tag and builds it with the same options.
- ft-camd gets its privileges from file capabilities (CAP_SYS_PTRACE,
CAP_PERFMON, CAP_DAC_READ_SEARCH) that hands/run.sh install sets with
sudo, and drops them once set up. It still works under sudo. It runs
on the host, linked statically, as frametop-camd.service. ft-hands
runs in the dev container as frametop-hands.service. Both start and
stop with SteamVR.
- Files move to /run/user/UID/frametop/ (cam-ring, hands, gestures),
not $XDG_RUNTIME_DIR, which a terminal in the Frametop desktop has
its own of. SIGUSR1 recordings go to ~/.local/share/frametop/hands.
- The calibration is read through /run/host in the container.
- Settings: HANDS_SWAP_SIDES and HANDS_CPUS in frametop.conf.
- install.sh offers hand tracking as an optional last step.
- The container gets jsoncpp-devel, glibc-static, and NumPy and OpenCV
for the Python tools.
- tools/ring.py reads the ring, and models/NOTICE credits the
Apache-2.0 models.
Checked: ft-handreplay gives identical summaries and byte-identical
depth dumps to frame-hands' fh-replay on both 2026-09-29 recordings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Frametop Input Settings' Gaze page gets two settings, saved in frametop.conf and
read again by ft-gazed when the file changes: Eye tracker (GAZE_TRACKER: SteamVR's
or our own, frame-eyes' fe-trackd) and Eye bias (GAZE_EYE: auto, left, right).
ft-gazed now combines the eyes, each calibrated on its own: SteamVR's set 2 eyes
with the probe's Left eye and Right eye calibrations, or our tracker's eyes as
they come. Without per-eye calibrations, or with --source, it keeps the older
one-source path. A pointer nudge finds its look from the raw gaze the helper
echoes back; with our tracker it goes to fe-trackd as a click.
The bias leans instead of choosing (gazecal.EyeWeights): on 306 live clicks the
eyes' errors partly cancelled, both together 0.65 deg off against 0.96 and 1.11
for either alone. Left or Right counts that eye twice; auto weights each eye by
its RMS miss at its last 20 nudges, since the calibration's fit picked the wrong
eye on SteamVR's test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ft-screens makes a panel for each spare output (frametop.float.N), hidden
until ft-floatd floats a window on it. The panel shows only the window's
rectangle of the buffer at the density of the screen it came from, its
popups and dialogs get small panels over it, pressing its title bar
carries it while KWin's pointer stays put, the corner tab resizes the
window in pixels, and two more buttons close it and put it back on the
desktop. The session adds FLOAT_SLOTS spare outputs to KWin and starts
ft-floatd; ft-layout arranges only the screens' outputs, and the pointer
helper treats the new panels like screens. Not yet tried in the headset.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A head-locked performance overlay kept catching the 3D mouse. It has no
input method, so SteamVR's laser passes through it, but the helper hit
tests every visible overlay with ComputeOverlayIntersection, and the dot
stuck to it whenever it crossed that corner of the view.
POINTER_IGNORE in frametop.conf now lists overlay keys the helper leaves
out of the collision, comma-separated shell patterns, so "vendor.app*"
covers a whole app, including panels it opens later. The laser starts
just before the cursor point, so an ignored panel nearer to you doesn't
catch it either.
Frametop Input Settings has a new Ignored panels page. It asks the
helper for SteamVR's overlays ("overlays", answered from the list's
thread once vrcmd has run again, even while the pointer is off), groups
them by app, and has a checkbox per panel and one for the whole app.
Frametop's own screens aren't offered, since ignoring one would leave
nothing to click the app on with the mouse. Entries for apps that
aren't open are listed so they can be removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A display mount that covers the proximity sensor makes the headset seem
worn, so SteamVR never turned its displays off and they stayed on all
night. The new power service, ft-powerd (frametop-power.service), goes by
use instead: after DISPLAY_OFF_MIN minutes in which the headset and
controllers didn't move and no input device was used, it turns the
backlight off, and the next movement or input turns it back on.
The new Power tab in Frametop Display Settings sets that time and has a
Stay awake while plugged in switch. The switch sets Steam's own "When
Plugged In and Idle -> Sleep after" to Never through Steam's UI, so the
Frame stays reachable remotely while the power button still works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A controller released the 3D mouse's pointer on a single pose sample
faster than 0.35 m/s or 2 rad/s, once the mouse had been still for
500 ms. The helper polls about every 8 ms, so one noisy sample was
enough: a knock on the desk, or a tracking jump when the headset's
cameras pick a resting controller up again (#1).
A controller now has to stay over the limit for 100 ms in a row, and
only samples with a normal tracking result (Running_OK) count. The new
POINTER_CONTROLLER_PICKUP setting (1 by default, 0.5 to 5) scales both
limits; it's a slider on the Pointer page of Frametop Input Settings
and applies live. A controller picked up for real still gets the laser
back through SteamVR's hand role, which follows its touch sensors. The
release log now records the speed and spin that triggered it, to tune
the default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Frame controllers aren't input devices on the host, so the pointer
helper reads them with SteamVR input (vrbuttons.h, actions/), one action
set per button, and sends presses to the relay, which does the mapped
action like for a mouse button. Only mapped buttons are taken, at an
overlay-global priority (SteamVR's experimental "Enable global input from
overlays"), and only outside games unless In games is on. Frametop Input
Settings gets a Controllers page to map them.
Gaze mode: a left press while the gaze has the pointer isn't sent at
once. The pointer stops, you drag it onto what you meant with the button
held, and the release clicks there; a press held still for
POINTER_GAZE_HOLD (0.5 s) becomes a real press, for drags. The dot shows
only while the mouse moves it (POINTER_GAZE_SHOW), while a press is held,
and as a pulse per click. Outside games the pointer stays on while gaze
mode is on. ft-gazed takes a third of each lesson's offset instead of all
of it: in the first live test one 6 degree lesson moved everything and put
the next target 7 degrees off. Frametop Input Settings gets a Gaze page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MAGIC pointing (Zhai et al. 1999), off by default: with POINTER_GAZE=1,
"gaze on", or a button mapped to the new gaze_toggle action, the cursor ray
is the corrected gaze from ft-gazed while the gaze has the pointer. Moving
the mouse takes the pointer from where the gaze left it; looking more than
POINTER_GAZE_RETAKE (5 deg) away with the mouse still gives it back. The
pointer is aimed at the gaze, never steered toward it, and only fresh gaze
moves it, so a stopped service or a blink leaves it where it is. A mouse
nudge of up to POINTER_GAZE_NUDGE_MAX (8 deg) before a click is sent to
ft-gazed as a lesson in the eye tracker's error.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Easing the reference toward the head all the time moved the cursor on
every small head movement, so a cursor parked in a corner of the view
drifted. Now nothing moves while the head stays within the leash. Once
the head has been past it for POINTER_LEASH_DELAY (0.2 s, so a glance
out and back doesn't count), the reference eases all the way to where
you face (POINTER_LEASH_RETURN) and the cursor lands back in its place
in the view, then the leash waits again. Leaning inside the leash no
longer moves the cursor either. Leash 0 stays head-locked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The leash only moved its reference when the head reached the leash's end,
so after turning back it sat up to the leash off, and recentring meant
overshooting with the head. The reference now eases toward the head's
facing (POINTER_LEASH_RETURN, 0.2 s) and never lags more than the leash,
so the cursor settles back to its place in the view once the head stops.
The mouse can now move the cursor up to POINTER_FOLLOW_REACH (70 degrees)
from the middle of the view, up from a fixed 40. Both are sliders on the
Pointer page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Off by default. With POINTER_FOLLOW=1 (a switch on the Pointer page of
Frametop Input Settings, or a mouse button mapped to Head follow on/off),
the cursor rides on a reference direction leashed POINTER_LEASH_DEG (10)
from where you face. Within the leash it stays put in the room; past it,
it turns with your head and keeps its offset. A leash of 0 locks it to
your view. Head roll is ignored, the cursor stays within 40 degrees of
the reference, and it holds still in the room while the left button is
down so the head can't nudge a click or a drag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An ungrabbed keyboard reached both sides at once: gamescope, which in VR
reads every input device itself (the SteamOS build's InputStealer), typed
into its focused app, and ft-screens typed into the desktop. Space in the
desktop paused Spotify on the dashboard, including every space frame-voice
dictated. And while the SteamVR dashboard was open, the desktop got no
keys at all.
Typing now follows the last click. ft-screens sees clicks on its own
screens; the pointer helper reports the panel under the dot on each mouse
press, so a click on any other panel sends typing to Steam. While typing
goes to the desktop and the screens are showing, ft-screens tells the
relay every second, and the relay grabs pass-through keyboards. A grab
waits until no key is down, and the relay lets go if ft-screens stops
reporting. Controller clicks on other panels aren't visible to overlay
apps, so they don't move typing (noted in the README).
A program that reads every keyboard for a hotkey loses a grabbed one.
Repeating the keys on another input device doesn't work, since gamescope
reads that too, so with SHARE_KEYS=1 the relay sends them to
@frametop_keys as datagrams with the device name. It's off by default:
any local process that binds that name first would get every key typed
into the desktop.
The docs now say gamescope reads keyboards and the headset's buttons
itself; they said SteamVR passed keys on to it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A Meta tap on a pass-through keyboard toggled the SteamVR dashboard, which
got in the way of using Meta on its own. It's now off unless
META_DASHBOARD=1 is in ~/.config/frametop.conf. Meta as a modifier
(Meta+Shift+R, Meta+Shift+H) is unaffected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Several KDE Plasma screens floating in SteamVR, each a real monitor of any
resolution and shape, shown by our own compositor (ft-screens), with a
layout, wrist pinning, and visibility modes; a Bluetooth mouse that drives
all of SteamVR as a room-anchored 3D pointer (input relay, ft-pointer
helper, ft_pointer SteamVR driver); two settings apps; and Bluetooth LE
fixes. Installs on the headset with ./install.sh.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
