285 Commits
Author SHA1 Message Date
apirroneandClaude Opus 5.5 7e45a4281b pad.status carries the pad's battery, read by padd from a Nintendo pad
The board's kernel has no hid-nintendo, so a Switch pad runs on hid-generic
in its simple HID mode and nothing on the robot knew its battery. Every
subcommand reply carries it, though: byte 2 of an 0x21 report.

- padd asks the connected Nintendo pad on its hidraw node once a minute
  (subcommand 0x50, which changes nothing), and writes the level and the
  charging bit to /run/padd/pad-battery.json. With hid-nintendo present it
  reads that driver's power_supply instead and sends nothing.
- configd adds it to the connected pad in pad.status, while fresh (5 min).
  It cannot ask the pad itself: PrivateDevices leaves no hidraw in its /dev.
- robotctl pad status and robotctl monitor's pad title show it.
- A udev rule gives padd the Nintendo pads' hidraw (group padd, 0660),
  installed by hooks/postinstall on every update; xtask checks it is packaged.

API v65: Pad.battery, optional.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 15:12:19 +09:00
apirroneandClaude Opus 5.5 911c0933e7 Merge main into autonomous-revive: soccer, map and autonomy move to API v58-v60
main took v57 (a join's progress). Type 5 tag writing keeps refusing
MIFARE Classic; pad-map's cfg.skill keeps soccer's bumper kicks.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-13 11:05:06 +01:00
apirroneandClaude Opus 5.5 7bc9fb2e26 Merge autonomous-mode (autod) into autonomous-revive: the brain is API v59
The October brain, brought up to main. Beyond the conflicts:

- The pad's handoff (silent until touched, lets go after pad_idle_s) is
  now pad_map::engage, since main moved padd's mapping into pad-map. A pad
  nobody was touching going away no longer sits the robot down: the brain
  was driving, not the pad.
- autod walks at the pad's [pad_drive] full stick rather than the removed
  robotd_params::drive constants.
- [map]'s template values are commented out like [laser] and [soccer]:
  they never shipped live, so they do not belong in EVER_SHIPPED.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 17:01:35 +01:00
apirroneandClaude Opus 5.5 5076bb6542 Merge maploc-v2 (maploc, mapd) into autonomous-revive: map.* is API v58
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 16:52:01 +01:00
Coralie 15dae7f1df Merge pull request #576 from pollen-robotics/roller-gamepad-mapping
Roller: A and X run no skill; pin the policy set to v10 (swizzle)
2026-11-12 17:26:57 +08:00
Pollen-CocoandClaude Opus 5.5 740bb23447 Pin the policy set to v10: the swizzle roller gait
v10 on the Hub differs from v9 in roller.onnx alone, now the swizzle skating
policy (Mjlab-Velocity-Swizzle-MicroDuck); the manifest and every other file
are v9's, byte for byte. Tested on a robot with `robotctl policy update
--version v10` before pinning.

The pin is a minimum: after the release, the post-install hook moves boards
on our set below v10 up to it, and a fresh board seeds v10.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-12 17:15:54 +08:00
apirrone 1cff8db55d Prompt name: read the SoC serial where configd does, per board
On a fresh beta or face-board-v1, the shell prompt said one duck-xxxx and
BLE (and the setup code) another, until a rename hid it. The prompt
snippet hashed /proc/device-tree/serial-number on every board, but since
34b6e7dc robotd-params reads a beta's serial from /proc/cpuinfo: the beta
U-Boot reads the OTP three bytes early. The prompt was the wrong one.

Pick the source from robotd.toml's [board] version, as soc_serial_at
does: device tree for zero3 (and unset), cpuinfo for beta and
face-board-v1, all-zero cpuinfo rejected. Checked the extracted snippet
against fake files with serial.rs's test vectors for each board.

Assisted-by: Claude:claude-opus-5-5
2026-11-12 10:22:34 +01:00
Pierre Rouanet 9cb079f11d Setup code: the robot's first name as its id, and robotctl to install and show it
The id in a setup code (md1:<id>:<key>) was whatever an operator passed
to btd's --secure-robot-id and the gateway's ROBOT_ID, so nothing on the
robot could print a QR for it. Use the robot's first name instead: the
duck-xxxx default name derived from the SoC serial. It moves from
configd into robotd-params::serial so configd, btd, duck-gateway and
robotctl all derive the same one, and a rename changes none of them.

robotctl system setup-code install reads the 64 hex characters on
stdin, refuses a missing data mount and an existing key without --force,
and commits the file root-only and atomically; it never prints the key.
robotctl system setup-code shows the md1: line and a QR on the terminal.
The robot never generates its key: the station does, per
factory-provisioning.md.

btd --secure-robot-id, duck-gateway --robot-id, ble-gateway.env and
scripts/generate-ble-dev-credential.sh are gone. A btd drop-in still
passing --secure-robot-id fails to start after this; remove the flag.

Assisted-by: Claude:claude-opus-5-5
2026-11-09 15:44:47 +01:00
Pierre Rouanet 723815ddb3 setup-board: let a phone re-pair after it forgot the robot
With pairing required, a phone bonds with the robot. Forgetting the
robot in the phone's Bluetooth settings drops the phone's keys but not
the robot's, and BlueZ's default JustWorksRepairing = never then refuses
the phone's fresh just-works pairing: PIN or Key Missing (0x06) until
the bond is removed on the robot over ssh.

Set JustWorksRepairing = always on every board. The bond is encryption,
not identity, so nothing is given away. Measured on a face-board-v1
beta with an iPhone: refused twice under never, re-paired with no
robot-side step under always.

Assisted-by: Claude:claude-opus-5-5
2026-11-09 15:03:00 +01:00
Thomas Wolf c829879b0f Bound gateway sessions and fix board artifact fixture 2026-11-08 20:55:57 +01:00
Thomas Wolf d27653315f Add opt-in authenticated local media gateway 2026-11-08 20:49:13 +01:00
Thomas Wolf 78635eb321 Document development secure app setup 2026-11-08 20:43:07 +01:00
apirroneandClaude Opus 5.5 b3901367f9 Pin the policy set to v9: the v7 velstand, because v8 was worse
v8's velstand turned out worse than v7's on the robot. v9 on the Hub is v7's
velstand.onnx and manifest.json, byte for byte. v8 stays tagged, since released
daemons pin it.

The pin is a minimum, so a daemon pinned to v8 would still hand v8 to a fresh
board. Raising it means the post-install hook moves v8 boards to v9.

The pad defaults stay as they are: tested with v7's velstand, they feel right.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 00:05:05 +01:00
apirrone 27cdd00a11 Merge remote-tracking branch 'origin/main' into pad-mapping 2026-11-06 16:04:04 +01:00
Pierre Rouanet 6f28f95f16 Merge pull request #533 from pollen-robotics/board-face-board-v1
Add the face-board-v1 board
2026-11-06 07:55:16 +01:00
drenayaz 32703ada6c Merge branch 'main' into pad-mapping 2026-11-05 11:58:17 +01:00
apirrone 6b97b714a0 emotions: ship the Microduck set inside every release
robotd/emotions/ holds the 13 emotions from microduck_emotions (431c73d),
cut by scripts/vendor-emotions.py, and every packaging site includes them as
emotions/ (xtask --include now takes a directory). robotd loads the shipped
set from the release, then [emotions] dir on top: a file there adds one or
replaces the shipped one of the same name. A robot installed or updated to
this build has them with nothing to copy. The simulator reads the
checkout's set.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit f6822d9894)
2026-11-05 11:40:45 +01:00
Rémi Fabre 6f7d600300 robotd: play emotions as skills (robot.do sad)
An emotion is two files in [emotions] dir (default /var/lib/robot/emotions): <name>.json, the
keyframes in the microduck_emotions format, and <name>.wav. robot.do <name> plays it, so it goes
on a pad button like any skill (robotctl pad bind y sad), from every client, with no new call.

While one plays, the loop puts its frame in place of the clients' head, body pose and mouth, with
a zero twist, for that tick: the same policy and safety as the sticks, and the clients' values
back when it ends. Started only on a robot the policy drives with no move in flight; a fall,
limp-fall, Start or a rest ends it and cuts its sound. "skill": "sit" sits the robot down.

The names are listed wherever robot.do's are (the refusal, pad bindings, the subscribe answer
robotctl pad bind checks). scripts/duck-sim reads them from ~/.cache/duck-sim/emotions. Docs: an
Emotions section in the cheat sheet, a row in robotd-design 4.5.

Assisted-by: Claude:claude-opus-5-5
(cherry picked from commit 9bcdc89f6e)
2026-11-05 11:40:45 +01:00
Pierre Rouanet f12becac5a Add the face-board-v1 board
The beta main board with the V1.0 face board (Awinic LED drivers in place of
the PCA9535 expander) was declared "beta", so robotctl showed the main board
it is built from rather than the board it is. It gets its own name,
face-board-v1, hw_rev 3, detected from pollen,microduck-face-board-v1.

Everything decided by the main board (body and head IMU, camera, ToF and NFC
on SPI, SoC serial) goes through Board::is_beta_main_board(), so the new
board behaves exactly like a beta. install.sh and provision-board.sh accept
the name.

Assisted-by: Claude:claude-opus-5-5
2026-11-05 11:38:32 +01:00
apirroneandClaude Opus 5.5 7f93a90364 scripts: a truncated .deb no longer poisons the sysroot cache
`curl -o` wrote straight to the cached path and `[ -s "$deb" ]` accepted any
non-empty file, so a download cut short failed `ar` on every later run until
someone found the file and deleted it.

The cache is now trusted only at the size the package index states, and a
download reaches its final name only once complete. Size rather than SHA256
because `sha256sum` is not on macOS by default and this script runs there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-05 10:11:28 +01:00
Pierre Rouanet b64def7b73 updaterd: automatic rollback is off unless asked for
`[update_gate] auto_rollback` in robotd.toml, default false, a row in
`robotctl configure`. It gates every path that reverts without being
asked:

- the health gate after `update apply`, `select` and `rollback`;
- the boot counter on a trial that never confirmed;
- `robot-boot-check` handing over to `robot-rescue`.

Off, the gate still runs and decides. A release that fails stays live,
reported `Stuck` with `robotctl update rollback <component>` in the
reason; the trial is cleared, and the journal entry is aborted rather
than rolled back, so `known_bad` does not blacklist it. The boot check
names the failed units and the manual `robot-rescue --reboot`, and exits
2. `--dry-run` is unchanged, since the OS slot gate reads it.

Releases now reach robots as tested Yocto images, and a separate daemon
update is the dev-board case, where an unasked revert replacing the
code under whoever holds the robot does more damage than it prevents.
A soft deprecation: the code stays, tested, one setting away.

`updaterd` reads the key from `robotd_config` (new, defaults to
/etc/robot/robotd.toml), both layers, at the moment of deciding, like
`allow_community` and the board. Test fixtures turn it on where they
test what a revert does.

Assisted-by: Claude:claude-opus-5-5
2026-11-04 12:38:25 +01:00
Helge Winkler 6eb8525135 duck-sim: the twin walks — velstand, on the robot with backlash
The simulated duck never took a step. `scripts/duck-sim drive` at 0.15 m/s moved it ~4 cm in 8 s,
and "still upright after walking" was all anyone had checked. Two mismatches with what the gait
was trained on, each confirmed in microduck_rl's own scripts/infer_policy.py, without robotd:

- duck-sim named the pre-v5 alpha_walking + alpha_stand pair. A robot's own defaults
  (deploy/robotd.toml) are walk = velstand.onnx with stand off, and the old pair does not walk in
  MuJoCo at all.
- velstand was trained as Mjlab-VelStand-Rough-Backlash and stands still on the robot without
  backlash, which is what scene.xml includes. On scene_backlash.xml it walks and turns, so that is
  the default scene now.

`drive` now reports how far the duck actually went, from the simulator's trunk position rather
than odometry, and says so when it is under a quarter of what was asked. Measured from a cold start:
0.15 m/s forward gives 0.06; 0.8 rad/s on the spot gives 50 deg/s; 0.1 m/s with 0.5 rad/s gives
24 deg/s. The forward shortfall is still open (the body server has no BAM actuators); the other
built-in scenes carry robots without backlash and do not walk yet. docs/design/simulation.md §11.

Assisted-by: Claude Code <noreply@anthropic.com>
2026-11-01 16:17:56 +01:00
Pierre Rouanet a4b88154e3 Merge remote-tracking branch 'origin/main' into telemetry 2026-10-30 15:50:50 +01:00
Fabien Danieau 36569991a0 telemetry: the microduck PostHog key, and which policy each slot runs
With the project key set, telemetry is live on any robot running this
build with the switch on.

updaterd asks robotd for robot.policies every minute: one `policies`
event per run names what every slot runs, and `policy_loaded` reports
a slot whose policy changed. A policy is `org/name@revision/file` when
it came from the Hub, which is always public since both the official
set and policy.fetch download without a token; a file put on the robot
by hand is only `local`, never its path.

systemd-test.sh keeps its container's updaterd from reporting.

Assisted-by: Claude:claude-opus-5-5
2026-10-30 10:32:07 +01:00
Fabien Danieau 08cdd65191 updater: anonymous usage telemetry to PostHog, with an opt-out
Same design as Reachy Mini's daemon. updaterd sends daemon_started,
update_result (on a version change) and daily_heartbeat to PostHog EU,
with a random robot id, the board, versions and arch. On by default;
telemetry.status / telemetry.set (API v47) are the owner's switch,
routed over BLE and WebRTC, and the console drawer has a checkbox.
Off forgets the id. MICRODUCK_TELEMETRY=0 or DO_NOT_TRACK=1 forces it
off; duck-sim sets the first.

The PostHog key is still empty, so nothing is sent until a microduck
project exists. docs/robot/telemetry.md owns what is and is not sent.

Assisted-by: Claude:claude-opus-5-5
2026-10-30 10:12:01 +01:00
Pierre Rouanet 0357f5976e robot-boot-check: decide once the daemons settle, not 180 s in
The boot check slept until 180 s after boot and looked once. On a Yocto
board every member is up by 16 s, so a healthy boot waited nearly three
minutes for a verdict nothing needed it to wait for — and the OS's own
gate, which copies the deadline, held the OS update page on "rebooting"
for as long.

The timer now starts the script a second in, and it looks every 2 s
until the first of: a member that did not come up (hand over at once;
the predicate is unchanged), every member `active` for 10 s straight
(the release came up — about 25 s in on a Yocto board), or the deadline.

The deadline is `[update_gate] boot_deadline_s` in robotd.toml, 60 s by
default. It is when a boot that never settles is judged, by restart
count, and updaterd's 10 s RestartSec needs about 45 s to show a crash
loop. The script clamps it to 20-300 rather than robotd-params refusing
it, since every daemon reading the file exits on a refused value.
`robotctl configure` says it applies at the next boot.

`--dry-run` now exits 3 where it would have handed over, so
microduck-slot-health can ask the same question instead of copying it.

Assisted-by: Claude:claude-opus-5-5
2026-10-29 17:28:01 +01:00
apirroneandClaude Opus 5.5 6816d1307b robotctl endurance: a battery and heat test that runs until the pack is empty
`sudo robotctl endurance` walks, stands and sits at random within half a
metre of where it started, and every 10 s logs the charge, every thermal
zone, the CPU clock and throttle, servo temperature and current, and
odometry to /var/lib/endurance/<hostname>-<time>/. It runs as the
robotctl-endurance unit so the ssh session can close; `status` and `stop`
manage it, `--dry-run` logs without moving (the idle-drain baseline).
scripts/endurance-plot.py draws a run on the laptop.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-26 12:00:13 +01:00
apirroneandClaude Opus 5.5 8d7ae67848 provision: phase 2 always fetches setup-rkaiq rather than the persisted copy
The persisted /usr/local/sbin/robot-setup-rkaiq on a board predates the
previous commit's fix and stays until a release replaces it, so preferring
it left every board provisioned once before the fix unable to finish a
re-run — the fix would only ever reach boards that did not need it. Fetch
the script and its shim source fresh, as a first run already does; phase 2
needs the network for install.sh regardless.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-17 13:09:57 +02:00
apirroneandClaude Opus 5.5 35a34da23b setup-rkaiq: the persisted copy builds from the shim source it kept
setup-rkaiq.sh persists itself to /usr/local/sbin/robot-setup-rkaiq and
its shim source to /usr/local/lib/rkaiq-modinfo-shim.c, then looks for the
source beside $0 — /usr/local/sbin, where nothing puts it. So the copy it
tells people to re-run, and that provision.sh's phase 2 prefers, died on
every run: a board could be provisioned once and never again. That is what
every re-provision of a coworker's alpha hit on 2026-10-17, before
install.sh, behind a replayed log.

Fall back to the kept source when none sits beside the script, and do not
install that source onto itself when building from it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-17 13:09:09 +02:00
apirroneandClaude Opus 5.5 e3687a1675 provisioning: a re-run reports its own result, not the last attempt's log
A re-run of provision-board.sh on a board whose phase 2 had failed showed
that old failure as its result. The state file outlives a failed phase 2,
so provision.sh skips phase 1 and the reboot and does phase 2 inline, in
the ssh session. The launcher could not tell: it ignores ssh's status
(a phase 1 drops the connection), waited for a reboot that never came,
then streamed /var/lib/robot/provision.log from byte 0 — which only phase 1
empties, so it was the previous attempt's, with its old timestamps. On
2026-10-17 that replayed a fixed 'response too large' from the day before
across a day of re-runs, while the runs themselves resolved 0.17.6.

provision.sh now records how a launched run ended in
/var/lib/robot/provision.outcome, keyed by a run id the launcher passes:
'rebooting' just before the reboot, 'done' in finish, and 'ended <status>'
from an EXIT trap for anything else (signals become exits so dash runs it).
The launcher reads it right after ssh returns: done reports health and
stops, ended dies pointing at the session output, and only a reboot or no
answer goes on to wait and stream. An inline phase 2 also empties the log
first, with a line saying where its output went, so anyone reading it by
hand — or through an older launcher — no longer sees the stale attempt.

The board always fetches provision.sh from the ref, so any board left in
this state is covered by the next run. shellcheck clean; the outcome
recording and the launcher's dispatch were exercised in dash, not yet on
a board.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-17 13:07:50 +02:00
apirrone 936ab005c9 Merge remote-tracking branch 'origin/main' into maploc-scaling 2026-10-16 12:12:53 +02:00
Pierre Rouanet 910b411b4b duckctl: push the Homebrew formula on each release
After the PyPI publish, a homebrew job renders the formula for the version
just released from PyPI's wheels and pushes it to pierre-rouanet/homebrew-tools,
so brew upgrade duckctl follows the releases. It writes with a deploy key
scoped to the tap, held by a homebrew environment.

Assisted-by: Claude:claude-opus-5-5
2026-10-15 12:47:13 +02:00
apirroneandClaude Opus 5.5 f7e2f10949 Merge origin/main into maploc-v2
main took API v42–v44 for the vet book, so map.* moves to v45. netpadd
joins mapd in the release units, the restart order and configd's MANAGED
list (now ten); the Call variant count is 77 with both sides' routes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 10:10:35 +02:00
Pierre Rouanet ad7453d368 Merge main into udp-pad
padd's mapping lives in pad-map on this branch, so main's v8 pad defaults
(vx ±0.4, vyaw ±1.0, bumpers unbound) reach padd through robotd-params and
pad-map's tests now expect them. padd's moved tests are dropped from the
conflict; deploy/robotd.toml takes main's [pad_drive] and keeps [netpad].

Assisted-by: Claude:claude-opus-5-5
2026-10-12 14:52:42 +02:00
apirroneandClaude Opus 5.5 4b47953c0b pad defaults for the v8 velstand: faster drive, slower turn, no kicks; policies pin to v8
The v8 velstand tracks yaw commands, so full stick now asks for what it follows well on the
robot: vx ±0.4 (was ±0.3), vyaw ±1.0 (was ±1.5); vy unchanged. The bumpers ship unbound
instead of kicking — the kick skills stay in the set, `robotctl pad bind lb kick_left` puts
one back. The policy set floor moves to v8 (Cargo.toml and seed-policies.sh).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-12 13:57:37 +02:00
Pierre Rouanet bdd19368c4 Merge branch 'main' into udp-pad 2026-10-12 11:20:10 +02:00
Pierre Rouanet 41a45c4ecb duckctl: one release per branch, beside the daemon
The duckctl builds move into dev.yml and publish onto
`daemon-dev-<branch>`, rather than a `duckctl-dev-<branch>` of their own.
Separate workflows could not share a release, because dev.yml deletes and
recreates it. One workflow can: the daemon job now hands its signed dist
to a publish job that waits for both. Only the daemon is required there,
so a failed duckctl build leaves its file out and never holds back what
boards install.

The updater fetches assets by exact name, so these files sit beside the
board's without being seen. duckctl.yml and the prune case for
`duckctl-dev-` go.

Assisted-by: Claude:claude-opus-5-5
2026-10-12 11:04:07 +02:00
Pierre Rouanet fd5b549b77 duckctl: a prerelease per branch, and a one-line install
Every branch push now publishes `duckctl-dev-<branch>`, beside the
daemon's `daemon-dev-<branch>`, so a branch's client is a download for
anyone without a Rust toolchain or a GitHub login:
`scripts/install-duckctl.sh [branch]` on Linux and macOS, a zip on
Windows. A release of its own rather than assets on the daemon's, which
dev.yml recreates on every push in parallel with this. The updater lists
only its own prefixes, and prune-dev-releases removes these with their
branch.

Assisted-by: Claude:claude-opus-5-5
2026-10-12 10:54:09 +02:00
apirrone 66384dd6ba mapd: the onboard map as a service, map.* (API v42)
mapd subscribes to robot.state and tof.frame like any client, feeds the maploc
mapper on its own thread, keeps the map in /var/lib/mapd/map.json across
reboots, and answers map.status, map.stream, map.grid and map.wipe on
/run/mapd/map.sock. It commands nothing: the robot walks without it, and the
head moves by robotd's own idle glances.

The map file is replaced atomically, saved every autosave_s, when the pose has
moved 30 cm and on shutdown; a file it cannot read is set aside as
map.json.bad-<secs>, never overwritten.

[map] in robotd.toml (on by default); robotctl map / map show / map wipe;
routed over WebRTC by mediad, refused over BLE. Packaged in both release
workflows and dev-push, named in install.sh, configd's unit list and
updater.toml. docs/design/mapping-design.md owns the mechanism.

Checked end to end against a recorded living-room session served over real
sockets: same 21 stops as the offline replay, saved on SIGTERM, and after a
restart relocalized within four stops onto the first run's nodes (2-3 cm).

Assisted-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-08 16:16:02 +02:00
drenayazandClaude Opus 5.5 82ff106a45 Merge origin/main into udp-pad
padd took two changes on main while it was being split into pad-map: the robot sitting down
when a driving pad drops out (#372), and the Home button toggling the flashlight (#375). Both
land in the split shape. The pad-loss state machine and the reading of the robot's answers move
to pad_map::loss so netpadd can share them; Home is pad_map::Buttons bit 12, which was reserved
on the wire and is now the middle button.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-07 17:38:23 +02:00
drenayazandClaude Opus 5.5 a40c706999 netpadd: unit, user and packaging; padd stands down on [netpad] enabled
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-07 11:38:20 +02:00
apirroneandClaude Opus 5.5 a71387476a robotctl: draw the battery black, as it is
The CAD export paints the NP-F970 the same light grey as the shells, so
in the monitor it read as part of the body. The MJCF belongs to the app
repository, so the bake script carries a per-mesh colour override
instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-06 19:17:19 +02:00
apirroneandClaude Opus 5.5 6201412cb7 robotctl: draw the battery in the monitor's 3D view
The bake script listed np_f970 (the NP-F970 pack) among the parts hidden
inside shells, but the battery rides outside the trunk, so the monitor
drew a robot without one. Unhide it and rebake from the same MJCF (a
rebake without the change was byte-identical to the committed blob, so
the diff is only the battery: +1 mesh, +306 triangles).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-06 19:15:41 +02:00
apirroneandClaude Opus 5.5 75496c00d8 Merge nfc-pad-pairing into beta-nfc
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011N2umWfRzHL71umm3VKfc7
2026-10-06 16:00:06 +02:00
Pierre Rouanet 6ecd11b085 Merge branch 'main' into beta-board 2026-10-06 11:10:21 +02:00
apirroneandClaude Opus 5.5 9cd5b3e72a Bring the Rust floor back to 1.89, and keep it under Yocto's 1.94
The Yocto image builds each release from source with the Rust its OE release
ships -- 1.94.1 on wrynose -- and 0.15.1's rust-version of 1.99 (for the
atomics' try_update in mediad) made it unbuildable there: the image had to
seed 0.15.0 instead.

Both call sites become compare-exchange loops, which mean the same thing on
every version and trip neither 1.99's deprecation of fetch_update under CI's
-D warnings nor 1.94's lack of try_update. rust-version returns to 1.89, and
CONTRIBUTING now says the number is a ceiling as well as a floor.

Checked: `cargo +1.94.1 check --workspace --all-targets`, stable clippy
--workspace --all-targets -D warnings, mediad's tests (122 pass).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-05 18:37:02 +02:00
Pierre Rouanet 4ba09e0c6c Declare the board in robotd.toml, and retire one with a single constant
`[board] version` ("zero3" | "beta", unset is zero3) says which electronic
board a robot is built on. It is declared, not detected: provisioning writes
it (`provision-board.sh --board`), `robotctl configure` edits it, and
`robotctl health` prints it and questions it when /proc/device-tree/compatible
names the other board.

The board is the robot's hardware revision for the updater (zero3 1, beta 2),
read from robotd.toml before every check. `hw_rev` in updater.toml applies
only when no board is declared: a beta provisioned against a release whose
template still carried `hw_rev = 1` would otherwise stay revision 1 for good.

Retiring a board is `Board::last_release`. From the release that sets it,
health on that board warns that updates end there; `xtask package` derives
`min_hw_rev` from the table, so the first release after it is refused by the
updater the board already runs. The `--min-hw-rev` flag is gone: nothing
passed it, and it is what the table now answers.

Nothing a beta does differently yet: no audio, camera, ToF or LED change.

Assisted-by: Claude:claude-opus-5-5
2026-10-05 15:42:56 +02:00
apirroneandClaude Opus 5.5 a72beaf6d1 Prepare release 0.15.2
Bump the workspace to 0.15.2 and the policy set floor to v7 (velstand
steers the trunk's roll/pitch while standing; sitstand no longer tips
backward on abrupt head moves, from v6).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-05 15:08:16 +02:00
apirrone 3c88c472a3 robotd: pause the policy while the robot is held ([pickup], off by default)
A policy keeps stepping when the duck is lifted, so a robot picked up mid-walk
thrashes in the hand until somebody presses Start. `[pickup] enabled = true`
runs a small classifier over the last second of what the loop already reads
(gyro, gravity, joint positions/velocities, last commanded target) and pauses
the walking/standing networks while a hand carries the robot: the legs ramp to
the walking policy's own standing stance at the policy gain, and the policy
takes the robot back ~0.2 s after it is set down on its feet.

- duck_control::pickup: feature row, 1 s window, ONNX detector, hysteresis
  latch and pause pose -- the contract with the training code in microduck_rl
  (pickup_detector branch); model vendored at duck-control/models and packaged
  at all three release sites, with a tripwire test
- robotd/src/pickup.rs: the loop's side; watches only walk/stand in walk mode,
  any other owner (skill, sit, limp-fall, disable, shutdown) ends a pause, a
  failing detector lets go; off means nothing is loaded or run
- robotd-params [pickup]: enabled (feature switch in robotctl configure,
  restarts robotd), model, pause/resume thresholds (validated as a band)
- docs: robotd-design.md 2.4.2, cheatsheet "Picking the robot up"

Trained in simulation only. Closed loop in the RL sim: pick-ups detected 100 %
(median 0.20 s), set-downs resumed median 0.18 s, under ~1 false pause per hour
of walking. End to end against duck-body (robotd --sim, velstand, a scripted
hand): three pick-ups paused in 0.22-0.28 s, three set-downs resumed in
0.20-0.24 s, no false pause while walking, standing or turning.

Assisted-by: Claude:claude-opus-5-5
2026-10-03 21:26:09 +02:00
Pierre Rouanet 82bbad2f79 mediad: build on Rust 1.99, which is now the floor
Stable 1.99 deprecates the atomics' `fetch_update` in favour of `try_update`, and CI builds
with `-D warnings` on a floating stable toolchain, so `main` went red on the toolchain bump
alone. Both call sites move to `try_update`, which only exists from 1.99, so `rust-version`
and CONTRIBUTING go to 1.99 with them.

1.99's clippy also flags the one-element `for` over `["create-data-channel"]` in
`open_control_channel`; it is a plain `if` now.

Assisted-by: Claude:claude-opus-5-5
2026-10-01 16:10:14 +02:00