Skip to content

Skycontroller NG firmware image (nap 2.0.0-rc3) notes

Correction. The firmware image analysed here (nap-linux-2.0.0-rc3) is the NG model’s (model id 0x0913), not the Skycontroller 2’s. Wherever this page says “the controller” it means the NG controller, and a mention of the Skycontroller 2 is about the other model, which this image does not describe. See the Skycontroller mapping.

Working notes: facts, questions, references. Not a write-up yet. Source: firmware/skycontroller_nap_update.plf, sha256 dbba796e…fc83 (matches firmware/README.md). Extracted with tools/third_party/extract_plf.py to firmware/extracted/skycontroller-nap/. Static analysis only. Nothing here touched hardware.

Field Value
version 2.0.0
target_platform 6 (same as Bebop 2 image)
target_application 103 (Bebop 2 image is 121)
hardware_compatibility 183847359 (Bebop 2 image: 2241106895)
sections / filesystem entries 1211 / 1209
size 28,997,204 bytes
kernel modules path /lib/modules/3.4.11+/ (Linux 3.4)

Same PLF container and header layout as the Bebop 2 image, so the one extractor handles both. The target_application field is the cheapest way to tell product images apart.

What it runs (EVIDENCE: etc/boxinit.d/*.rc)

Section titled “What it runs (EVIDENCE: etc/boxinit.d/*.rc)”
Service Notes
mppd Main daemon. Links libarsdk, uses both an ARSDK “mux” backend and a “net” backend. Class mpp.
sensorsd-mpp Sensor daemon for the controller itself (class main).
hudd HUD renderer, Qt5 on -platform eglfs (libQt5Gui/Widgets, libMali). Disabled by default.
aoad Android Open Accessory daemon, the USB link to a phone.
wifid (STA) --mode STA --ip 192.168.42.3 --dhclient, chip qca9375. This is the link to the drone.
wifid-ap --mode AP --ip 192.168.44.1, chip wl18xx (TI WiLink). Started when gps.socket.ready=1. Disabled by default.
gpsd, btgpsdemux GPS, including a Bluetooth GPS demux.
adbd Started only when ro.debuggable=1. USB gadget VID 19cf (Parrot), function adb.

The drone side of the Wi-Fi link is 192.168.42.1 (Bebop 2 subnet), so the controller joins the drone’s network as a client at 192.168.42.3.

Protocol surface (EVIDENCE: strings in usr/bin/mppd)

Section titled “Protocol surface (EVIDENCE: strings in usr/bin/mppd)”

mppd carries the ARSDK command descriptors for the whole stack, not just piloting:

  • Ardrone3_*: piloting, camera, media record, picture settings, Piloting_UserTakeOff, Piloting_NavigateHome
  • Skyctrl_*: CoPiloting_SetPilotingSource, AccessPointSettings_WifiSecurity, calibration (MagnetoCalibrationQualityUpdates), SettingsState_ProductSerialChanged
  • Mapper_*: Axis_mapping_item, Map_button_action, Grab, Grab_state, Set_inverted, Reset_mapping, Application_axis_event, Application_button_event
  • Drone_manager_* (e.g. Forget), Follow_me_*, Common_MavlinkState_MavlinkFilePlayingStateChanged
  • transport: arsdk_backend_mux_new, arsdk_backend_net_start_listen, arsdk_publisher_avahi_new, arsdk_discovery_start

So the controller is an ARSDK peer on both sides: an ARSDK “device” toward the phone (over USB mux) and an ARSDK “controller” toward the drone (over Wi-Fi, discovered through Avahi). The Mapper commands mean stick/button to command mapping is configured on the controller and reported to the app as events, rather than the app polling raw sticks.

  • How are stick positions encoded into Ardrone3_Piloting_PCMD frames, and at what rate?
  • Is the controller to drone Wi-Fi link authenticated beyond WPA2 (anything in Drone_manager)?
  • What does sensorsd-mpp read (sticks over ADC/I2C, IMU, compass)? Not looked at yet.
  • CoPiloting_SetPilotingSource: how does piloting authority get handed between controller and app?
  • Why is wifid-ap gated on gps.socket.ready? (looks like a start-order trick, not a feature)
  • Does the Skycontroller carry the same shpoison/debug-hook style button combos as the Bebop 2? No evidence yet.
  • Does the 2.0.0 image contain a Bebop-specific compat table (which drone firmware versions it will pair with)?
  • firmware/extracted/skycontroller-nap/rootfs/etc/boxinit.d/ for service definitions
  • firmware/extracted/skycontroller-nap/rootfs/usr/bin/mppd (153 KB ARM ELF, not stripped of ARSDK symbol names)
  • firmware/extracted/skycontroller-nap/rootfs/usr/bin/{sensorsd-mpp,hudd,aoad,wifid}
  • arsdk-cli is shipped in usr/bin (host-side ARSDK client, useful to read for command syntax)

Stick and button path (added after the stick-path analysis; agent notes in skycontroller-stick-path.md)

Section titled “Stick and button path (added after the stick-path analysis; agent notes in skycontroller-stick-path.md)”

Checked directly against the extracted Skycontroller rootfs unless marked agent-cited.

  • The sticks and buttons are not read by sensorsd-mpp. That daemon links libiio and handles the controller’s own IMU and magnetometer (properties persist.gyro.bias, persist.magneto.bias) and talks to mppd over an abstract Unix socket @sensorsd (both binaries contain the name).
  • The sticks arrive from a USB “bottom board”: lib/firmware/xmega.hex (36,930 bytes) is flashed by usr/bin/bottom_board_update; the MCU is an ATxmega32C4 (verified: bottom_board_update calls dfu-programmer atxmega32c4 with xmega.hex; an agent had read it from the updater first).
  • udev rules in lib/udev/rules.d/: 55-juba-gamepad.rules matches an input device named mpp2_gamepad (and qpnp_pon, the power key) as event* nodes; 56-juba-bb.rules matches hidraw* nodes for USB vendor 03eb (Atmel). libmppd_hw_nap.so (libevdev) and libnapbb.so are the consumers.
  • mppd references per-unit factory properties ro.factory.js.centers and ro.factory.js.min_max (stick centres and ranges). No file in the image sets them, and no user stick-calibration file was found.

Default mapping (etc/mppd/<type>/mapping_bebop_2.cfg)

Section titled “Default mapping (etc/mppd/<type>/mapping_bebop_2.cfg)”

Axes (identical in the file I read): left stick left/right = yaw, left up/down = gaz (throttle), right stick left/right = roll, right up/down = pitch, right mini-stick = camera pan and tilt. The axis action list in the file defines 16 = roll, 17 = pitch, 18 = yaw, 19 = gaz, 20 = camera_pan, 21 = camera_tilt. Expo is on for the four main axes.

mppd picks the directory from ro.factory.type (string /etc/mppd/{ro.factory.type} in mppd; the two directories shipped are bebop and milos). The two Bebop 2 button tables differ:

Action bebop directory milos directory
emergency button Z not bound in the buttons table
center_camera button B button Z
cycle_hud button A button L1
app_0 button L1 button B
return_home, takeoff_land, video_record, take_picture Y, X, C, R1 Y, X, C, R1

In the milos default for the Bebop 2 there is no emergency-stop button in the buttons table (not a hazard note). I parsed only the buttons array. Another binding path, a hard-coded action in mppd, or a user override under /var/lib/mppd/ (agent inference from .cfg.tmp strings) could still provide one. Which directory a given controller uses, and whether the milos one is a given product’s default, is not established.

Command generation (agent-cited from decompiling mppd, not re-derived)

Section titled “Command generation (agent-cited from decompiling mppd, not re-derived)”
  • PCMD is encoded as flag, roll, pitch, yaw, gaz plus a u32 timestamp (ms & 0xffffff) | (seq << 24) (FUN_000227d8); the flag is cleared after 5 consecutive zero-stick ticks (about 250 ms).
  • A 50 ms periodic timer (FUN_00022a14) sends PCMD at 20 Hz. The wire byte layout comes from libarsdk and was not read.
  • Piloting source (FUN_00022aac): source 1 forwards the app’s stored PCMD re-stamped with mppd’s own timestamp and sequence; otherwise the controller’s own sticks are sent. Source 0 is the default and a drone disconnect resets it. No authorization check on the handoff was seen.
  • Failsafe: no controller-side stale-PCMD watchdog, RSSI trigger or link-lost handling was found. INFERENCE: loss-of-link behaviour is entirely on the drone, which is not yet confirmed.
  • HID report layout and the xmega protocol; evdev-to-signed-char scaling.
  • Which mapping directory a given controller selects, and the value of ro.factory.type.
  • The PCMD descriptor in libarsdk; whether the drone times out a repeated stale PCMD.
  • Who sets ro.factory.js.*.

Identity of this image (checked 2026-10-01)

Section titled “Identity of this image (checked 2026-10-01)”

etc/build.prop: ro.parrot.build.project=nap, ro.parrot.build.variant=linux, ro.parrot.build.uid=nap-linux-2.0.0-rc3, ro.parrot.build.version=2.0.0-rc3. The PLF header says version 2.0.0, but the build inside is a release candidate (rc3), so this may not be the shipped release. The image never calls itself “Skycontroller 2”. That label earlier in these notes was inferred from the services (HUD, Android Open Accessory, ARSDK mux). The udev rules mention both nap (03eb:2407) and mpp (03eb:2406) hardware, and the gamepad is named mpp2_gamepad, so which Skycontroller generation this update targets is NOT established. The image alone does not say which generation it targets.

Corrections found by the evidence-ledger check (verified 2026-10-02)

Section titled “Corrections found by the evidence-ledger check (verified 2026-10-02)”
  • The image does name its product: etc/boxinit.rc sets ro.product.model "Skycontroller" with model id 0x0913, and builds the USB product name from the model and the build version (Skycontroller v2.0.0-rc3 in the form the script uses). It never says which generation. Earlier text here saying the image “never names its product” was wrong; the accurate statement is that it never names a generation.
  • The part on the bottom board is an ATxmega32C4: the update script bottom_board_update runs dfu-programmer atxmega32c4 with /lib/firmware/xmega.hex.
  • ro.factory.type, which picks the mapping folder, appears in nap_factory_reader next to the path /customer_prod/type.txt; /customer_prod is an empty folder in the image, so the value lives on the unit.

Comments

Spotted a mistake, or tried this on your own Bebop? Leave a note. Comments are kept on this site's own server.