Skycontroller 2: what the live unit says against what the `nap` image says
Written 2026-10-04. The live column is second-hand: it comes from the skycontroller2 session, which read a live Skycontroller 2 over its shell and on the wire (its notes: skycontroller2-video.md, sections “Measured 2026-10-03”, “Telemetry carried in the relayed stream”, “Controller inputs read live over adb”; raw captures are gitignored there and the pcap holds a GPS position, so do not copy it). Nothing in this file came from a device this session touched. Live readings are Verified on that one unit only. The image column is from firmware/extracted/skycontroller-nap (build nap-linux-2.0.0-rc3, built 2017-08-09), read by this session, with the services report (skycontroller-services-network.md) as Agent-cited support.
Serial, SSID suffix, MACs and Wi-Fi keys are absent from this file by design.
The question
Section titled “The question”Is the nap image the firmware the live Skycontroller 2 runs? Answer: not the same build, and not the same product entry (see the resolution section). Table below first.
| Item | Live unit (second-hand) | nap image (read here) |
Reading |
|---|---|---|---|
| Model string | adb banner ro.product.model=Skycontroller 2 (hand-built CNXN reply) |
etc/boxinit.rc sets ro.product.model to Skycontroller; the string “Skycontroller 2” is not in the image |
Differ |
| Hostname | kernel line Linux mpp |
etc/boxinit.rc sets hostname nap |
Differ |
| Kernel | 3.4.11+, built 2017-10-24 | Kernel blob not yet unpacked here; image build date 2017-08-09 | Live unit is newer than the image (inference: later build, not an older one) |
| Bottom-board gamepad | USB 03eb:2406, name “www.parrot.com MPP GAMEPAD”, evdev mpp-type |
udev rules in the image match both 03eb:2406 (mpp) and 03eb:2407 (nap) |
The unit’s board is the mpp id the image also knows about |
| Video relay | GET /video on TCP 7711 returns HTTP 200 with an SDP body (c=IN IP4 addr:..., m=video <port> RTP/AVP 96, a=rtpmap:96 H264/90000); RTP then flows from the controller’s port 5004 |
usr/bin/mppd contains the same template: GET /video HTTP/1, HTTP/1.1 200 OK, Content-Type: application/x-www-form-urlencoded, c=IN IP4 addr:%s, m=video %d RTP/AVP 96, a=rtpmap:96 H264/90000 |
Agree. The port number is in the image too: mppd builds a sockaddr_in from movw #0x1f1e (htons of 7711) in both builds, and the RTP source port 5004 the same way (#0x8c13). An earlier version of this row said the image did not confirm the port; that was a search method error (ports are byte-swapped immediates) |
| adb | TCP 9050, no AUTH step, shell is uid 0, adbd running | service.adb.tcp.port=9050 in etc/default.nap.prop, ro.debuggable=1, ro.adb.secure set nowhere, adbd has no user line |
Agree; ro.debuggable was not read on the unit, so the image’s debug default is not proved for it |
| FTP | TCP 21 open, banner 220 Operation successful |
inetd serves BusyBox ftpd on 21 and 51, root | Agree on port 21; the greeting text Operation successful is also a string in the image’s BusyBox 1.25.0 (bin/busybox), consistent with the same ftpd, though a banner is not proof of the server (login behaviour on the unit was not tried) |
| RTP | PT 96, H.264 constrained baseline level 4.0, 856x480, about 30 fps; Parrot frame-metadata extension id 0x5032 | not yet examined | Open |
| Stick axes | ABS_X/Y/Z/RX at -100..100, ABS_RY the left above-trigger control; keys 288 to 294 plus 298/299 (scan codes 90001 to 90007, 9000b, 9000c); keys 300..303 and 720..725 fire as axes pass about two thirds of travel | mppd mapping files exist under etc/mppd/{bebop,milos}; the evdev-layer map is not in the image text |
Not comparable yet: the live map is the kernel input layer, not an mppd mapping, and the unit’s active mapping folder was not read |
Resolution from the image’s own tables (Verified, re-derived 2026-10-04)
Section titled “Resolution from the image’s own tables (Verified, re-derived 2026-10-04)”The image’s usr/lib/libarsdk.so carries two tables. The device-type name table (arsdk_device_type_str, jump table decoded for all 23 ids) gives 0x903 SKYCTRL, 0x90f SKYCTRL_2, 0x913 SKYCTRL_NG, 0x915 SKYCTRL_2P, with 0x901 BEBOP and 0x90c BEBOP_2 agreeing with the Bebop images’ own model ids (the control). The update-file table pairs 0x903 with nap_update.plf, 0x90f with mpp_update.plf and 0x915 with mpp2_update.tar. This image announces model id 0x0913 and ships as nap_update.plf. The live Skycontroller 2 reports hostname mpp, model Skycontroller 2 and the mpp gamepad id. Those line up: in the SDK’s own naming a Skycontroller 2 is the mpp platform (0x90f), and the nap image is a different product entry (0x913 SKYCTRL_NG, default SSID prefix nap-ng-). The nap image is therefore not the Skycontroller 2’s firmware. What the NG product is remains open; the hardware agent’s guess (original controller hardware with new firmware) is a recollection and not evidence.
A further consequence, checked in the image: mppd links only libmppd_hw_nap.so, which handles the nap_gamepad and nap_bb types and has no mpp_gamepad; the live Skycontroller 2 enumerates as 03eb:2406, the mpp id. So even by hardware support this image could not drive that unit’s input board. Same conclusion from a second direction.
Consequence for the mapping folders: etc/mppd/{bebop,milos} in this image belong to the NG firmware, where ro.factory.type is read from /customer_prod/type.txt or falls back to a serial-prefix test (agent-read). The earlier line in the public story, that the milos folder “looks like” the second-generation controller’s dropped emergency button, ties that folder to the Skycontroller 2 and is not supported by anything above. Which mapping the live Skycontroller 2 uses is unread (its firmware is the mpp build, which we do not have).
The live unit’s own system volume, copied 2026-10-04 (Verified)
Section titled “The live unit’s own system volume, copied 2026-10-04 (Verified)”With Ryan’s approval the skycontroller2 session copied the system volume (bin, sbin, lib, usr, etc; 963 entries) of the live Skycontroller 2 read-only over adb and checked it (tar sha256 matches what it reported, member count 963, md5 of all 521 regular files equal on unit and workstation). The copy lives in the extracted system volume (gitignored, not in this repo; the capture folder’s recon-1.txt carries the unit’s serial in the kernel command line and must not be quoted). My own identifier scan of the copy: the serial found in zero files (control: the same scan finds a planted copy), no key assignments, and MAC-shaped strings only in vendor Wi-Fi firmware, calibration and test tools (placeholders and vendor defaults, not read as the unit’s). Not copied: /factory, /data, /update, and the kernel; /customer_prod does not exist at the top level of this unit.
Facts read from it, and what they settle:
etc/build.prop:ro.parrot.build.uid=mpp-linux-1.0.7(original uidmpp-linux-1.0.7-rc4), productmpp, built 2017-10-24. So the unit runs mpp-linux 1.0.7, notnap-linux-2.0.0-rc3.etc/default.mppboard.prop:ro.debuggable=1,service.adb.tcp.port=9050. So the debug default the nap image shipped is also this unit’s own, read from its file.etc/mppd/holdsdrone_manager.cfgand four flat mapping files (bebop, bebop_2, chimera, disco). There are nobebop/andmilos/folders on this build, so thero.factory.typeselection seen in the nap image does not exist here.- The Bebop 2 mapping on the unit,
etc/mppd/mapping_bebop_2.cfg, binds no emergency action (action 26 is bound in none of the four files; control: action 17 takeoff/land is bound in each). The nap image’sbebopfolder binds it; itsmilosfolder does not, but themilosfile differs from this unit’s (it names a Z button and a cycle-HUD action this unit’s file does not have). Ryan’s statement that the Skycontroller 2 dropped the emergency button is therefore consistent with the unit’s own file, and themilosfolder was never evidence for it. - The unit’s file list differs from the nap image in hardware: Broadcom bcm43526 Wi-Fi instead of TI and Atheros, a
libmppd_hw_mpphardware library,usbnetd,iosd,macgyverd,usbctrl-mon,memtester, bottom-board firmwarebottomboard.hex(and a.dv.hex); no GPS daemon, no HUD daemon, no dnsmasq or hostapd. Details come from three agent reports in progress.
What this changes
Section titled “What this changes”- The earlier line in
skycontroller-stick-path.mdthat calledmppthe “Skycontroller 1 board” is withdrawn (edited in that file). A live Skycontroller 2 uses themppgamepad id. - Public pages already say the image’s generation is not established; the live readings support that wording and add a reason (model string and hostname differ). They do not say what the
napimage is. - The image’s
mppdis the same video-relay code as the live unit’s, which makes the image useful for understanding how the live unit works even if it is not the same build.
Still unread on the live unit (needs Ryan’s say-so, read-only)
Section titled “Still unread on the live unit (needs Ryan’s say-so, read-only)”ro.parrot.build.version or build uid, ro.factory.type (which mapping folder), ro.product.model.id, ro.debuggable, /sys/kernel/hsis/pcbrev. The skycontroller2 session has no property reader in its adb shell PATH and was approved for a read-only shell for the input work only.
Comments
Spotted a mistake, or tried this on your own Bebop? Leave a note. Comments are kept on this site's own server.