Skip to content

Evidence ledger

The file evidence in each row was re-checked against the extracted files by static means only: file reads, string searches, byte parsing and disassembly listings. No firmware was executed. Rows marked Confirmed independently also rest on readings that another session took from a live aircraft. Some Evidence cells in the motor-board group end with a name tag in code font: id: x names a row that others rely on, and basis: x, y lists the rows this one depends on. When a row changes, search this page for its id to find the rows that need rechecking. A basis means “look there when either row changes”; it does not make the named row the more authoritative one, and when two rows disagree the files decide. Only rows that depend on a derived finding carry a basis; a row that rests on a single file read has none, because re-reading the file is the check. The build checks that every reference in these tags resolves; it cannot check that a row names the right dependencies. A row that rests on one file read says so with basis: none, and in a section that uses tags at all, a checked row with no declaration draws a warning at build time, so “never looked” and “looked, nothing underneath” read differently. Sections with no tags are not covered, and a none can itself go stale if a row later comes to depend on another and nobody revisits it, so a declaration is a statement someone made, not a guarantee of independence. All rows were read through together for consistency with one another on 2026-10-03; that read checks that rows agree, and is not a fresh check of every claim against the files.

Status Meaning
Verified Checked against the file or bytes.
Confirmed independently A second route reached the same answer.
Agent-cited An automated reading that was spot-checked but not fully re-derived.
Corrected An earlier statement that was wrong, with what replaced it.
Open Not established.
Prefix Root
B2 Bebop 2 family 4.7.1 image, bebop-2-4.7.1/rootfs
B1 Bebop 1 4.0.6 image, bebop-1-4.0.6/rootfs
SC Skycontroller image, skycontroller-nap/rootfs
meta metadata.json beside the rootfs folder of the named image
project A file or note kept in the project, not in a Parrot image

Addresses without a prefix are code addresses in the file named in the same cell. Motor-board addresses are offsets in the flat image rebuilt from BLDC.cyacd.

Claim Status Evidence
Three Parrot update images were analysed: Bebop 1 4.0.6, Bebop 2 family 4.7.1 and a Skycontroller image (2.0.0 in its header). All use the PLF container (magic PLF!, header version 13), and one extractor that imports only the Python standard library unpacks all three. Verified version and header_version in the three meta files; magic bytes read from all three .plf files; project tools/third_party/extract_plf.py, import lines 6 to 16.
The header field target_application tells the products apart (104, 121, 103). Section counts are 1183, 1313, 1211; filesystem entries are 1181, 1309, 1209. Verified meta of each image; length of sections.json and manifest.json.
The Bebop 2 root filesystem is about 40 MB. Verified du -sm on B2 returns 40.
Build ids: ardrone3-milos-4.7.1 (Bebop 2), ardrone3-mykonos3-4.0.6 (Bebop 1), nap-linux-2.0.0-rc3 (Skycontroller, a release candidate although its PLF header says 2.0.0). Verified etc/build.prop, key ro.parrot.build.uid, in B2, B1 and SC.
The Bebop 2 image was first unpacked in a container with no network. Agent-cited meta source is /in/fw.plf, consistent with a container; the network setting was not re-checked.
Two partition blobs of the Bebop 2 image are PLF containers with target_application 121; a third, 46,000 bytes, starts with an ARM exception vector table (0xe59ff018 in slots 0 to 4, 6 and 7; 0x00000000 in slot 5). Verified bebop-2-4.7.1/partitions/section-00002-type-12.bin (4,546,428 bytes), section-00004-type-3.bin (3,661,552 bytes), section-00003-type-7.bin.
Both PLF blobs contain a gzip-compressed ARM Linux kernel, 3.4.11+. The type-3 blob’s kernel was built Wed Jun 6 12:02:19 UTC 2018, the type-12 blob’s Tue Dec 1 10:23:38 CET 2015. Verified B2 partitions/section-00004-type-3.bin inner section 1 and section-00002-type-12.bin inner section 1, gzip stream at offset 0x5ae8 in each; the Linux version string in each decompressed payload.
The type-3 blob’s kernel is the one the Bebop 2 boots. Confirmed independently Its build date matches B2 etc/build.prop ro.build.date and the kernel line the live unit logs (3.4.11+, built 2018-06-06); the unit’s log is another session’s reading, not republished.
What the type-12 and type-3 blobs are flashed to or used for. Open The kernels’ presence does not say; the older type-12 kernel’s role is unknown.
The 4.7.1 image is a family image: five board config sets plus common. Verified B2 etc/colibry/: chimeraboard, common, milosblack, milosboard, milosthermal, mykonos3board.
Scripts and dragon-prog branch on the runtime property ro.hardware. Verified B2 etc/init.d/rcS:261, bin/ardrone3_shell.sh:8,13, bin/onoffbutton/shortpress_3.sh:22,39; usr/bin/dragon-prog strings wai_is_chimera, wai_is_milos, wai_is_mykonos3.
The live Bebop 2 unit loads the milosboard config set, with include path milosboard/default. Confirmed independently The unit’s own log file states “My hardware is milosboard” and the config path it read; reported by another session, one unit, not republished. No file in B2 assigns ro.hardware, and etc/build.prop states only ro.parrot.build.variant=milos.
Other Bebop 2 units load milosboard too. Open Only one unit was checked.
A machine that identifies as Milos also logs Mykonos3 board : exporting HSIS and init done lines. The kernel carries those two messages with the literal text “Mykonos3 board”, and a separate Milos board : exporting GPIOs line. Verified The decompressed type-3 kernel (see the kernel row above): the four strings. That the HSIS and init-done messages print on every board is an inference; the kernel code was not disassembled.
/sys/kernel/hsis/ is exported by the kernel, not by a module. The root filesystem reads hwrev, pcbrev, user_on_off and camera pin entries from it. Verified B2 etc/boxinit.rc:23-24, etc/init.d/rcS:22, etc/init.d/rcS_vision, etc/init.d/rcS_video; no module under B2 lib/modules contains the string hsis.
The Bebop 1 image ships only common, milosboard and mykonos3board, with build variant mykonos3. motorsRpmMax is 12200 in milosboard, milosblack and milosthermal, 11000 in mykonos3board, 7100 in chimeraboard. Verified B1 etc/colibry/, etc/build.prop; B2 etc/colibry/<set>/drone.cfg:23.
videoTiltRange is [-75, 30] in common and [-70, 35] in mykonos3board; no other set overrides it. The tilt range of a Bebop 2 is not settled. Open B2 etc/colibry/common/video.cfg:8, etc/colibry/mykonos3board/video.cfg:5; a value reported by a live unit matches neither file and may be a different quantity.
Claim Status Evidence
No file named libColibry*, libNavdata* or libHAL* exists in the Bebop 2 rootfs; dragon-prog reads the config sets at runtime. Verified find over B2 returns only etc/colibry and bin/colibrySend.sh; usr/bin/dragon-prog strings /etc/colibry/common/root.cfg and /etc/colibry/%s/root.cfg.
The estimator and controller are compiled into dragon-prog. Agent-cited Inferred from the missing libraries; dragon-prog holds 11 ControlGainsPID string hits.
libCondor.so exports condor::Prediction::shift_image, getPrediction and condor::Engine:: setBuffer, Compute, getTranslation, getIndicator, setAttitude, setCamera, setCamera2gc, setCameraGain; dragon-prog links it. Verified B2 usr/lib/libCondor.so, nm -D; readelf -d usr/bin/dragon-prog.
libCondor.so is the image engine for the downward camera and not the flight controller. Agent-cited Reading of the export list and the camera notes; setAttitude takes six floats.
The estimator is an 18-state Kalman filter inside dragon-prog. Agent-cited Function addresses FUN_0037d884, FUN_0037c0f8, FUN_0037a16c in the project note colibry-fusion.md.
The ordering of the 18 states. Open Not recovered.
adcFsSet is 198.3642578125 Hz in the default and fvt working modes and 200.0 in simulator. Verified B2 etc/colibry/milosboard/root.cfg:8-13.
Attitude gains KP, KI, KD: pitch (theta) 264.7408, 264.7408, 33.0926; roll (phi) 242.5376, 242.5376, 30.3349. Verified B2 etc/colibry/milosboard/controller.cfg:11-13,15-17.
Altitude “Default” gains kp, ki, kd are 4.170, 1.158, 5.800 and “Freefall” are 2.0, 0.0, 15.01; position gains are 0.8, 0.0, 1.6; speed gains are 0.2612, 0.1606, 0.0. Verified B2 etc/colibry/common/controller.cfg:75-77,80-82,124-126,146-148.
Mass 0.500 kg; inertia Jx, Jy, Jz 0.001491, 0.001458, 0.003328 kg m^2; motor range 3000 to 12200 RPM. Verified B2 etc/colibry/milosboard/drone.cfg:18-24.
motorA is 2.15e-8 N per RPM squared on all motors; motorB is 2.85e-10 N m per RPM squared with signs (-, +, -, +); Kv 1630 rpm per volt; PWM maximum 255; motor (x, y) in metres are m0 (+0.0875, -0.115), m1 (+0.0875, +0.115), m2 (-0.0875, +0.115), m3 (-0.0875, -0.115). Verified B2 etc/colibry/milosboard/drone.cfg, keys motorA, motorB, motorX, motorY, motorsKv, motorsPwmMax.
Maximum thrust is 12.8 N (2.61 times the 4.9 N weight) and hover needs about 7552 RPM per motor. Verified Computed from the constants above: 4 x 2.15e-8 x 12200^2 = 12.80 N; sqrt(4.905 / (4 x 2.15e-8)) = 7552.
User limits: tilt 15 degrees, rotation speed 180 deg/s, vertical speed 1.0 m/s, maximum vertical speed 6.0 m/s; reference tilt limit 35 degrees. Verified B2 etc/colibry/common/reference.cfg:7,25,26,113,140.
The classes QuadMixRPMController, BLDCMotorController, HeightLoopController, QuadQuaternionLoopController exist; dragon-prog holds 60 mangled ControlState strings (for example Flying, Hovering, Flip, PilotedPOI). Verified B2 usr/bin/dragon-prog, strings hit counts 1, 4, 1, 1 and 60.
The mixer is model-based control allocation over six inputs, not a fixed plus/minus table; under saturation channels are given up in the order altitude correction, yaw, tilt thrust, pitch, roll, hover term, capped at 99 iterations. Agent-cited Project note mixer-input-roles.md; process function 0x2a6e98, matrix builder 0x2a78d0.
Clamped RPMs are truncated to 16 bits by BLDCMotorController (0x4177ec) and passed to HAL_send_motor_command (0x43d060), called from ACQCON_colibry_Thread_Handler (0x4e12c). About 30 Mykonos3::ControlState classes cover take-off, landing, flip, freefall, cut-out, flyaway and emergency landing. Agent-cited Project notes control-law.md and mixer-input-roles.md; addresses not re-derived.
Roll torque signs are (+, -, -, +) and pitch torque signs are (+, +, -, -) over motors 0 to 3. Verified Recomputed from motorX and motorY with torque = r x F in a z-down frame with thrust along -z.
The yaw sign convention, which end is +x (nose or tail), and whether anything turns the attitude loop off. Open Arithmetic does not fix the first two; the agent reading of yaw depends on the builder’s convention; the last two were not determined.
Three short presses on the power button stop the stock flight program and start the arducopter service; ArduCopter runs only if a binary exists at the user-data path, and the image carries no /usr/bin/arducopter. Verified B2 bin/onoffbutton/shortpress_3.sh (calls kk, then prestart arducopter); usr/bin/arducopter.sh:4-12.
Four presses run shpoison_cli --start_debug; the telnet and adb service files are marked disabled; the telnet login script is exec /bin/sh -l. Verified B2 bin/onoffbutton/shortpress_4.sh; etc/boxinit.d/50-telnetd.rc, 50-adb.rc; bin/login.sh.
--start_debug leads to pstart telnetd and pstart adbd, and neither state survives a reboot. Agent-cited B2 bin/asix_setup.sh:62-63 starts both services; the call chain from shpoison_cli to that script was not traced.
The only file naming UDP ports 14550 and 14551 is the disabled arducopter service; stock dragon-prog contains neither number. Verified B2 etc/boxinit.d/50-arducopter.rc; grep -c on usr/bin/dragon-prog returns 0.
One Bebop 2 unit’s reported limits equal common config values for six keys that milosboard does not define: maximum tilt 35 degrees, maximum vertical speed 6 m/s (default 1 m/s), maximum pitch/roll rotation speed 300 deg/s, maximum distance 2000 m, geofence off. Confirmed independently B2 etc/colibry/common/reference.cfg:7,19,113,140 and common/event.cfg:70-71; the live values were reported by another session from the aircraft’s own state and are not republished. The field-to-key mapping is an inference from names and units.
Not every reported range comes from these keys: the reported maximum altitude is 150 m (config default usrConfigMaxAltitude 1000.0) and the yaw rotation speed range is 10 to 200 deg/s (config usrConfigRotationSpeed 180.0). Confirmed independently B2 etc/colibry/common/reference.cfg:26,139 (read by me); the reported ranges come from another session’s live read. Where the reported ranges originate is Open.
The reported camera tilt range of -83 to +17 degrees is the virtual gimbal range, not videoTiltRange. Which setpoint videoTiltRange governs is Open. Confirmed independently Another session sent a camera orientation command and read the same tilt value back from the aircraft; B2 etc/colibry/common/video.cfg:8 holds [-75.0, 30.0].
Claim Status Evidence
The motors are reached over I2C on /dev/i2c-1, not over a UART. Verified B2 lib/firmware/BLDC_Factory_Tests.sh:12; string /dev/i2c-1 in usr/bin/BLDC_Test_Bench. basis: none
The motor board answers at 7-bit I2C address 8. Confirmed independently The factory script checks i2c_cmd output for 0x8 (B2 lib/firmware/BLDC_Factory_Tests.sh:13,69,77); ArduPilot’s driver and hardware definition independently use 0x08 (RCOutput_Bebop.cpp:22, hwdef/bebop/hwdef.dat); the flight program’s HAL sets the same slave address (earlier analysis). basis: none
The board firmware ships as one Cypress .cyacd image with header line 1E0680690300 (silicon id 0x1E068069, revision 03, checksum type 00). Bebop 2: 46,917 bytes, 89 rows of 256 bytes, 22,784 bytes of payload, rows 0x19 to 0x7F. Bebop 1: 86 rows, 22,016 bytes, rows 0x18 to 0x7F. Verified Parsed from B2 and B1 lib/firmware/BLDC.cyacd. basis: none
Every row checksum is the two’s complement of the byte sum, 89 of 89 and 86 of 86; none of the 85 row addresses shared by the two images holds identical bytes. Verified Same files, recomputed and compared row by row. basis: none
Application entry is an LJMP at 0x1900 to 0x3EF3 (Bebop 2) and at 0x1800 to 0x3B88 (Bebop 1); the metadata row at 0x7FC0 repeats the entry address and begins with 0x8D or 0xAE. Verified Flat image rebuilt from the rows: bytes 02 3e f3 and 02 3b 88; bytes at 0x7FC0. basis: none
The core is an 8051 and the family is PSoC 3. Open 8051 behaviour is verified (C51 switch table, movx, djnz); the family rests on the 0x1E silicon-id prefix, and the exact part is not mapped.
BLDC_Host_Bootloader prints bootloader version, silicon id and revision, contains BLDC_VerifyCrcAppli, and prints a rejection message for bootloader versions below v3.30. Verified B2 usr/bin/BLDC_Host_Bootloader, strings. basis: none
The host bootloader accepts checksum type 0 or 1; its frame is 01 cmd len_lo len_hi data chk_lo chk_hi 17 with command codes 0x38, 0x32, 0x37, 0x39, 0x3A, 0x31, 0x3B. Agent-cited Project note bldc-decompile.md; matches Cypress’s published stock bootloader format as recalled, not re-derived. basis: none
BLDC_Test_Bench contains exactly two eor instructions, at 0x18424 and 0x189f0. Each sits in a byte loop that starts from an accumulator of 0; the first stores the result after the payload and the second compares a received reply against its trailing byte. The I2C setup issues ioctl 0x706 (I2C_SLAVE_FORCE) at 0x1c1b0 and 0x707 (I2C_RDWR) at 0x1c518. Verified llvm-objdump -d on B2 usr/bin/BLDC_Test_Bench, 0x183f8 to 0x18438, 0x189c8 to 0x18a04, movw r1, #0x706 and #0x707. id: host-tool-xor; basis: none
The only integrity check on the host-tool link is an 8-bit XOR; the tool’s error text nevertheless says “CRC error”. Verified No other eor and no teq in the listing; no crc name among dynamic symbols; the only CRC string is get_obs_speed(): CRC error: Obtained: %u Computed:%u. basis: host-tool-xor
The host tool can send 14 command bytes: 0x02, 0x06, 0x20, 0x2B, 0x40, 0x4D, 0x60, 0x80, 0x82, 0x8D, 0xA0, 0xC0, 0xE0, 0xFA. Frames have one command byte, no length byte and no terminator; 16-bit values are big-endian. Agent-cited Project note bldc-decompile.md; the MTB_BLDC_* routine names and the string BLDC returned an empty message are present. basis: none
The Bebop 2 firmware dispatch table lists ten command bytes (0x02, 0x20, 0x40, 0x4D, 0x60, 0x80, 0x82, 0x8D, 0xA0, 0xFA) and has no entry for 0x06, 0x2B, 0xC0 or 0xE0; unmatched bytes go to the default at 0x23A2. Verified Flat image: inline table after the C51 switch call at 0x1fc1, entries (address high, address low, command), terminator 00 00, default word 23 a2; 0x23a2 is clr 0x20.5. id: dispatch-table; basis: none
get_info writes version bytes 6, 7, R, 4 (Bebop 2) and 6, 3, R, 4 (Bebop 1), read as 6.7.R.4 and 6.3.R.4; no ASCII version string exists. Verified Flat image: 76 06 .. 76 07 .. 76 52 .. 76 04 at 0x220e (Bebop 2) and the 6, 3 variant at 0x2163 (Bebop 1); no 6.7.R string in either image. id: board-version; basis: none
A live Bebop 2 reports the same motor-controller version string through its normal protocol, so that unit runs the firmware the image ships. Confirmed independently Image bytes above (B2 lib/firmware/BLDC.cyacd) against the MotorSoftwareVersionChanged state message observed by the project; no I2C traffic involved. basis: board-version
The I2C-loss watchdog default is 1024 ticks (0x0400), a reload of 0 disables it, and its only call site (0x4a74) is inside the closed-loop routine. Verified Flat image: 0x53b7 loads 0x0A:0x0B with 04:00; routine 0x541f skips on zero; byte search for 12 54 1f finds only 0x4a74. id: watchdog-default; basis: none
On expiry the watchdog raises error 4 and does not reload the countdown; the raise routine ignores a new error that is not strictly greater than the current one; the closed-loop exit test leaves the loop only for a stop request or error 2 to 5. So while an error 8 or 9 is latched (the only errors above 4 that can be latched in closed loop, see below), an expiry cannot raise 4 and the loop continues until the condition clears, after which the next iteration raises 4. Verified 0x541f calls 0x572e with r7=4 and returns without reloading; 0x572e compares xdata 2 with r7; exit test at 0x4a47 to 0x4a5c, watchdog call at 0x4a74 after it. The error byte is a ratchet: the clear tests for 8 (0x4719) and 9 (0x4a61) are guarded by a compare against their own number. id: watchdog-suppression; basis: watchdog-default, start-gate, errors-start-up-only, error-8-clear, error-9-debounce
Dispatch sets the activity bit 0x20.5, cleared only at 0x2001 and at the unknown-command default 0x23a2; the copy to 0x20.4 is at 0x23b0; the exits of the 0x20 and 0xA0 handlers jump to 0x23a4, which falls through to that copy. Verified Byte search for d2 05, c2 05; ljmp 0x23a4 at 0x2171 and 0x2293; listing 0x23a2 to 0x23b4. id: activity-bit; basis: dispatch-table
A read-only poll (0x20 or 0xA0) feeds the watchdog. Verified The single exit of each poll handler (ljmp 0x23a4 at 0x2171 and 0x2293) falls through to the copy that credits the watchdog, and neither handler (0x2088 to 0x219c, 0x2200 to 0x2296) contains a clr 0x20.5. Other exit forms such as returns, and the other commands, were not enumerated. id: poll-feeds-watchdog; basis: dispatch-table, activity-bit
Over-temperature on the Bebop 2 raises error 8 at 95 C or above, or at or below -10 C, and error 8 does not leave closed loop. Verified 0x46ca to 0x4714: xrl #0x80, subb #0xdf, subb #0x76, mov r7,#8, lcall 0x572e; exit test as above. The basis link is for the second half of the claim only: that error 8 does not leave closed loop rests on the exit test described in the watchdog suppression row; the 95 C and -10 C thresholds do not depend on it. basis: watchdog-suppression
The over-temperature action is limited to the error and a sound. Agent-cited Project note bldc-cyacd.md; only the sampling routine was read. basis: none
Error 8 clears at the first temperature sample inside the window (above -10 C and below 95 C); there is no counter, timer or hysteresis. Verified 0x46fd to 0x4733: clear guarded by a compare against 8 at 0x4719; the later compares against 90 C and -4 C cannot change the result. The sampling interval (once per 1000 ticks) is agent-read. id: error-8-clear; basis: none
Whether any interrupt routine stops the motors on over-temperature. Open The fault interrupt at 0x3aa3 was read and does not touch temperature (it raises errors 2, 5 and 11 from fault bits); the other interrupt routines were not read.
The duration of one tick (timer reload 0xF9; the application never writes the clock dividers) and so the real watchdog timeout; whether the flight program depends on this watchdog; whether any hardware watchdog exists. Open “About one second” holds only if a tick is 1 ms, an assumption supported by the host tool’s “cycles x 1 ms” wording. The flight program does not send the timeout command 0x8D (see below), so the board runs on its default.
Reply sizes (get_info 14 bytes, 0x20 reply 15 bytes), a fixed 11-byte set_ref_speed frame, errors 1 to 11 each with a raise site (7, 12, 13 never raised), state 3 never set, hard-coded limits (reference clamp 1500 to 15000, slew 500 per tick, duty clamp 16 to 255, 3S battery only), and a Bebop 2 spin-up of about 200 ticks of alignment, a 600-tick quadratic ramp from 167 to 2500, then a 200-tick lock. Agent-cited Project note bldc-cyacd.md. basis: none
PWM frequency, motor timing, divider resistors and PID gains of the motor board. Open Not recovered.
gpio199 is written by both factory scripts and named in test-BLDC; the factory battery test uses a 9.5 V reference (VREF=95, decivolts). Verified B2 lib/firmware/autotest-mosfet.sh:22,23,52; lib/firmware/BLDC_Factory_Tests.sh:35; string in usr/bin/test-BLDC. basis: none
gpio199 is the motor-board reset line. Verified Settled by the kernel pin table, where the field named reset_psoc holds 199; see the Pin map group. The earlier “assumption from usage” status is retired. basis: reset-psoc-pin
test-BLDC rejects per-motor speeds above 16000; this is an argument bound and not a flight limit. Verified Strings motor1_speed <= 16000 to motor4_speed <= 16000 in B2 usr/bin/test-BLDC; flight range in etc/colibry/milosboard/drone.cfg:23-24. basis: none
The words “hall” and “back-EMF” or “zero cross” occur in BLDC_Test_Bench (2 and 3 string hits); the host tool describes a 45-entry settings table, 7 status names and 14 error names. Agent-cited strings -ci confirms the word hits; counts and names from project note bldc-decompile.md. How the board uses the words is not known, and the shipped board image has no handler for the settings command. basis: none
The motor board’s I2C address, its set of command bytes, the XOR check and the per-board maximum speeds agree with the independent Bebop driver in ArduPilot. Confirmed independently libraries/AP_HAL_Linux/RCOutput_Bebop.cpp and .h in ArduPilot (commit 4c98c92), read against the host-tool decode and B2 etc/colibry/*/drone.cfg. No frame layouts are reproduced here. basis: none
The set-speed frame’s pad byte (frame byte 9) is never read by the motor board: its value does not influence the watchdog, the error-3 path or any branch; it only takes part in the XOR check. Verified Flat image 0x1fe6 to 0x2085: a ten-byte XOR loop over idata 0x65 to 0x6e into 0x6f, speeds read from 0x66 to 0x6d; no other access to 0x6e found in the application; the receive interrupt (0x2b31) stores bytes straight into the buffer. Scoped to the 4.7.1 image; Bebop 1 has the same structure. id: pad-ignored; basis: dispatch-table
The board does not check the length of a set-speed frame: the dispatcher and the handler go straight into the ten-byte XOR loop, and the receive interrupt calls the dispatcher without comparing a byte count. Verified 0x1fb5 to 0x1ffa; the dispatcher call at 0x2bac inside 0x2b31, reached when status register 0x01fc reads 0x11. That this status means the end of a write is an assumption. id: no-length-check; basis: dispatch-table
The board clamps each commanded speed to between 1500 and 15000 (0x05DC and 0x3A98), which is above the 12200 the flight configuration allows. Verified Flat image 0x2043 to 0x2065; B2 etc/colibry/milosboard/drone.cfg:23-24. id: board-clamp; basis: none
Error 9 is a debounce, not a time bound: the battery routine counts up to 128 while the pack is in bounds (8.8 V to 13.4 V) and down while it is out of bounds, raises 9 when the count is 0, and 9 clears once the count exceeds 64. Verified Flat image 0x4991 to 0x49e7; counter at xdata 0x94:0x95, initial value 128 at 0x4070; clear at 0x4a64. id: error-9-debounce; basis: none
Errors 10 and 11 cannot be raised in closed loop: they come from a start-up routine called once before the main loop, and the fault interrupt raises 11 only in state 0 and error 5 in every other state. Verified Start-up routine 0x4be9 (single caller 0x56d8); raise sites 0x51fe (10) and 0x3361 (11); interrupt 0x3aa3 to 0x3aec. id: errors-start-up-only; basis: none
The flight program does not send the timeout command 0x8D to the motor board, so the board runs on its power-up watchdog default. Agent-cited Earlier analysis of dragon-prog; no immediate with the value 141 appears in its motor-board code range 0x440000 to 0x460000, where each command it does send is found. id: flight-no-8d; basis: none
The flight program’s stop at orderly shutdown is one 0x60 whose write result is checked; no status read follows it in the shutdown function or its cleanup call. When the write fails, the error branch only sets a link-failed flag and a stop-progress code and returns an error, which the shutdown function does not test. Verified dragon-prog 0x4f468 to 0x4f55c; branch at 0x45b8f4 to 0x45b91c. What the HAL call under 0x43d060 does internally was not traced. id: shutdown-stop; basis: none
The start command (0x40) is ignored unless the error number is 0 and the state is idle, so no start can happen while any error is latched; a start only clears the stop-request byte. Verified Flat image 0x219c to 0x21b4: non-zero error byte (xdata 2) leaves, state byte (xdata 0) not 1 leaves, otherwise xdata 3 is cleared. id: start-gate; basis: none
In flight, the stock flight program forwards board errors 8 and 9 as an alert to the app and a blackbox record, and no stop, landing or retry for them was found in the code that reads the motor result. Errors 2 to 5 clear the motors-spinning word and lead to the Emergency state. At takeoff, an error still present after the clear command refuses the start, and a refused start with idle motors leads to Emergency. Agent-cited B2 usr/bin/dragon-prog: the routines at 0x444800 and 0x45aab0 that separate errno 2 to 5 from 8 and 9 were re-read and match; the consumers, the Emergency wiring and the takeoff path are the agent’s reading. Only direct readers of the result were enumerated: whole-structure copies, other processes and what posts the later emergency events were not. basis: watchdog-suppression, flight-no-8d
While error 8 or 9 is latched the board stays in its closed-loop state until a stop request arrives, because that routine’s only way out is a stop request or error 2 to 5 and no other direct write to the state byte exists in the image (writes through a computed address were not searched). So, taking the status in the poll reply to be this same byte, the board status the flight program watches does not change on its own while one of them is latched. Verified Flat image: the routine at 0x4a47 to 0x4a9d has one ret and no branch out of its body; the state byte (xdata 0) has six writers in the image (state 2 at 0x23d4, 5 at 0x3889, 1 at 0x465e, 4 at 0x4a4b, 6 at 0x5778, and the start-up clear at 0x4bec), none of them in the loop or its callees. Not covered: state written through an address built from a computed register; that the status in the poll reply is this same byte is taken from earlier notes. id: closed-loop-state-stable; basis: none
Claim Status Evidence
/calib holds 13 files: an 8,192-byte camera EEPROM image, four F-theta lens tables, one distortion table, five ftd_*.txt and two userTuning*.txt ISP tables; /data and /factory are empty. Verified B2 calib/, data/, factory/.
The only separate imaging libraries are libarstream2.so, libh1enc.so, liboyster.so and libCondor.so; the first three are NEEDED by dragon-prog. Verified B2 usr/lib/; readelf -d usr/bin/dragon-prog; find for *dxo*, *alpaga*, *isp*, *vision* returns no imaging library.
dragon-prog contains 106 MT9F002 hits, 9 mt9v117 hits (lowercase driver function names; the upper-case form has 0), 50 dxo hits, 11 alpaga hits, and the sensor part names ICM20608, MS5607 and AK8963 (for example icm20608_init_commands, HAL_thread_ms5607_entry_point, HAL_ak8963_start). Verified B2 usr/bin/dragon-prog, strings and strings -ci.
Front sensor MT9F002 is addressed at I2C 0x10, downward sensor MT9V117 at 0x5D, on /dev/i2c-0. Verified B2 usr/bin/mk3_cam0_common.sh:5; usr/bin/mk3_cam2_common.sh:5; usr/bin/mk3_cam_common.sh:5.
The imaging code (ISP, auto-exposure, vision) is built into dragon-prog. Front chain: sensor, ISP, gyro-to-frame sync, crop prediction, fisheye correction, H.264 on the Hantro H1 encoder, RTP through libarstream2; a 160x120 luma image from a 320x240 capture feeds Condor. Agent-cited Project note camera-isp.md; consistent with the missing library files above.
Which lens table a given unit uses, the video-device mapping and the setCamera argument order. Open Listed as open in camera-isp.md.
The GPS firmware is a u-blox 8 image (526,548 bytes): starts UBX8, FWVER=SPG 3.01, PROTVER=18.00; the same bytes sit under @flasher/gps/ beside a 746,680-byte run binary containing PUBX and UBX protocol strings. Verified B2 etc/gps_firmware.bin; cmp against @flasher/gps/gps_firmware.bin is identical; strings on @flasher/gps/run.
/dev/ttyPA1 is the GPS port. Verified B2 etc/gps_config.txt:1, etc/init.d/rcS_gps:10, @flasher/gps/path.
The exact GPS module part number. Open The image identifies a generation, not a module.
Claim Status Evidence
The kernel exports 22 attributes under /sys/kernel/hsis, each a field of one integer structure mk3_hsis at 0xc06b3aa0, with one shared show function. Verified B2 partitions/section-00004-type-3.bin, inner section 1, gzip payload at 0x5ae8 decompressed: attribute table at file offset 0x6abe50 (20-byte entries, name pointers resolve at base 0xC0008000). basis: none
GPIO 199 is the motor-board reset line (reset_psoc). The factory scripts toggle gpio199, and the path user_gpio/RESET_PSOC appears as a string in the bootloader tool, the motor test tools and dragon-prog; in dragon-prog no code references it. Verified The structure cell for reset_psoc holds 199; B2 lib/firmware/BLDC_Factory_Tests.sh:67,75 toggles gpio199; the path is a string in dragon-prog, BLDC_Host_Bootloader, test-BLDC, check_motors, AR_Drone3_IMU_calibration; a scan for position-independent references finds none to the two strings in dragon-prog and two to the FAN path used as a control. A reference in the other tools’ code was not checked. id: reset-psoc-pin; basis: none
The Milos entry routine writes 83 to reset_gnss and 0 to super_led_pwm. Verified Kernel disassembly at 0xc06741ac: stores to offsets 0x54 and 0x24 of 0xc06b3aa0, then branches to 0xc0673ce4. basis: none
PWM channel n is a function option on kernel pad AAI_n (GPIO 119 plus n), for channels 6, 8, 9 and 11. Verified Pad strings AAI_06__GPIO_125__PWM_06, AAI_08__GPIO_127__PWM_08, AAI_09__GPIO_128__PWM_09, AAI_11__GPIO_130__PWM_11 in the kernel. basis: none
The camera clock scripts agree with the camera sensors’ master clocks: 23 ns (43 MHz) on the vertical camera and 77 ns (13 MHz) on the front camera. Verified B2 etc/init.d/rcS_vision:11-17, rcS_video:11-19; arithmetic. The sensor pairing is inferred from the clock figures. basis: none
On the Milos path hwrev equals pcbrev plus 11, and pcbrev comes from GPIOs 138 to 140. Agent-cited Read from the kernel code by an automated analysis; a unit’s reported pcbrev 3 and hwrev 14 (relayed by another session) fit the rule. basis: none
Which pad table applies to this chip revision; what servo1_pwm to servo4_pwm drive; what HSIS stands for. Open The kernel has two pad tables that disagree on GPIOs 199 to 204; no kernel attribute exists for the servo outputs; no file expands the acronym.
ArduPilot’s Bebop hardware definition independently uses PWM 6 for the IMU heater (125000 ns period) and PWM 9 for the vertical camera clock (43.33 MHz), the same channels the kernel’s pin table gives. Confirmed independently libraries/AP_HAL_Linux/hwdef/bebop/hwdef.dat in ArduPilot (commit 4c98c92) read against the kernel table decode and B2 etc/init.d/rcS_IO:8-14, rcS_vision:11-17. basis: none
Claim Status Evidence
mppd is the main daemon, sensorsd-mpp the sensor daemon, aoad the Android accessory daemon, and hudd the HUD renderer on -platform eglfs, disabled by default. Verified SC etc/boxinit.d/99-mppd.rc, 60-sensorsd-mpp.rc, 90-aoad.rc, 99-hudd.rc.
wifid runs in station mode on chip qca9375; wifid-ap runs in AP mode on chip wl18xx, is disabled by default and starts when gps.socket.ready=1; adbd starts only when ro.debuggable=1; the kernel modules live under /lib/modules/3.4.11+/. Verified SC etc/boxinit.d/50-wifi.rc, 50-adb.rc.
mppd carries ARSDK descriptors for the whole stack (Ardrone3_, Skyctrl_, Mapper_, Drone_manager) and both a mux and a net backend. Verified SC usr/bin/mppd strings, for example g_arsdk_cmd_desc_Ardrone3_Piloting_PCMD, arsdk_backend_mux_new, arsdk_backend_net_start_listen.
sensorsd-mpp links libiio, uses persist.gyro.bias and persist.magneto.bias, and shares the abstract socket name @sensorsd with mppd; the IMU is named mpu6050 and the magnetometer ak8963. Verified SC usr/bin/sensorsd-mpp strings and readelf -d; etc/boxinit.rc ro.sensors.imu.name, ro.sensors.magneto.name.
sensorsd-mpp reads only the controller’s own IMU and magnetometer, not the sticks; the sticks and buttons reach mppd from the USB bottom board. Agent-cited Project note skycontroller-stick-path.md; the binary links libiio and no libevdev or USB library; the link between the mpp2_gamepad input device and the board is a reading, not a traced path.
lib/firmware/xmega.hex (36,930 bytes) is flashed by bottom_board_update, which names the part atxmega32c4. Verified SC usr/bin/bottom_board_update:3-5.
udev matches an input device named mpp2_gamepad and the power key qpnp_pon, matches input nodes for vendor 03eb products 2406 and 2407, and matches hidraw nodes for the same ids. Verified SC lib/udev/rules.d/55-juba-gamepad.rules, 56-juba-bb.rules.
mppd references ro.factory.js.centers and ro.factory.js.min_max (stick centres and ranges); no file in the image sets them. Who sets them, and how the sticks are calibrated, is not known. Open /usr/bin/grep -rl over SC finds the strings only in usr/bin/mppd; no setter and no calibration file found.
ro.factory.type is set at boot by nap_factory_reader from /customer_prod/type.txt, and that directory is empty in the image. mppd selects /etc/mppd/{ro.factory.type}; two folders ship, bebop and milos. Verified SC etc/boxinit.rc:18; strings in usr/bin/nap_factory_reader and usr/bin/mppd; customer_prod/ and etc/mppd/ listings.
Default Bebop 2 axes: left stick yaw and throttle (gaz), right stick roll and pitch, right mini-stick camera pan and tilt; expo strength 1 on the four main axes and linear on the mini-sticks; identical in both folders. Verified SC etc/mppd/{bebop,milos}/mapping_bebop_2.cfg; action codes 16 roll, 17 pitch, 18 yaw, 19 gaz, 20 pan, 21 tilt.
Buttons, bebop folder: emergency Z, center_camera B, cycle_hud A, app_0 L1. Buttons, milos folder: center_camera Z, cycle_hud L1, app_0 B, and no emergency entry in the buttons array. In both: return_home Y, takeoff_land X, video_record C, take_picture R1. Verified SC etc/mppd/bebop/mapping_bebop_2.cfg:71-102; parsed buttons array of etc/mppd/milos/mapping_bebop_2.cfg.
Whether another path (a hard-coded action in mppd or a user override under /var/lib/mppd/) provides an emergency binding for milos, and which folder a given controller uses. Open Override path inferred from .cfg.tmp strings, not examined; the value of ro.factory.type lives on the unit, not in the image.
PCMD is encoded as flag, roll, pitch, yaw, gaz plus a u32 timestamp built from the low 24 bits of the millisecond clock and a sequence number in the top byte (FUN_000227d8); the flag clears after 5 consecutive zero-stick ticks; a 50 ms timer (FUN_00022a14) sends it at 20 Hz. Agent-cited Project note skycontroller-stick-path.md; decompile of mppd.
Piloting source 1 forwards the app’s stored PCMD re-stamped by mppd; source 0 is the default and a drone disconnect resets it; no authorization check on the handoff was seen; the controller has no stale-PCMD watchdog, RSSI trigger or link-lost handling. Agent-cited Same note, FUN_00022aac; absence of evidence only.
Loss-of-link behaviour on the drone side, whether the drone times out a repeated stale PCMD, the HID report layout, the xmega protocol, and the PCMD wire layout in libarsdk. Open Not examined.
The image announces model id 0x0913, which its own ARSDK library names SKYCTRL_NG, not SKYCTRL_2 (0x090f). The same library files nap_update.plf under 0x0903 (SKYCTRL), mpp_update.plf under 0x090f (SKYCTRL_2) and mpp2_update.tar under 0x0915. So this image is not the update file of the entry the SDK calls SKYCTRL_2. The image sets the model string Skycontroller and the hostname nap and never contains the string “Skycontroller 2”. One owner’s Skycontroller 2, read by another session, reports the model string Skycontroller 2 and the hostname mpp. What product the nap image belongs to is not established. Verified SC usr/lib/libarsdk.so: arsdk_device_type_str at 0x1fc08 (all 23 ids decoded from its jump table; 0x901 BEBOP and 0x90c BEBOP_2 agree with the Bebop images’ own model ids) and the update-file table (id paired with filename pointer); SC etc/boxinit.rc. The unit’s values are one unit’s, relayed, and are not firmware we hold.
What was said Status What the files show Evidence
The firmware contains libColibry, libNavdata and libHAL, and libCondor is the flight controller with setAttitude(roll, pitch, yaw, thrust). Corrected “Colibry” is a config namespace read by dragon-prog; none of the three libraries exists. libCondor exports image-registration code, setAttitude takes six floats, and the four labels had no source. B2 etc/colibry/; find over B2; usr/lib/libCondor.so, nm -D.
The estimator runs at about 90 Hz, and ACQCON_colibry_Thread_Handler is the control loop. Corrected adcFsSet is 198.3642578125 Hz and the 90 Hz figure had no source. The handler symbol exists once; the agent reading is that it copies frames and dispatches callbacks, with the filter elsewhere in dragon-prog. B2 etc/colibry/milosboard/root.cfg:8-13; usr/bin/dragon-prog, strings; project note colibry-fusion.md.
The ColibryState struct and the motor mixing table (“Front-Left +roll +pitch +yaw”) describe the firmware. Corrected Both were guesses; the config gives a rectangular layout with signs recomputed above. B2 etc/colibry/milosboard/drone.cfg.
Maximum motor speed is 16,000 RPM, and the motors are four speed controllers on a UART. Corrected 16000 is a command-line argument bound in test-BLDC; the flight config says 3000 to 12200. The motors are one board on I2C /dev/i2c-1; /dev/ttyPA1 is the GPS port. B2 usr/bin/test-BLDC; etc/colibry/milosboard/drone.cfg:23-24; lib/firmware/BLDC_Factory_Tests.sh:12; etc/gps_config.txt:1.
The motor-board link uses command numbers 0x00 to 0x09, a frame [cmd][params][CRC_LO][CRC_HI] and CRC-16-CCITT. Corrected Commands are single bytes (14 on the host tool, 10 handled by the 4.7.1 board image) with no length byte; the host tool has two eor instructions and no CRC routine; the image rows use a summation checksum. B2 usr/bin/BLDC_Test_Bench at 0x18424, 0x189f0; flat image table at 0x1fc4.
The motor board has a 120 C over-temperature cutoff, a 20 percent PWM soft-start over 500 ms and a 10 kHz PID loop. Corrected The 4.7.1 image raises an error at 95 C or above, or at or below -10 C, and that routine does not stop the motors; spin-up is alignment, ramp and lock counted in ticks; the loop tick is a timer reload 0xF9 whose clock is not derivable, so the period is unknown. Flat image 0x46ca to 0x4714; project note bldc-cyacd.md.
The motor-board silicon is PSoC 5LP. Corrected The 0x1E prefix points to the 8051-based PSoC 3 family; the exact part is not mapped. B2 lib/firmware/BLDC.cyacd:1; 8051 code in the flat image.
The bootloader requirement of version 3.30 or later “had no source”. Corrected The string (< v3.30) appears in the host bootloader’s rejection message. It constrains the board’s bootloader, which is not in the .cyacd file, and says nothing about the application. B2 usr/bin/BLDC_Host_Bootloader, strings.
Sensor I2C addresses and bus numbers in the first draft came from the firmware; the GPS is on /dev/i2c-0 and /dev/ttyPA1; the GPS model is “TBD”. Corrected Part names come from strings and the addresses were datasheet defaults (camera addresses were later read from scripts); only /dev/ttyPA1 is evidenced for GPS and /dev/i2c-0 is the camera bus; the bundled firmware is a u-blox 8 image. B2 usr/bin/mk3_cam0_common.sh:5, usr/bin/mk3_cam_common.sh:5, etc/gps_config.txt:1, etc/gps_firmware.bin.
The stock firmware serves MAVLink on UDP 14550 and 14551. Corrected The ports appear only in the disabled arducopter service. B2 etc/boxinit.d/50-arducopter.rc.
The tilt range [-70, 35] belongs to “the Bebop 2 board”. Corrected That range is in mykonos3board, the Bebop 1 platform set; the Bebop 2 tilt range is open. B2 etc/colibry/mykonos3board/video.cfg:5; B1 etc/build.prop.
0x2a63b0 is the mixer’s own config reader and the 16-bit truncation at 0x4177ec belongs to the mixer. Corrected 0x2a63b0 is a shared reader for the motorsParameters block used by ten classes; the truncation belongs to BLDCMotorController. Project note mixer-input-roles.md; addresses not re-derived.
A usage note for BLDC_Test_Bench said one option stops the motors. Corrected The tool’s help text lists a separate stop option and describes the clear-error option as clearing an error flag only. Help-text strings in B2 usr/bin/BLDC_Test_Bench.
section-00003-type-7.bin begins with eight ldr pc, [pc, #0x18] words. Corrected Seven such words; slot 5 is zero (the reserved vector slot). bebop-2-4.7.1/partitions/section-00003-type-7.bin, words 0 to 7.
The Skycontroller image never names its product, and the label “Skycontroller 2” was read from it. Corrected etc/boxinit.rc sets the model string Skycontroller; the string “Skycontroller 2” is absent. The label was inferred from the services (HUD, accessory daemon, mux backend). SC etc/boxinit.rc:21; etc/boxinit.d/.

Comments

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