ArduPilot against PX4¶

My drone runs ArduPilot, not PX4: what changes?

The drone pipelines read a MAVLink log with read_mavlink_log, and the seven fields it yields are the same on both autopilots. What changes is the message the autopilot emits for each field, and sometimes whether it emits one at all. This page lists, for each field, the MAVLink message and the field it comes from on ArduPilot and on PX4, and shows one runnable example per step.

The names below were checked in the two dialects of pymavlink, pymavlink.dialects.v20.ardupilotmega and pymavlink.dialects.v20.common, by importing them and printing the field names. Koubaa et al. 2019, the MAVLink survey, names only HEARTBEAT explicitly (page 3); the other names come from the dialect files, which are the authoritative source.

ALFA is ArduPilot (Keipour, Mousaei and Scherer 2020, section 2, page 2), so the ALFA reader of pipelines/drones/alfa.py is the one to reuse, not the ULog one.

1. The seven fields¶

Field	ArduPilot message and field	PX4 message and field
`t`	`time_usec` or `time_boot_ms` of any message	`time_usec` or `time_boot_ms` of any message
`gyro`	`HIGHRES_IMU.xgyro, ygyro, zgyro`; `ATTITUDE.rollspeed, pitchspeed, yawspeed` when `HIGHRES_IMU` is absent	`HIGHRES_IMU.xgyro, ygyro, zgyro`
`acc`	`HIGHRES_IMU.xacc, yacc, zacc`	`HIGHRES_IMU.xacc, yacc, zacc`
`gps_v`	`GPS_RAW_INT.vel, cog`	`GPS_RAW_INT.vel, cog`
`motor`	`SERVO_OUTPUT_RAW.servo1_raw ... servo4_raw`	`SERVO_OUTPUT_RAW.servo1_raw ... servo4_raw`
`battery_v`	`BATTERY_STATUS.voltages[0]`	`BATTERY_STATUS.voltages[0]`
`commanded`	`ATTITUDE_TARGET.body_roll_rate, body_pitch_rate, body_yaw_rate`	`ATTITUDE_TARGET.body_roll_rate, body_pitch_rate, body_yaw_rate`

The two messages that differ are HIGHRES_IMU and ATTITUDE: ArduPilot logs both (the HIGHRES_IMU only when the high-resolution stream is enabled), PX4 logs HIGHRES_IMU. When only ATTITUDE is present, the gyro comes from it and acc stays None.

2. The gyro and the acceleration¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/imu.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_highres_imu_message(
        1000000, 0.0, 0.0, 9.81, 0.1, 0.2, 0.3,
        0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0, 0))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/imu.tlog"))
print([round(float(v), 3) for v in s["gyro"]],
      [round(float(v), 3) for v in s["acc"]])

[0.1, 0.2, 0.3] [0.0, 0.0, 9.81]

The message writes the gyro in rad/s and the acceleration in m/s^2, and read_mavlink_log copies them unchanged. This is the PX4 case: HIGHRES_IMU is the message both autopilots can log.

3. The gyro from ATTITUDE when HIGHRES_IMU is absent¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/att.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_attitude_message(
        1000, 0.0, 0.0, 0.0, 0.7, 0.8, 0.9))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/att.tlog"))
print([round(float(v), 3) for v in s["gyro"]], s["acc"])

[0.7, 0.8, 0.9] None

ArduPilot logs ATTITUDE even when the high-resolution stream is off. read_mavlink_log uses it only when HIGHRES_IMU is absent, and the acceleration stays None: on that log the field cannot be read with the current reader. That is a proposal for io.py, not a change: SCALED_IMU2.xacc, yacc, zacc is the message ArduPilot logs, and adding it to the reader would fill acc without touching the other fields.

4. The GPS velocity¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/gps.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_gps_raw_int_message(
        1000000, 3, 0, 0, 0, 0, 0, 300, 9000, 10,
        0, 0, 0, 0, 0, 0))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/gps.tlog"))
print([round(float(v), 3) for v in s["gps_v"]])

[0.0, 3.0, 0.0]

vel is in cm/s and cog in centidegrees. The reader converts both and rotates the velocity to NED: 300 cm/s with a course of 90 degrees is 3 m/s east.

5. The motors¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/servo.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_servo_output_raw_message(
        1000000, 0, 1500, 1600, 1260, 1520, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/servo.tlog"))
print([round(float(v), 3) for v in s["motor"]])

[1.5, 1.6, 1.26, 1.52]

The four servo*_raw channels are divided by 1000, so 1500 is 1.5 and 1260 is 1.26. The scale is the one of the synthetic stream, not a normalisation to [0, 1]: a servo output in [1000, 2000] maps to [1.0, 2.0].

6. The battery¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/bat.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_attitude_message(
        1000, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0))
    mav.send(mavlink.MAVLink_battery_status_message(
        0, 0, 0, 0, [12600, 0, 0, 0, 0, 0, 0, 0, 0, 0],
        0, 0, 0, 0, 0, 0, [0, 0, 0, 0], 0, 0))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/bat.tlog"))
print(round(float(s["battery_v"]), 3))

12.6

BATTERY_STATUS.voltages[0] is in mV, divided by 1000. The ATTITUDE message comes first because BATTERY_STATUS has no timestamp of its own: the reader gives it the last timestamp seen.

7. The commanded rate¶

from pymavlink.dialects.v20 import ardupilotmega as mavlink

with open("/tmp/cmd.tlog", "wb") as f:
    mav = mavlink.MAVLink(f)
    mav.srcSystem = 1
    mav.srcComponent = 1
    mav.send(mavlink.MAVLink_attitude_target_message(
        1000, 0, [1.0, 0.0, 0.0, 0.0], 0.1, 0.2, 0.3, 0.0))

from pipelines.drones.io import read_mavlink_log

s = next(read_mavlink_log("/tmp/cmd.tlog"))
print([round(float(v), 3) for v in s["commanded"]])

[0.1, 0.2, 0.3]

ATTITUDE_TARGET carries the reference rate that the autopilot is tracking. On a log without it, commanded is None and the actuator-fault pipeline cannot compute the residual.

Details¶

References¶

Keipour, A., Mousaei, M., Scherer, S. (2020). ALFA: A Dataset for UAV Fault and Anomaly Detection. The International Journal of Robotics Research. arXiv:1907.06268v2. Section 2, page 2: ALFA is ArduPilot. Section 3.1, page 3: the processed data are one CSV per topic.
Koubaa, A. et al. (2019). Micro Air Vehicle Link (MAVLink) in a Nutshell: A Survey. IEEE Access. arXiv:1906.10641. Page 3: HEARTBEAT is the only message the survey names explicitly.
The dialect files pymavlink.dialects.v20.ardupilotmega and pymavlink.dialects.v20.common are the source of every message and field name in the table.

Choices stated explicitly¶

The mapping from field to message is the one of pipelines/drones/io.py, not a new one. The table states it, it does not propose a different reader.
The eight code blocks write their own tlog with pymavlink and read it back with read_mavlink_log: no file on disk is needed, and each block is self-contained.
motor is divided by 1000, as in io.py, so the values are in [1.0, 2.0] on a board log and not in [0, 1].
SCALED_IMU2 is named as the proposal for acc when HIGHRES_IMU is absent, because ArduPilot logs it. It is not added to io.py here: B says no change to io.py.

Development diary¶

The first version of the page read an ALFA tlog and showed the first samples. The tlog contains RAW_IMU and SCALED_IMU2, not HIGHRES_IMU, and read_mavlink_log reads only the latter: the gyro comes from ATTITUDE, the acceleration is None. The page now says that, and the examples write a tlog with the message each field needs, so the reader is shown on the case it handles.