Paper Link: https://www.science.org/doi/epdf/10.1126/scirobotics.ade4538
Video Link: https://www.youtube.com/watch?v=lrEJnJrRJsQ&t=218s
Including the CAD files of PULSAR and the swashplateless mechanism.
Including the hardware design files of the magnetic encoder (AS5600).
- Power the magnetic encoder and config it using a I2C bus (please follow the AS5600 datasheet, you can use Arduino or Raspberry Pi or other devices you want, the encoder configuration is not involved in this PX4 firmware), and make sure the 910-Hz PWM signal is outputed (e.g., verified by a oscilloscope).
- Connect the power port of magnetic encoder to a 5V port of the Pixhawk 4 Mini.
- Connect the PWM signal of magnetic encoder to the CAP1 port of the Pixhawk 4 Mini.
- Connect the ESC (off-the-shelf ESC is ok as long as it supports dshot protocol) to the MAIN OUT 1 port of the Pixhawk 4 Mini.
- Clone the repository
- Config your compile environment https://docs.px4.io/main/en/dev_setup/dev_env.html
- Build the code and upload to Pixhawk 4 mini through USB connection
cd PULSAR
git submodule update --init
cd 3-PX4_Firmware/PULSAR_PX4
make px4_fmu-v5_default upload
- Load the parameter file "PULSAR_PX4_parameters.params" to the Pixhawk 4 mini by QGroundControl or other tools (very important)
- Keeping the USB connection, in the MAVLINK Console of QGroundControl, type
px4_cn start
- If the encoder works well, two rotor angles (angle_compensated and angle_raw) will be printed in the MAVLINK terminal. Rotate the positive blade of the swashplateless mechanism to align with the negative direction of the body y axis of PULSAR (facing to LiDAR's front side, right hand side is the nevative y axis), and record the value of the printed "angle_raw".
- Modified the MACRO (SENSOR_ROTOR_ANGLE_BIAS) with the recorded value (see 3.2.2).
- Compile the code and upload again.
- Keeping the same blade's position and starting the angle print in MAVLINK again, make sure the "angle_compensated" is close to 0 deg or 360 deg.
- To stop the print, type
px4_cn stop
The files are located in "src/drivers/pwm_input". Please make sure that the encoder 910-Hz PWM signal is connected to the CAP1 of the Pixhawk 4 Mini. Changing the frequency should also modify the timer frequency in "PULSAR_PX4/src/drivers/pwm_input/pwm_input.cpp".
The mixer files are located in the "PULSAR_PX4/src/drivers/dshot/swashplateless".
SENSOR_ROTOR_ANGLE_BIAS (unit is degree) is used for compensating the encoder installation offset angle and it must be set properly.
MOTOR_DELAY_ANLGE_BIAS (unit is radian) is used for compensating the lag angle of the swashplateless mechanism. This value is different when using different motor or propeller and can be determined on a test stand with 6-axis force sensor by analyzing the moment data. If you use the propulsion system same as PULSAR (i.e., T-MOTOR MN5006 KV450 and MF1302 propeller), you don't need to change this value.
The data of swashplateless mechanism are published for logging.
The repalcement is conducted in "PULSAR_PX4/src/drivers/dshot/dshot.cpp". Please make sure that the ESC signal wire is connected to the MAIN OUT1 of the Pixhawk 4 Mini. The dshot has been enabled if you had loaded the parameter file into the Pixhawk 4 mini.
The attitude controller is implemented in "PULSAR_PX4/src/modules/mc_att_control/AttitudeControl/AttitudeControl.cpp". The attitude setpoint and feedback are published for data logging.
The angular velocity controller is implemented in "PULSAR_PX4/src/modules/sa_rate_control_me/sa_rate_control_me_main.cpp". The angular velocity setpoint and feedback both in world frame are published for data logging.
There are some modifications in "PULSAR_PX4/src/modules/mc_pos_control" and "PULSAR_PX4/src/lib/flight_tasks". This ensure the position command from remote controller (RC) is available in world frame rather than in body frame.
Some ulog topics are added in "PULSAR_PX4/src/modules/logger/logged_topics.cpp" and "PULSAR_PX4/msg".
Different from the standard version of FAST-LIO2, the FAST-LIO2 in this repository contains coordinate transformations of both odometry and point clouds because the LiDAR (Livox AVIA) is installed on PULSAR with a 90-degree rotated orientation. Refer to https://github.com/hku-mars/FAST_LIO for more configuration details.
To transmit the odometry from FAST-LIO2 to PX4, MAVROS should be run with FAST-LIO2 at the same time
roslaunch mavros px4.launch
roslaunch fastlio mapping_avia_indoor.launch # indoor environment
roslaunch fastlio mapping_avia_outdoor.launch # outdoor environment
In addition, Point-LIO is more suitable for PULSAR since it can handle higher self-rotation rate.
The new version of the planner used in PULSAR is in https://github.com/hku-mars/dyn_small_obs_avoidance. Other planners can also be used for PULSAR if they can receive the odometry and point clouds provided by FAST-LIO2, Point-LIO, or other LIOs.
The incremental kd-forest is a data strcuture which contains many ikd-trees. You may contact Yixi Cai if there is any issue.