tobidelbruck
commented
6 years ago I see line 628 of https://github.com/uzh-rpg/rpg_dvs_ros/blob/master/davis_ros_driver/src/driver.cpp should convert from caer g units to m/s^2 unit. (I don't know what the switch is there for, since caer should deliver the IMU data in same form for each type of DAVIS).
if (davis_info_.chipID == DAVIS_CHIP_DAVIS346B)
{
// convert from g's to m/s^2 and align axes with camera frame
msg.linear_acceleration.x = -caerIMU6EventGetAccelY(event) * STANDARD_GRAVITY;
msg.linear_acceleration.y = -caerIMU6EventGetAccelX(event) * STANDARD_GRAVITY;
msg.linear_acceleration.z = -caerIMU6EventGetAccelZ(event) * STANDARD_GRAVITY;
// convert from deg/s to rad/s and align axes with camera frame
msg.angular_velocity.x = -caerIMU6EventGetGyroY(event) / 180.0 * M_PI;
msg.angular_velocity.y = -caerIMU6EventGetGyroX(event) / 180.0 * M_PI;
msg.angular_velocity.z = -caerIMU6EventGetGyroZ(event) / 180.0 * M_PI;
}
else
{
// convert from g's to m/s^2 and align axes with camera frame
msg.linear_acceleration.x = -caerIMU6EventGetAccelX(event) * STANDARD_GRAVITY;
msg.linear_acceleration.y = caerIMU6EventGetAccelY(event) * STANDARD_GRAVITY;
msg.linear_acceleration.z = -caerIMU6EventGetAccelZ(event) * STANDARD_GRAVITY;
// convert from deg/s to rad/s and align axes with camera frame
msg.angular_velocity.x = -caerIMU6EventGetGyroX(event) / 180.0 * M_PI;
msg.angular_velocity.y = caerIMU6EventGetGyroY(event) / 180.0 * M_PI;
msg.angular_velocity.z = -caerIMU6EventGetGyroZ(event) / 180.0 * M_PI;
}
Line 112 of https://github.com/uzh-rpg/rpg_dvs_ros/blob/master/davis_ros_driver/include/davis_ros_driver/driver.h defines ` static constexpr double STANDARD_GRAVITY = 9.81;
`
https://github.com/inilabs/libcaer/blob/master/include/events/imu6.h defines cAER units for acceleration as g's. Therefore I think that the g units of cAER are being multiplied by another factor of g in this code, rather than being divided by it. I must be missing something.
guillermogb
I wonder if it has been checked that the units for the DAVIS IMU acceleration are really correct? I am seeing in one recording about -0.9 m/s^2 for y component, which should be -9.8m/s^2 or about -1g. This comes up in developing a native-java ROS bag reader for jaer, where I need to convert back from ros IMU units to internal jAER units, which are in g for acceleration and deg/s for gyro rotation (and are the native units for the invensense IMU, which is why these units were unfortunately used to begin with).
Does the ROS IMU acceleration have gravity component removed from y component? This wouldn't make too much sense but could it be why y is too small? It doesn't really make sense, since y component is always the one with significant negative component, and corresponds to how the camera was mounted in the recording. I haven't looked at the source code yet, sorry for this laziness.