ros-drivers
Mounting Advice for ROS
Hey, I have a question regarding the mounting orientation of a PhdigetSpatial Precision 3/3/3 on a mobile robot. We have a four-wheeled rectangularly shaped robot platform and want to mount the IMU in the center.
In which orientation should the IMU be mounted so that it complies with the ROS convention of X-forward, Y-left, Z-up? As I understand from the documentation, the X axis goes from left to right, Y from front to back, w.r.t. the USB Port.
Can we mount it in any direction and apply a matching static transform (with static_transform_publisher or imu_pipeline/imu_transformer) or should we place it in the correct orientation from the beginning?
Thanks for your advice!
On my old PhidgetSpatial 1044 3/3/3, the axes are printed onto the board; IIRC, axis 0=x, 1=y, 2=z. You can easily figure this out (or verify it) yourself for your IMU. Just install this driver, turn the IMU onto each of the 6 sides and echo the raw IMU values published by the driver. If the linear acceleration for axis x is positive and roughly +9.81 N, and the linear acceleration for y and z are roughly 0, then this is the direction of the positive x axis. Same for the y and z axes.
The only thing that really matters is the direction of the z axis. The ROS standard is to use the ENU convention for IMUs (x/y/z = East/North/Up), so the z axis should point up. For my IMU, this means mounting it "upside down" (i.e., with the underside of the PCB pointing up). Some packages, like robot_localization, assume the IMU data to follow the ENU convention. However, if you mount your IMU in a different orientation, you should be able to transform it using imu_transformer.
Thank you for this detailed answer! Could you please tell me, why only the Z axis matter? The Axis are printed on the devices exactly as you say, which means that with the USB Port of the device pointing to the rear of the robot, Y points forward, X points left and Z points down. Rolled upside down, X points right. Don't I need to rotate again around Z to make X point forward?
Thank you for your time!
If you want, you can make X point forward. This has the advantage that if you look at the linear_acceleration field and angular_velocity of the Imu message, they follow ROS standard (X forward, Y left, Z up).
What I meant by "only Z matters" is that in my use case, I almost never look at the Imu message directly. I feed it to a filter (in my case imu_filter_madgwick) that fills the orientation field. (NB: If you have a newer Phidgets, you can set the use_orientation parameter to true so you don't need an external filter, see README).
The orientation field reports the orientation of the IMU with respect to a reference frame. That reference frame is ENU if you enable the magnetometer. However, I never enable the magnetometer because on small robots, the disturbance from the electric motors is too large. Without a magnetometer, the orientation of the x and y axes of the reference frame is arbitrary (it's just the initial orientation of the axes when you started the filter) because there is no reference measurement for these axes; the Z axis is special because the gravity vector gives us a reference measurement that tells us which way the Z axis should point. This is what I meant by "only Z matters", which was not 100% correct.
BTW, later stages in the pipeline like robot_localization and amcl don't care about the relative orientation of the x and y axes anyway, they only look at the relative change in angles.