About sMartins

Hello Bosch-Community, We use Adafruit_BNO055.h (and Adafruit_Sensor.h) to gather IMU-Data. We set no registers and just use: bno = Adafruit_BNO055(55, 0x28, imuConfig.theWire); // and after successful initialisation //get all data from BNO 055 sensors_event_t angVelocityData, linearAccelData; bno.getEvent(&angVelocityData, Adafruit_BNO055::VECTOR_GYROSCOPE); bno.getEvent(&linearAccelData, Adafruit_BNO055::VECTOR_LINEARACCEL); imu::Quaternion quat = bno.getQuat(); to gather data as json. It works! Now I want to generate a rotation matrix from your quaternion data to transform your linear_x, _y, _z acceleration from sensor coordinate system into ENU coordinate system. I'm not really interested into the orientation of the sensor. I just want the linear accelerations in ENU (or NED) reference frame. For this I analysed the attached data, where I wiggle the IMU across a table in sensor X (almost North), wait, turn 90° cw, wiggle again in X, wait, 90° cw, wiggle in X.   from scipy.spatial.transform import Rotation as R import numpy as np for measurement in data: raw_quat = np.array([measurement['quat_w'], measurement['quat_x'], measurement['quat_y'], measurement['quat_z']]) correct_quat = np.array([measurement['quat_w'], measurement['quat_z'], measurement['quat_y'], measurement['quat_x']]) raw_rotation = R.from_quat(raw_quat) corrected_rotation = R.from_quat(correct_quat) # Quaternion to rotation matrix corrected_rotation_matrix = corrected_rotation.as_matrix() # IMU measurements accel = np.array([measurement['linear_x'], measurement['linear_y'], measurement['linear_z']]) # Transform IMU measurements to NED (ENU??) frame accel_ned = np.dot(corrected_rotation_matrix, accel)   This only works as expected in the "correct_quat" case, where I swapped z and x quaternion (PS: not true if I use the quaternion_rotation_matrix instead of R.from_quat). Please answer me, if this is correct or if I am doing something horribly wrong. If I use raw_quat (PS: with R.from_quat), it does not calculate correctly and linear_x stays untouched. Kind regards, Martin PS/Edit: I think the scipy Rotation implementation is a bit opaque. Therefore let's use this implementation of quaternion to rotation matrix implementation as basis for our discussion (from https://automaticaddison.com/how-to-convert-a-quaternion-to-a-rotation-matrix/). My main question is, if this is the correct approach for linear acceleration transformation to ned/enu, or if I am missing something besides calibration, filtering and smoothing? import numpy as np def quaternion_rotation_matrix(Q): """ Covert a quaternion into a full three-dimensional rotation matrix. Input :param Q: A 4 element array representing the quaternion (q0,q1,q2,q3) Output :return: A 3x3 element matrix representing the full 3D rotation matrix. This rotation matrix converts a point in the local reference frame to a point in the global reference frame. """ # Extract the values from Q q0 = Q[0] q1 = Q[1] q2 = Q[2] q3 = Q[3] # First row of the rotation matrix r00 = 2 * (q0 * q0 + q1 * q1) - 1 r01 = 2 * (q1 * q2 - q0 * q3) r02 = 2 * (q1 * q3 + q0 * q2) # Second row of the rotation matrix r10 = 2 * (q1 * q2 + q0 * q3) r11 = 2 * (q0 * q0 + q2 * q2) - 1 r12 = 2 * (q2 * q3 - q0 * q1) # Third row of the rotation matrix r20 = 2 * (q1 * q3 - q0 * q2) r21 = 2 * (q2 * q3 + q0 * q1) r22 = 2 * (q0 * q0 + q3 * q3) - 1 # 3x3 rotation matrix rot_matrix = np.array([[r00, r01, r02], [r10, r11, r12], [r20, r21, r22]]) return rot_matrix For your convenience, graphs of the raw and processed data. Euler(Euler) and Euler(Quats) was generated by R.from_euler/from_quat and R.as_euler('xyz', degrees=True). Attached you also find an Animation of the Rotation and "correct" Rotation. As of PS I added the a plot of the original quat(w, x, y, z) using the quaternion_rotation_matrix function above the old image with quat(w, z, y, x). Old plot x and z axis exchanged   ... View more