Lab 2: IMU Setup + Motion Data

Set up the IMU

IMU Connection Wiring

Blinking LED Test

IMU Working Output

Output at {-90, 0, 90} degrees (Pitch + Roll)

I recorded accelerometer data at three known orientations at -90, 0, and 90 degrees for both pitch and roll. Then I compared the results to the expected pitch/roll values by finding the conversion factor. I took that value and multiplied it to the recorded accelerometer data to obtain proper values for pitch and roll.

Accelerometer Code

case GET_ACC: {
    unsigned long starttime = millis();
    
    for (int i = 0; i < arraylength;) {
        if (myICM.dataReady()) {
            myICM.getAGMT();
        }
        
        // Conversion factors
        const float pitch_full_range = 180.0 / (86.0 + 88.4);
        const float roll_full_range  = 180.0 / (89.0 + 87.4);
        
        float accelerometer_pitch = atan2(myICM.accY(), myICM.accZ()) * 180.0 / M_PI * pitch_full_range;
        float accelerometer_roll  = atan2(myICM.accX(), myICM.accZ()) * 180.0 / M_PI * roll_full_range;

        time_list[i] = (float)(millis() - starttime);
        Acc_pitch_list[i] = accelerometer_pitch;
        Acc_roll_list[i] = accelerometer_roll;

        i++;
    }
    
    // Send data over Bluetooth
    for (int i = 0; i < arraylength;) {
        tx_estring_value.clear();
        tx_estring_value.append(time_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Acc_pitch_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Acc_roll_list[i]);
        tx_characteristic_string.writeValue(tx_estring_value.c_str());
        i++;
    }
    break;
}

Pitch vs Time

Roll vs Time

Accelerometer Accuracy Discussion

My accelerometer readings were initially imprecise. The main problem was that the total angle range from -90 to 90 degrees was actually less than 180 degrees. To combat this, I took the desired range of 180 and divided it by the accelerometer readings at the real -90 and +90 degrees. This gave me a scaling factor where I could obtain the proper range by multiplying this by the raw accelerometer angle.

Fourier Transform Analysis

Raw Pitch and Roll: Induced Noise

Fourier Transform Analysis

The raw pitch and roll plots show noisy behavior with induced noise. The Fourier transform plots of the sampled data from 0 to 0.214 seconds reveal the frequency components of this noise.

Low Pass Filter Analysis

Filtered Data

Sampling Frequency Calculation

RC = 1/(2π × fc) = 1/(2π × 5) = 0.0318
α = ts/(ts + RC) = 0.214/(0.214 + 0.0318) = 0.871

To implement the Low Pass Filter for the accelerometer, I first calculated the sampling rate by finding the mean difference between time points, resulting in 1163 Hz. The LPF does an excellent job at smoothing the data and eliminating random spikes.

Pitch, Roll, and Yaw Documentation

I tested multiple IMU positions and documented how pitch, roll, and yaw changed based on the board orientation.

Gyroscope Code

case GET_GYRO: {
    const unsigned long starttime = millis();
    
    for (int i = 0; i < arraylength;) {
        if (myICM.dataReady()) {
            myICM.getAGMT();
        }
        
        // Accelerometer calculations
        const float pitch_full_range = 180.0 / (86.0 + 88.4);
        const float roll_full_range  = 180.0 / (89.0 + 87.4);
        
        float accelerometer_pitch = atan2(myICM.accY(), myICM.accZ()) * 180.0 / M_PI * pitch_full_range;
        accelerometer_roll = atan2(myICM.accX(), myICM.accZ()) * 180.0 / M_PI * roll_full_range;
        
        // Gyroscope calculations
        float dt = (float(millis()) - starttime) / 1000.0;
        float gx = myICM.gyrX();
        float gy = myICM.gyrY();
        float gz = myICM.gyrZ();
        
        // Store data
        time_list[i] = (float)(millis() - starttime);
        Acc_pitch_list[i] = accelerometer_pitch;
        Acc_roll_list[i] = accelerometer_roll;
        
        if (i == 0) {
            Gyro_pitch_list[i] = 0;
            Gyro_roll_list[i] = 0;
            Gyro_yaw_list[i] = 0;
        } else {
            Gyro_pitch_list[i] += gx * dt;
            Gyro_roll_list[i] += gy * dt;
            Gyro_yaw_list[i] += gz * dt;
        }
        
        // Complementary filter
        comp_gyro_pitch[i] = (comp_gyro_pitch[i] * (1 - alpha)) + (accelerometer_pitch * alpha);
        comp_gyro_roll[i] = (comp_gyro_roll[i] * (1 - alpha)) + (accelerometer_roll * alpha);
        
        i++;
    }
    
    // Send data via Bluetooth
    for (int i = 0; i < arraylength; i++) {
        tx_estring_value.clear();
        tx_estring_value.append(time_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Acc_pitch_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Acc_roll_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Gyro_pitch_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Gyro_roll_list[i]);
        tx_estring_value.append("|");
        tx_estring_value.append(Gyro_yaw_list[i]);
        tx_characteristic_string.writeValue(tx_estring_value.c_str());
    }
    break;
}

Position 1 Output

Position 2 Output

Without Complementary Filter

Complementary Filter Accuracy & Range

I used a complementary filter to combine gyroscope data (fast response) with accelerometer data (stable long-term reference). I tuned the blending factor to reduce drift while keeping motion response quick and accurate.

Sampling Speed Analysis

I measured how fast I could sample and transmit IMU data, and noted where bottlenecks occurred (printing, BLE packet size, processing time, etc.).

Time-stamped IMU Data Storage

I collected time-stamped readings (t, ax, ay, az, gx, gy, gz) and stored them in arrays for later processing and transmission.

Array Storage Implementation

5 Seconds of Bluetooth Data

Video Evidence & Observations

I recorded a short stunt run to demonstrate the system working and documented observations from the IMU data during the maneuver.