IMU Jitter on OAK-D

ShivaRamoudith
Which sensors are you using and at which refresh rates?

Here are the frequencies possible.

Thanks,
Jaka

ShivaRamoudith
Possible frequencies:
Accelerometer: 15Hz, 31Hz, 62Hz, 125Hz, 250Hz 500Hz
Gyroscope: 25Hz, 33Hz, 50Hz, 100Hz, 200Hz, 400Hz

Setting to 200Hz will set accelerometer to 250Hz..

Thanks,
Jaka

ShivaRamoudith
Show a mre please.

The small jitter (₀.35 std) is expected because of the BNO interrupt inconsistencies.

Thanks,
Jaka

ShivaRamoudith
Remove

if cv2.waitKey(1) == ord('q'):
            break

it's too slow.

Thanks,
Jaka

Hi ShivaRamoudith

Try this:

import cv2
import depthai as dai
import matplotlib.pyplot as plt
import numpy as np

# Create pipeline
pipeline = dai.Pipeline()

# Define sources and outputs
imu = pipeline.create(dai.node.IMU)
xlinkOut = pipeline.create(dai.node.XLinkOut)
pipeline.setXLinkChunkSize(0)

xlinkOut.setStreamName("imu")

# Enable GYROSCOPE_RAW at 200 Hz rate
imu.enableIMUSensor(dai.IMUSensor.GYROSCOPE_RAW, 100)

# Set batch report threshold and max batch reports
imu.setBatchReportThreshold(1)
imu.setMaxBatchReports(10)

# Link plugins IMU -> XLINK
imu.out.link(xlinkOut.input)

# Initialize list to store time deltas
time_deltas = []

# Pipeline is defined, now we can connect to the device
with dai.Device(pipeline) as device:

    # Output queue for imu bulk packets
    imuQueue = device.getOutputQueue(name="imu", maxSize=50, blocking=False)  # Reduce maxSize for more frequent updates
    lastTimestamp = None

    while len(time_deltas) < 1000:
        imuData: dai.IMUData = imuQueue.get()  # blocking call, will wait until new data has arrived
        currentTimestamp = imuData.getTimestamp().total_seconds()
    
        if lastTimestamp is not None:
            delta = (currentTimestamp - lastTimestamp)  # Calculate time delta
            time_deltas.append(delta)

        lastTimestamp = currentTimestamp

        

# Calculate mean and standard deviation
time_deltas_ms = [x * 1000 for x in time_deltas]
mean_delta = np.mean(time_deltas_ms)
std_delta = np.std(time_deltas_ms)

# Create scatter plot
plt.figure(figsize=(10, 6))
plt.scatter(range(len(time_deltas_ms)), time_deltas_ms, c='blue', label='Time Deltas', s=10)
plt.axhline(mean_delta, color='red', linestyle='--', label=f'Mean = {mean_delta:.2f} ms')
plt.text(0, 0, f"Mean: {mean_delta:.2f} ms\nStd: {std_delta:.2f} ms", fontsize=12, bbox=dict(facecolor='white', alpha=0.5))
plt.xlabel('Measurement Index')
plt.ylabel('Time Delta (ms)')
plt.title('Scatter Plot of Time Deltas Between Consecutive IMU Packets')
plt.legend()
plt.show()

Thanks,
Jaka

ShivaRamoudith
Pretty much meaning what? The variance is expected due to how the IMU works. If you get timestamps that are multples of the base time difference, it means the host side loop is unable to keep up with the IMU frequency.

Thanks,
Jaka

ShivaRamoudith
That would be correct, but the IMU packets are discarded if they are not read in time. You can see this by increasing the queue size to eg. 100. It will take more time for measurements to become unstable.

Thanks,
Jaka

ShivaRamoudith
Even with 10000 datapoints I am not getting any issues. It's likely that some link in the pipeline is incapable of keeping up with the speed of data, potentially dropping frames. You can try setting all queues to blocking and setMaxBatchReports to a higher number so IMU itself doesn't drop frames if they are not read fast enough.

Thanks,
Jaka