Skip to content

ImuMonitor#

Description#

ImuMonitor provides a functionality to detect anomalous behaviours of an IMU by comparing with rigidly fixed redundant IMUs. It also contains tools for setting valid parameters of anomaly detection. It is composed of the following packages: - imu_monitor - imu_anomaly_simulator - imu_anomaly_analyzer

The imu_monitor detects anomalous behaviours of an IMU and publishes its state to /diagnostics. The imu_anomaly_simulator can generate anomalous data from normal IMU data. The imu_anomaly_analyzer helps you to analyze anomaly of the IMU. You can set and validate parameters of imu_monitor using imu_anomaly_simulator and imu_anomaly_analyzer.

imu_monitor#

Algorithms#

  1. Transform IMU data from sensor frame to base_link_frame.
  2. Save recent sample_buffer_size yaw rotation speeds(angular_velocity.z) of each IMU into buffer.
  3. Estimate rotation speed of each IMU by calculating the average of the buffer elements.
  4. Find outlier candidate with the largest difference to the average among the rotation speeds.
  5. Calculate difference between the outlier candidate and the average of all speeds besides the outlier candidate.(= imu_outlier_deviation)
  6. imu_outlier_deviation with a larger value than deviation_threshold is regarded as anomaly.

Parameters#

Edit config/imu_anomaly_detector.param.launch to set the following parameters.

Name Type Description
deviation_threshold float deviation_threshold in Algorithms
sample_buffer_size int sample_buffer_size in Algorithms
imu_count int The number of input IMUs
base_link_frame string base_link_frame in Algorithms

Input topics#

Name Type Description
~/input/imu0, ~/input/imu1, ... sensor_msgs::Imu Input IMUs. Use imu_count parameter to set the number of IMUs

Output topics#

Name Type Description
/diagnostics diagnostic_msgs::DiagnosticArray Diagnostics topic

Diagnostics topic#

[summary]

level message
OK OK
WARN Warning

[values]

key value
anomalous_imu_id 0 ~ (imu_count - 1)
imu_outlier_deviation imu_outlier_deviation in Algorithms

imu_anomaly_simulator#

The imu_anomaly_simulator node provides a functionality to simulate anomalous IMU messages from normal IMU messages.

Input topics#

Name Type Description
~/input/imu sensor_msgs::Imu Input normal IMU messages which are converted into anomalous messages.

Output topics#

Name Type Description
~/output/imu sensor_msgs::Imu Output IMU.
~/output/is_anomaly std_msgs::Bool Indicates that the simulator is publishing anomalous IMU messages.

Control of simulator#

imu_anomaly_simulator is controlled by the controller node. imu_anomaly_simulator repeatedly takes the following steps by controller: 1. Output normal data for a random duration. 2. Output anomalous data according to first element of error_patterns defined in config file. 3. In the next step, next element of error_patterns is used instead of first element.

You can change behaviour of controller by editing configuration file. Example configuration file is in imu_anomaly_simulator/config/. Configurable variables are the following: - normal_phase_duration_min: Minimum duration in seconds of step1 - normal_phase_duration_max: Maximum duration in seconds of step1 - error_patterns: Array of error patterns. Each element in the array should has the following hashes. - duration: Duration to keep outputting - bias, coeff, noise_coeff: Parameters of anomalous data generation which are composed of x, y, z.

imu_anomaly_analyzer#

The imu_anomaly_analyzer node provides topics which are useful to analyze imu_monitor.

Input topics#

Name Type Description
/diagnostics diagnostic_msgs::DiagnosticArray Diagnostics topic

Output topics#

Name Type Description
~/output/is_anomaly_detected std_msgs::Bool Indicates that IMU Monitor detects anomaly
~/output/imu_outlier_deviation std_msgs::Float32 imu_outlier_deviation in Algorithms
~/output/max_imu_outlier_deviation std_msgs::Float32 Max value of imu_outlier_deviation

How to Use#

Setup config file and launch file#

  1. Set imu_count in imu_anomaly_detector.param.yaml to the number of IMUs in your system.
  2. Create launch file by modifying imu_monitor.launch.xml, which is for a system with three IMUs(one IMU and two redundant IMUs).

Integrate to your system#

Launch imu_monitor nodes and state of IMUs will be published to /diagnostics topic. If you have three or more IMUs in your system, you can continue to run your system even when one of IMUs is anomalous by avoiding using a IMU with anomalous_imu_id. Although you cannot distinguish which IMU is anomalous if you have only two IMUs in your system, you can use information that any of IMUs is anomalous for a fail-safe.

Tune Parameters#

The deviation_threshold in imu_anomaly_detector.param.yaml should always be set to a value larger than imu_outlier_deviation when IMUs are in normal states. The default parameter is set to almost valid value. If anomaly is detected unintendedly, change the threshold according to the max value of imu_outlier_deviation with imu_anomaly_analyzer. 

# While running your ROS nodes
ros2 launch imu_anomaly_analyzer imu_anomaly_analyzer.launch.xml
# [imu_anomaly_analyzer-1] Max imu_outlier_deviation: 0.001347
In this case, you should set deviation_threshold more than 0.001347, double value of imu_outlier_deviation would be a safe value(e.g. 0.002694)

If you want to check anomaly is properly detected in a anomalous condition, you can create anomalous IMU messages with imu_anomaly_simulator. You can check whether anomalies are detected properly by comparing output/is_anomaly_detected topic from analyzer and output/is_anomaly topic from simulator. Plot tools such as PlotJuggler would be useful for comparison. 

# While running your ROS nodes
ros2 launch imu_anomaly_simulator imu_anomaly_simulator.launch.xml

# In another terminal
ros2 launch imu_anomaly_analyzer imu_anomaly_analyzer.launch.xml

# In another terminal
ros2 bag record /imu_anomaly_simulator/is_anomaly /imu_anomaly_analyzer/is_anomaly_detected

# Open recorded rosbag with PlotJuggler and compare `is_anomaly_detected`(detection result) and `is_anomaly`(ground truth).