control_performance_analysis

Purpose

control_performance_analysis is the package to analyze the tracking performance of a control module and monitor the driving status of the vehicle.

This package is used as a tool to quantify the results of the control module. That's why it doesn't interfere with the core logic of autonomous driving.

Based on the various input from planning, control, and vehicle, it publishes the result of analysis as control_performance_analysis::msg::ErrorStamped defined in this package.

All results in ErrorStamped message are calculated in Frenet Frame of curve. Errors and velocity errors are calculated by using paper below.

Werling, Moritz & Groell, Lutz & Bretthauer, Georg. (2010). Invariant Trajectory Tracking With a Full-Size Autonomous Road Vehicle. Robotics, IEEE Transactions on. 26. 758 - 765. 10.1109/TRO.2010.2052325.

If you are interested in calculations, you can see the error and error velocity calculations in section C. Asymptotical Trajectory Tracking With Orientation Control.

Error acceleration calculations are made based on the velocity calculations above. You can see below the calculation of error acceleration.

Input / Output

Input topics

Name	Type	Description
/planning/scenario_planning/trajectory	autoware_auto_planning_msgs::msg::Trajectory	Output trajectory from planning module.
/control/command/control_cmd	autoware_auto_control_msgs::msg::AckermannControlCommand	Output control command from control module.
/vehicle/status/steering_status	autoware_auto_vehicle_msgs::msg::SteeringReport	Steering information from vehicle.
/localization/kinematic_state	nav_msgs::msg::Odometry	Use twist from odometry.
/tf	tf2_msgs::msg::TFMessage	Extract ego pose from tf.

Output topics

Name	Type	Description
/control_performance/performance_vars	control_performance_analysis::msg::ErrorStamped	The result of the performance analysis.
/control_performance/driving_status	control_performance_analysis::msg::DrivingMonitorStamped	Driving status (acceleration, jerk etc.) monitoring

Outputs

control_performance_analysis::msg::DrivingMonitorStamped

Name	Type	Description
longitudinal_acceleration	float	[m / s^2]
longitudinal_jerk	float	[m / s^3]
lateral_acceleration	float	[m / s^2]
lateral_jerk	float	[m / s^3]
desired_steering_angle	float	[rad]
controller_processing_time	float	Timestamp between last two control command messages [ms]

control_performance_analysis::msg::ErrorStamped

Name	Type	Description
lateral_error	float	[m]
lateral_error_velocity	float	[m / s]
lateral_error_acceleration	float	[m / s^2]
longitudinal_error	float	[m]
longitudinal_error_velocity	float	[m / s]
longitudinal_error_acceleration	float	[m / s^2]
heading_error	float	[rad]
heading_error_velocity	float	[rad / s]
control_effort_energy	float	[u * R * u^T]
error_energy	float	lateral_error^2 + heading_error^2
value_approximation	float	V = xPx' ; Value function from DARE Lyap matrix P
curvature_estimate	float	[1 / m]
curvature_estimate_pp	float	[1 / m]
vehicle_velocity_error	float	[m / s]
tracking_curvature_discontinuity_ability	float	Measures the ability to tracking the curvature changes [abs(delta(curvature)) / (1 + abs(delta(lateral_error))]

Parameters

Name	Type	Description
curvature_interval_length	double	Used for estimating current curvature
prevent_zero_division_value	double	Value to avoid zero division. Default is 0.001
odom_interval	unsigned integer	Interval between odom messages, increase it for smoother curve.
acceptable_max_distance_to_waypoint	double	Maximum distance between trajectory point and vehicle [m]
acceptable_max_yaw_difference_rad	double	Maximum yaw difference between trajectory point and vehicle [rad]
low_pass_filter_gain	double	Low pass filter gain

Usage

• After launched simulation and control module, launch the control_performance_analysis.launch.xml.
• You should be able to see the driving monitor and error variables in topics.
• If you want to visualize the results, you can use Plotjuggler and use config/controller_monitor.xml as layout.
• After import the layout, please specify the topics that are listed below.

• /localization/kinematic_state
• /vehicle/status/steering_status
• /control_performance/driving_status
• /control_performance/performance_vars

• In Plotjuggler you can export the statistic (max, min, average) values as csv file. Use that statistics to compare the control modules.

Future Improvements

• Implement a LPF by cut-off frequency, differential equation and discrete state space update.