search

ros-controls / ros2_controllers

Generic robotic controllers to accompany ros2_control
https://control.ros.org
Apache License 2.0
350 stars 314 forks source link

Closed loop diff_drive_controller #210

Open jonatanolofsson opened 3 years ago

jonatanolofsson commented 3 years ago

In gazebo, with varying friction parameters et.c., I can't for the life of me get a consistent behavior of the configuration of the diff_drive_controller. In fact, I struggle to get consistency between 1) reported wheel odometry 2) IMU measurements 3) Turn radius calculations from desired linear and angular rates.

The problems seem to come down to that the actual wheel speed is inconsistent with the requested wheel speed due to friction, slip et.c. The solution to this seems to be to close the loop (when open_loop=false) so that the requested wheel speed is attained. My thoughts was to add a PID to either left/right or linear/angular commands. Or am I missing something? Am I using it wrong? I no-one else experiencing these issues?

In the screenshot below, I hand-tuned the wheel multiplier to give a reasonable IMU response at 0.3 rad/s, but when issuing 0.15rad/s it's waay off. Gyro_vs_Wheels

Also - what's up with the multiplier parameters? I can't make any physical sense of them and by their performance they (at least wheel_separator....) seems as though it varies with angular speed. Are they just there in lieu of a PID, to "compensate" for errors?

shonigmann commented 3 years ago

Another potential source of discrepancy, beyond using open-loop commands, is that gazebo_ros2_control's position and velocity interfaces act as perfect command followers, with infinite accelerations and discontinuous behavior (e.g. if you set a wheel velocity, the wheel will be at that velocity on the next time step). I wouldn't be surprised if that was having an impact on the odometry. that said, I agree that closing the loop on the controller would probably help too.

there is an effort-interface based position controller in the works that I believe will offer more realistic behavior in Gazebo (and could be modified to be an effort-based velocity controller), without needing to tweak arbitrary multipliers, but it is a work in progress. I've been trying to build on it myself and have uncovered a number of other issues along the way, so progress has been slower than expected.

Timple commented 3 years ago

Would it be enough if the velocity loops around the wheels where closed?

In that case the diff_drive_controller can remain open-loop again. As then the wheels turn as expected.

destogl commented 3 years ago

Would it be enough if the velocity loops around the wheels where closed?

I think this is how most people are using it.

cangozpi commented 1 year ago

Hİ, I have the same issues as were initially described. Is there an already existing implementation/configuration for ros2 control's differential_drive that I can use to fix these issues ?