Cannot start controller after stopping controller manually

[tkmtnt7000]

Hi. When I used https://github.com/fetchrobotics/robot_controllers/blob/010855f0e6a8e058a1cbd2ad69c324cd97f0d815/robot_controllers_interface/scripts/stop_controller.py to stop a controller and tried to start the controller by using https://github.com/fetchrobotics/robot_controllers/blob/010855f0e6a8e058a1cbd2ad69c324cd97f0d815/robot_controllers_interface/scripts/start_controller.py, I got the following error and did not start the controller.

fetch@fetch1075:~/ros/melodic/src/fetchrobotics/robot_controllers/robot_controllers_interface$ rosrun robot_controllers_interface stop_controller.py arm_controller/follow_joint_trajectory
[INFO] [WallTime: 1636960585.632305] [node:/stop_robot_controllers] [func:<module>]: Connecting to /query_controller_states...
[INFO] [WallTime: 1636960585.794209] [node:/stop_robot_controllers] [func:<module>]: Done.
[INFO] [WallTime: 1636960585.795474] [node:/stop_robot_controllers] [func:<module>]: Requesting that arm_controller/follow_joint_trajectory be stopped...
[INFO] [WallTime: 1636960585.797601] [node:/stop_robot_controllers] [func:<module>]: Done.
fetch@fetch1075:~/ros/melodic/src/fetchrobotics/robot_controllers/robot_controllers_interface$ rosrun robot_controllers_interface start_controller.py arm_controller/follow_joint_trajectory 
[INFO] [WallTime: 1636960615.499979] [node:/start_robot_controllers] [func:<module>]: Connecting to /query_controller_states...
[INFO] [WallTime: 1636960615.754447] [node:/start_robot_controllers] [func:<module>]: Done.
[INFO] [WallTime: 1636960615.756414] [node:/start_robot_controllers] [func:<module>]: Requesting that arm_controller/follow_joint_trajectory be started...
[ERROR] [WallTime: 1636960615.759133] [node:/start_robot_controllers] [func:<module>]: Unable to start arm_controller/follow_joint_trajectory

/var/log/ros/robot.log output

[ERROR] [1636960615.757763468] [/robot_driver:ros.robot_controllers.FollowJointTrajectoryController]: Unable to start, action server is not active.

https://github.com/fetchrobotics/robot_controllers/blob/010855f0e6a8e058a1cbd2ad69c324cd97f0d815/robot_controllers/src/follow_joint_trajectory.cpp#L145-L150 I wonder if checking isActive() is proper when starting contollers. I think isActive() shows status whether an active goal already exists or not. https://github.com/ros/actionlib/blob/10025114a28d94581b6fac68bd391eb2a1bc3d32/include/actionlib/server/simple_action_server_imp.h#L227-L236

Cc:@708yamaguchi @knorth55

[mikeferguson]

This is actually the expected behavior:

• The follow joint trajectory action requires an active goal to actually start - since otherwise it has no idea what to command the joints.
• To start the controller, send an action goal - it autostarts when a goal is received.
• Stopping the controller mainly exists because there is a "stop_with_action" parameter which defines whether this controller should "hold" the final position in the action or terminate immediately. The default is that "stop_with_action" is set to FALSE, so the arm will hold the final position in the trajectory (usually this is what you want, otherwise you are relying on the gravity compensation alone to hold the robot arm in that final position - which will usually fail if you are holding something heavy in the gripper). However, if the arm is holding that position rigidly, and you want to manually reposition the arm, you need to stop the controller to revert to pure gravity compensation (there is a button the joystick that allows you to do this as well - since it is pretty common when developing that you want to get the arm out of the way).
• The other reason to stop the controller would be to change modes - say you want to switch to a cartesian controller to move the arm a bit.

Some other controllers really should keep running all the time (base controller for instance publishes odometry, and you want that all the time, so the controller runs all the time) and so they can actually be started via script without commanding a goal.

[tkmtnt7000]

Thank you for your prompt response. I understand this is the expected behavior. I learned a lot.

What I was trying to do is moving freely from gravity compensation mode to position control mode and vice versa via script. I wondered why opposite commands do not start the controller and sending a trajectory is needed to start the controller, but now I clearly understand the reason.

Thanks a lot !