Revision history [back]

It's this really annoying regression in Gazebo: https://github.com/ros-simulation/gazebo_ros_pkgs/issues/612 . Basically everyone using ros_control (and many other plugins) + Gazebo is affected by this, and it looks like it won't get fixed in ROS Kinetic / Gazebo 7. :(

For workarounds, see the issue I've linked to.

It's this really annoying regression in Gazebo: https://github.com/ros-simulation/gazebo_ros_pkgs/issues/612 . Basically everyone using ros_control (and many other plugins) + Gazebo is affected by this, and it looks like it won't get fixed in ROS Kinetic / Gazebo 7. :(

For workarounds, see the issue I've linked to.

It's this really annoying regression in Gazebo: https://github.com/ros-simulation/gazebo_ros_pkgs/issues/612 . Basically everyone using ros_control (and many other plugins) + Gazebo is affected by this, and it looks like it won't get fixed in ROS Kinetic / Gazebo 7. :(

For workarounds, see the issue I've linked to.

The problem is that there is a regression in the Joint::SetPosition() method in Gazebo. Previously, it only did just that (it set the joint angle). Later, the implementation of that method was changed so that it also reset the link velocity of the link connected to the joint to zero - this is why your robot is falling so slowly: Gravity pulls the robot down, but the joint controller always involuntarily stops that falling motion when it sets the joint angle. Since the gravity force now doesn't work properly, the mobile base loses proper contact to the ground (you can see that when you visualize contacts in the Gazebo GUI); this is why the base doesn't properly respond any more.

If you do any of the workarounds in the issue (adding pid_gains, using the VelocityJointInterface or EffortJointInterface instead of the PositionJointInterface), the ros_control Gazebo plugin won't call the Joint::SetPosition() method any more, thereby avoiding the problem.

Please note that you also have to do the workaround for any other Gazebo plugin that might call the Joint::SetPosition() method, such as the mimic joint plugin.

It's this really annoying regression in Gazebo: https://github.com/ros-simulation/gazebo_ros_pkgs/issues/612 . Basically everyone using ros_control (and many other plugins) + Gazebo is affected by this, and it looks like it won't get fixed in ROS Kinetic / Gazebo 7. :(

For workarounds, see the issue I've linked to.

The problem is that there is a regression in the Joint::SetPosition() method in Gazebo. Previously, it only did just that (it set the joint angle). Later, the implementation of that method was changed so that it also reset the link velocity of the link connected to the joint to zero - this is why your robot is falling so slowly: Gravity pulls the robot down, but the joint controller always involuntarily stops that falling motion when it sets the joint angle. Since the gravity force now doesn't work properly, the mobile base loses proper contact to the ground (you can see that when you visualize contacts in the Gazebo GUI); this is why the base doesn't properly respond any more.

If you do any of the workarounds in the issue (adding pid_gains, using the VelocityJointInterface or EffortJointInterface instead of the PositionJointInterface), the ros_control Gazebo plugin won't call the Joint::SetPosition() method any more, thereby avoiding the problem.

Please note that you also have to do the workaround for any other Gazebo plugin that might call the Joint::SetPosition() method, such as the mimic joint plugin.

Edit 2018-03-21: Now there is a third option to avoid this bug (the first two being (1) adding pid_gains and (2) using the VelocityJointInterface or EffortJointInterface. ). Option (3) is to install Gazebo 9 (from deb packages, or from source) and the newest version of gazebo_ros_pkgs (from source until 2.5.15 is released). This bugfix has been added in PR 688, but only for Gazebo 9.