Avoid nan command to be processed

[vincentrou]

No description provided.

[bmagyar]

Ouch, great find!
Could you please also add a test that fails if this feature is not implemented?

[vincentrou]

When I send a twist with a NaN in the linear.x, the robot goes backward full speed in gazebo and in real.
But I do not reproduce it with the test class... (published odom always at zero)

[bmagyar]

This will need a rebase and that test. Sorry but I'd really prefer not to add any fixes that are not backed up by a test.

[bmagyar]

I tried to run the test without the fix and I can confirm that this test doesn't really catch the issue.

Could you please try setting the param publish_cmd and see what gets through there?

For more: https://github.com/ros-controls/ros_controllers/blob/kinetic-devel/diff_drive_controller/src/diff_drive_controller.cpp#L292

[bmagyar]

Also we've got a conflict. Could you rebase the PR on the latest kinetic-devel please?

[bmagyar]

Ping @vincentrou

[vincentrou]

I rebase the PR on kinect in december but I will not have time to do more on this.

[martiniil]

I picked these commits for melodic. However I see a problem: The robot gets stuck during execution of the previous test case testNaN here

  // stop robot (will generate NaN)
  stop();

Due to

        const double lp = left_wheel_joints_[i].getPosition();
        const double rp = right_wheel_joints_[i].getPosition();
        if (std::isnan(lp) || std::isnan(rp))
          return;

the commands are no longer updated even after start(). I am not sure what the intended behaviour (of the controller) is here.

[martiniil]

I tried to run the test without the fix and I can confirm that this test doesn't really catch the issue.

Could you please try setting the param publish_cmd and see what gets through there?

For more: https://github.com/ros-controls/ros_controllers/blob/kinetic-devel/diff_drive_controller/src/diff_drive_controller.cpp#L292

The test passes since the command is never updated during this test case. See my comment above.

[bmagyar]

Closing this in favour of #469