Robot not able to move if move_base goal is in its back

[cdondrup]

See this picture as an example:

The robot was stuck there unable to move. The real problem is that move_base happily produces new paths and does not respond with an error which could trigger recovery behaviours. No real idea what could cause that or how to fix it but @Jailander ask me to put it here to discuss the issue.

[Jailander]

@bfalacerda @nilsbore have you seen this?? any ideas, it seems to be a purely move_base issue not sure what to do here, except adding another monitor for timing out navigation when the robot doesn't move for a long time?

[bfalacerda]

if the robot is not rotating then he should be producing dwa planner failures. I have never seen move_base getting stuck indefinitely. Are you sure there's not another problem? Is there anything being published to /cmd_vel?

Btw, if that's the zig zag corridor, I think you'll have a pretty tough time having the robot go through that edge without getting himself a bit more in the middle. The edge itself looks pretty close to the wall

[Jailander]

I have seen that in both simulation and in the robot now :/ I think it happens when the goal is exactly 180 degrees from the robot pose, maybe the WayPoints are too straight ;)

Funny enough that is not the "zig zag" corridor, in that corridor the robot navigates on the middle of it since its narrow, the problem is in wider corridor where we have to stick to the walls anyway

[nilsbore]

Haven't seen this before I think. One thing to look for next time this happens is the value of min_vel_x of the planner. If that is not reset properly by backtracking I guess this could happen.

[nilsbore]

But then it should move backwards so I guess not.

[Jailander]

Ok, I'll try to reproduce and comment again

[bfalacerda]

I just tried to replicate that in simulation, giving it a goal that is exactly behind the robot (got him into 0 rotation, and then reduced the x value for the new goal), and he does it properly

[cdondrup]

If seen that twice at aaf already now... Next time it happens I'll have a look at the parameters and cmd_vel