Costmap Safety Check Doesn't Quite Work as Expected

[Jmeyer1292]

Greetings. MoveBaseFlex is an excellent piece of software. Thank you for the hard work in creating and maintaining it.

I wanted to report an issue: In brief the safetyChecks are not working as expected.

• Setup: Robot with MoveBaseFlex and a Time-Elastic-Band local controller
• Experiment: Set an expected sensor data rate in your costmap parameters. Do NOT publish sensor data to the costmap.
• Expectation: The robot does not move. If sensor data is provided, it would keep going.
• Observation: The robot stutters and slowly creeps forward. See figure 1 for a visualization of the bag file.

Blue is a velocity command (linear/x). Red is measured robot twist (linear/x). Figure 1

A very quick glance through the source code shows this section of code as the point where our checks are applied.

It appears that the safety check fails (as expected), a zero velocity is published, and the thread sleeps. It then proceeds to go and run the controller and issue a command anyway. I suggest the inclusion of a continue in the loop where the safety checks fail.

I'm happy to make this edit, but I acknowledge that this may have been done on purpose. I can certainly see the value in still running controllers at regular intervals and just letting them know that things are kinda broken.

Again thank you for being open source maintainers.

[corot]

Thanks for the thorough analysis. I agree, at first glance, looks like a continue; is missing. I'll try myself and see if there's any good reason for not having it (we wrote that logic long time ago, and I don't remember the rationale)