Revision history [back]

Fascinating problem. It looks like you have a few points which could be failing.

• Are you sure that the cliff is detected at the correct distance?
• If it is, is the costmap updated correctly?
• If it is, is the local planner not acting properly?

If all of the above work, I believe that some type of safety monitor is being introduced to the nav2 stack (see discourse).

Until then, I think the in-system way to solve something like this is to have a behavior(previously recovery) which looks for proximity to cliffs, and design the behavior tree such that it runs the follow_path unless the proximity_detector returns true, at which point it stops (and engages breaks, or actively publishes 0 velocity, or whatever).