Fix premature mission termination in mapless coverage navigation

[Copilot]

is_target_safe() treats unknown costmap cells as unsafe, which is wrong for mapless rolling-window navigation. Since most of the costmap starts unobserved (cost=-1), nearly all targets get rejected → retry limit exhausted → mission ends immediately without visiting any waypoints.

is_target_safe() — core fix

The old logic had two problems:

• Returns False if any neighbor cell is out of the costmap grid (cost is None → return False)
• Requires at least one known cell (return has_known_cells), so targets in unobserved space always fail

# Before: unknown space = unsafe
if cost is None:
    return False        # out-of-bounds neighbor → immediate reject
if cost < 0:
    continue            # skip unknown
has_known_cells = True
...
return has_known_cells  # False when all cells unknown

# After: unknown space = potentially free (let move_base handle actual path safety)
if cost is None:
    continue            # skip out-of-bounds neighbors
if cost < 0:
    continue            # skip unknown cells
if cost >= self.target_cost_threshold:
    return False        # only reject on known obstacles
return True

This is consistent with NavfnROS/allow_unknown=true already set in the launch config.

Other changes

• Costmap settling: Configurable costmap_settle_time param (default 3s, was hardcoded 1s) — gives the lidar time to populate the rolling window before target generation
• Recovery behaviors: Enabled recovery_behavior_enabled=true in move_base so it can recover from stuck states
• Retry improvement: Refresh targets from updated costmap after clearing costmaps during retry
• Diagnostics: Log initial target census (total/pending/active) at mission start

Tests

14 unit tests for is_target_safe(), target generation, and sweep ordering — runs without ROS via lightweight stubs.

Original prompt

I would like to understand the navigation stack super well right now... in the when i run it, there seems to be a problem where it doesnt navigate to waypoints and then stop, and it always finishes prematurely. Help me review the codebase and help me understand the navigation logic so that i can make a reliable navigation. It is starting in relatively the middle of a 2 x 2 walled area is where i am testing it and there are 2 plant obstacles and it should try to get relatively good coverage , right now i think it thinks its unsafe even though lots of space to move

---

📱 Kick off Copilot coding agent tasks wherever you are with GitHub Mobile, available on iOS and Android.

[mtaruno]

give me all the paramrters i can try tuning and an explanation of what the parameter is when testing. Do u think i shd just try to tune parameters and eventually it would run well?