[huxiaoyi-ovo]

Summary

Potential-field planning crashes on an obstacle-free map because the field bounds call min(ox) / max(ox) on empty lists and the repulsive potential tries to index the nearest obstacle even when none exists.

This change:

• Builds potential-field bounds from the start, goal, and any obstacles, so an empty map remains valid.
• Returns zero repulsive potential when there are no obstacles.

The behavior for maps with obstacles is unchanged.

Regression tests

• A map without obstacles has zero repulsion and a finite, purely attractive potential field.
• An obstacle-free end-to-end planning call progresses from (0, 0) toward (5, 5) and reaches within one grid resolution of the goal.
• The original obstacle-based example remains in the existing tests.

Scope: 2 files (the planner and its tests), no API changes or new dependencies.

CI validation: fork PR #3 (Linux/macOS/Windows jobs launched).

AI assistance: ChatGPT (OpenAI GPT-6) helped inspect the edge case, implement the fix, and prepare regression tests.