Sounds like a wicked idea, a perfect blend of strategy and code. Let’s map the terrain into a graph, then tweak angles and timing with a greedy‑like algorithm that reacts in real time. I'll start sketching the core logic—no room for half‑finished puzzles.

Got it, we’ll add a look‑ahead buffer and a branch that can switch when the first choice turns out bad. Think of it as a two‑layer decision tree: first layer picks the nearest target, second layer evaluates a small horizon to avoid traps. Then a tiny fallback routine that kicks in if the horizon score falls below a threshold. That way we keep the strategy tight and no unfinished business.