Informed search · 7 min read
Jump Point Search
Skips straight corridors, expands only at turns. An A* Search optimisation that jumps straight corridors and only expands at forced turns, then rasterises jumps into steps.
Why this category
Search that orders expansion using extra knowledge: path cost (including soft patrol risk), straight-line distance to the exit, jump pruning, or both.
How it explores
Exploring: 2 of 31 states checked
Why this category
Jump Point Search is Informed. It is a pruned form of A* Search on uniform-cost grids: the same distance-to-exit heuristic, plus symmetry pruning that skips cells in open corridors until a forced neighbour or the goal appears.
Learned death cells at any phase are walls for the spatial plan. Soft patrol proximity is not part of the jump scoring in this demo — the information is heuristic plus pruning.
How it picks the next move
From the current jump point, scan each cardinal direction until blocked, until the exit, or until a forced neighbour (a wall beside the travel axis with an open step past it). Those jump endpoints enter an A* Search-style open set keyed by g + h.
When the exit is reached, the list of jump points is rasterised into orthogonal tick-by-tick moves that still respect the death calendar during the walk.
What it stores after a fail
Identical calendar: "x,y,phase" from the catching cone. Next Jump Point Search rebuilds the spatial unsafe set from every phase and plans again.
A corridor that only dies at one phase still looks permanently blocked to Jump Point Search here, which is stricter than space-time A* Search. Surviving a cell still teaches nothing.
More informed search
Try it live
Run this algorithm on a built-in level, or paint your own grid and guards.