Pathfinding
Algorithms
Seven ways to cross a patrolled grid. They all solve the same space-time problem; the category says what information shapes the next decision. Click a card for the full walkthrough of planning, walking, and learning from deaths.
Prefer a live level? Open the stealth demo or the level editor.
Naive
Naive
No graph search. Each tick follows a local rule (random pick, keep a hand on the wall, or ride a potential field) with no plan of the full route.
Naive
Random Walk
No graph search, no memory of where it has been, just a coin flip among safe neighbours each tick.
DetailsNaive
Wall Follower
A classic hand-on-the-wall reflex with a real guarantee on loop-free mazes, and no map when the walls braid.
DetailsNaive
Potential Fields
A local force field: attract to the exit, repel from death phases and walls, no graph search.
DetailsUninformed search
Uninformed search
Real search over space-time states, but blind to where the exit is. Order comes only from the frontier structure (queue, stack, or iterative deepening), not from a goal heuristic.
Uninformed search
Breadth-First Search
A first-in-first-out queue over space-time states that guarantees fewest ticks, with no goal heuristic and no soft sense of risk.
DetailsUninformed search
Depth-First Search
A stack instead of a queue: systematic but goal-blind, great at finding a route and poor at finding a good one.
DetailsUninformed search
Iterative Deepening Depth-First Search
Depth-First Search repeated with a rising depth limit so the first success is at the shallowest workable depth.
DetailsUninformed search
Bidirectional Search
Two Breadth-First Search fronts meet in the middle on the spatial grid, then the joined route is walked tick by tick.
DetailsInformed search
Informed search
Search that orders expansion using extra knowledge: path cost (including soft patrol risk), straight-line distance to the exit, jump pruning, or both.
Informed search
Dijkstra's Algorithm
A priority queue ordered by accumulated cost, including soft patrol risk, still without aiming at the exit.
DetailsInformed search
Greedy Best-First Search
Always expands the state closest to the exit, ignores sunk cost, and often plans fast while hugging danger.
DetailsInformed search
A* Search
Expand by cost-so-far plus distance-to-exit, respect soft patrol risk, and replan when death phases land.
DetailsInformed search
Jump Point Search
An A* Search optimisation that jumps straight corridors and only expands at forced turns, then rasterises jumps into steps.
DetailsInformed search
Theta*
A* Search with line-of-sight parent shortcuts, then Bresenham-style rasterisation into orthogonal moves.
DetailsInformed search
D* Lite
A teaching stand-in for D* Lite: reuse a cached route until the death map grows, then replan with A* Search.
DetailsSampling / stochastic
Sampling / stochastic
Sampling / stochastic. Grow random trees or roll out simulated futures instead of exhaustively expanding every state.
Learning
Learning
Learning. Update heuristics or action values across attempts from what failed before.
Constraint search
Constraint search
Constraint search. Resolve conflicts by adding bans and replanning (CBS-style).