AI: Steering & Pathfinding

Game AI isn’t about being smart — it’s about being fun to play against. Most enemies are a state machine (lesson 15) plus two skills: moving convincingly toward or away from things, and finding a path around obstacles.

Steering: moving like a living thing

The vector recipe from lesson 5 (“subtract, normalize, scale”) gives a desired velocity. Instead of setting velocity to it directly, steering behaviours nudge the current velocity toward it (a capped acceleration — lesson 6). The result is smooth curves and natural-looking overshoot.

const desired = scale(normalize(sub(target, me.pos)), MAX_SPEED)  // seek
let steer = sub(desired, me.vel)                                   // how to get there
steer = limit(steer, MAX_FORCE * dt)
me.vel = add(me.vel, steer)

Pathfinding: getting around walls

Seek walks straight into walls. To navigate a maze, treat the tilemap (lesson 12) as a graph: each walkable cell is a node, connected to its neighbours. Then search it:

• Breadth-first search (BFS) explores outward in rings. Guaranteed shortest path when every step costs the same — but it explores in all directions.
• Dijkstra is BFS with different step costs (mud costs 3, road costs 1).
• A* is Dijkstra plus a heuristic: a guess of the remaining distance (e.g. Manhattan distance). It explores toward the goal first and usually checks far fewer cells.

// A*: always expand the open cell with the lowest f = g + h
//   g = steps from the start so far,  h = estimated steps to the goal
open = [start]
while (open.length) {
  const cur = popLowest(open, c => g[c] + manhattan(c, goal))
  if (cur === goal) return rebuildPath(cameFrom, goal)
  for (const n of neighbours(cur)) if (!blocked(n) && !seen(n)) {
    cameFrom[n] = cur; g[n] = g[cur] + 1; open.push(n)
  }
}

Putting it together: an enemy brain

switch (enemy.state) {                       // FSM from lesson 15
  case 'patrol': followWaypoints(); if (canSee(player)) enemy.state = 'chase'; break
  case 'chase':  followPath(aStar(enemy, player)); if (near(player)) enemy.state = 'attack'; break
  case 'attack': swing(); if (!near(player)) enemy.state = 'chase'; break
}