Engine Lab

Raw JSON harness

The engine loads a world and an agent's brain from data, then loops: observe → decide → act → update → render. Every layer of the brain is editable below — change goals, states, rules, functions, or memory, press Apply & Reset, and the same engine executes what you wrote.

WorldBrain

Speed1/s

wallvisitedgoalstart▶ robot

Debug panel⌄Idle

Decision pipeline

Tick0

Active goalreach-goal

Active stateexplore

Triggered ruleexplore-advance

Executed functionstep

Returned actionmoveForward

Why this action?

WHEN !wallAhead

This was the first top-to-bottom rule whose WHEN condition was true and whose next action was executable.

World observation

Position(1, 1)

DirectionE

Wall ahead / left / rightno / yes / yes

Goal visibleno

Visited tiles1

Energy400

Holding on goal0 / 3

Memory1

notesLeft-hand wall follower. Memory is unused by this brain but is loaded and shown in the debug panel to prove the channel works.

rules.json — Reactions: when → do, plus optional state / setState / priority. Condition flags: wallAhead, wallLeft, wallRight, goalVisible, goalAhead, atGoal, visitedAhead, visitedCount, energy, tick, always, never. Available bricks: .

{
  "rules": [
    {
      "id": "settle-at-goal",
      "when": "atGoal",
      "do": "hold",
      "priority": 200,
      "description": "Standing on the goal — hold position until the arrival is confirmed. Applies in every state."
    },
    {
      "id": "explore-spot-goal",
      "state": "explore",
      "when": "goalAhead",
      "do": "step",
      "setState": "seek",
      "priority": 100,
      "description": "Goal came into straight view — switch to seeking and step toward it."
    },
    {
      "id": "explore-advance",
      "state": "explore",
      "when": "!wallAhead",
      "do": "step",
      "priority": 60,
      "description": "Path ahead is clear — keep moving forward (straight beats turning)."
    },
    {
      "id": "explore-hug-left",
      "state": "explore",
      "when": "!wallLeft",
      "do": "pivot-left",
      "priority": 50,
      "description": "Blocked ahead but the left opens up — turn into it (left-hand bias)."
    },
    {
      "id": "explore-turn-right",
      "state": "explore",
      "when": "!wallRight",
      "do": "pivot-right",
      "priority": 40,
      "description": "Blocked ahead and left — the right is open, turn there."
    },
    {
      "id": "explore-dead-end",
      "state": "explore",
      "when": "always",
      "do": "pivot-right",
      "priority": 10,
      "description": "Dead end — turn around (two right turns over two ticks)."
    },
    {
      "id": "seek-advance",
      "state": "seek",
      "when": "goalAhead",
      "do": "step",
      "priority": 100,
      "description": "Goal still straight ahead — keep moving toward it."
    },
    {
      "id": "seek-lost",
      "state": "seek",
      "when": "always",
      "do": "pivot-right",
      "setState": "explore",
      "priority": 10,
      "description": "Lost sight of the goal — resume corridor-following."
    }
  ]
}

Edit any layer, then Apply & Reset — the same engine runs your brain.