As a starting point, it’s already interesting to think about what I will call “White Sugarlawn”: a simpler, more refined variant of Sugarlawn with some of the more annoying (from a computational perspective) gameplay features removed:
- the room graph is as in the real game, but movement is unobstructed: no locked doors requiring an extra move to unlock, no chicken suit issues, etc;
- you have 120 turns, with each turn consisting of one of the following three actions: moving from your room to an adjacent room, picking up any number of items (subject to the carry limit) from the current room; dropping any number of held items into the current room;
- all loot items are in plain sight (in the starting location listed in the table here: Sugarlawn Hints and Walkthrough) and can be picked up without complications (no requirement to unlock safes, etc);
- items can only be carried (not worn) and all items are subject to the carry limit;
- no partial credit for bringing an item to the Foyer; they must be dropped in the correct room to score any points;
- no exit or secret bonuses.
The saving grace, when it comes to algorithms for solving White Sugarlawn, is that the room graph is far from worst-case: it is very sparse, with only two nontrivial cycles. That said my first attempt at a solution was stopped in its tracks by the realization that it can be optimal to “park” items you’ve picked up in some intermediate room to save inventory space, and then come back to pick them up later. The naive state space thus contains on the order of 50^50 states which is utterly intractable.