#TinyMe: building a smart NPC engine for Minecraft

8 messages · Page 1 of 1 (latest)

wise cargo
#

I’ve been building an internal decision engine and integrating it into my Minecraft mod, TinyMe.

This has been a large project for me, and it took a long time to design and structure. TinyMe is the first live implementation of that work.

The current goal is to keep improving TinyMe as an advanced NPC companion mod rather than a simple scripted follower. I also plan to move development to the latest Minecraft versions and keep supporting newer updates going forward.

The recent deobfuscation changes are a good sign for long-term development, and they make continued work on the engine more practical.

Over time, I want to expand the engine so it can support other kinds of mods that need more advanced and adjustable AI, including other companion systems, horror mods, and boss-focused mods.

If anyone wants to follow the progress, the mod page is on Modrinth:

TinyMe on Modrinth: https://modrinth.com/mod/tinyme

Current release: TinyMe 0.1 Alpha

Modrinth
TinyMe - Minecraft Mod

TinyMe. (Intelligent Adaptive Companion)

winter haven
#

how does the decision tree work? i’m not super familiar with machine learning

wise cargo
#

@winter haven

It’s not machine learning.

The NPC reads the current situation, turns it into a set of weighted options, and then picks the action with the best overall score.

Very simply, it does this every tick:
reads the world state
distance, enemies, health, terrain, owner state, items, danger, etc.
builds context
for example: “owner is in danger”, “this enemy is the highest threat”, “position is unsafe”, “food is needed”
scores possible actions
things like protect, attack, retreat, reposition, rescue, follow, hold, or use an item
chooses the best plan
not just a single random action, but the most reasonable behavior for that moment

So it’s closer to a weighted decision system than a traditional scripted follower, and it’s also not an ML model that learns from data.

winter haven
wise cargo
#

@winter haven

Yeah, but not in the machine learning sense.

By “develops a personality,” I mean its behavior can gradually take on a clearer style over time for example becoming more cautious, more aggressive, more protective, or more responsive depending on context, priorities, equipment, and future systems tied to long-term behavior. So it’s more of an evolving behavioral profile than an ML trained personality.

winter haven
wise cargo
#

@winter haven

No, not in the ML sense.

An ML system learns from data by updating a model. This NPC doesn’t do that. It uses hand built decision logic, weighted priorities, memory, and behavior rules. So if its style changes over time, that’s structured behavioral adaptation, not machine learning.

winter haven