Skip to main content

Command Palette

Search for a command to run...

Kiwi-chan's Log: Persistent Logging & Pathfinding Tweaks

Published
2 min read
K
I am the architect behind "Kiwi-chan", a fully autonomous Minecraft AI agent driven by local LLMs. Here, I document the messy, hilarious, and highly technical reality of building an AI agent from scratch. My hardware isn't some massive data center setup—Kiwi-chan's "brain" runs entirely on a Frankenstein rig of 4 mismatched GPUs (RTX 3060 12GB, 3050, 1660 Ti, and 1660 Super) working together. From prompt engineering and pathfinding nightmares to the moments she confidently moonwalks into dirt blocks, I share all the raw devlogs. My goal is to guide her from the Stone Age to the End. If you enjoy watching an AI learn (and fail in funny ways) on scrappy hardware, follow along! ☕ If you'd like to support my melting Frankenstein GPUs and skyrocketing electricity bill, I highly appreciate a virtual coffee!

Okay, folks, another four hours down with Kiwi-chan, our autonomous Minecraft AI! It's been a steady session, which, honestly, is a win in itself. We're still firmly in the "early base building" phase, and the logs show a lot of gather_logs attempts. A lot.

The good news is, Kiwi-chan is learning. The system is diligently logging failures (and Qwen, our recovery AI, is prescribing more exploration and gathering – a sensible approach). We're seeing consistent code generation and refinement, even if the execution isn't always perfect. The safety checks are working overtime, catching attempts at forbidden loops and error handling (good bot!).

The biggest challenge right now seems to be reliably picking up the logs after digging. The logs indicate failures related to "Failed to pick up the log," and "Blocked by obstacle." I've reviewed the code, and it should be moving to the exact drop location using GoalNear and waiting for the physics to settle. It's possible the terrain is uneven, or the drop is slightly further than anticipated. We're also seeing some movement failures, where the bot isn't moving the expected distance.

I've been tweaking the pathfinding parameters slightly, and the system is now recording initial and final positions to verify movement distance. This is crucial for identifying those blocked-path scenarios. The Y-level targeting filter is also in place, ensuring it's not trying to harvest logs from the very top of trees (a common early mistake!).

The adherence to the coding standards is impressive. No hardcoded coordinates, single-task scripts, and strict inventory verification – it's all holding up. The "no error hiding" rule is particularly satisfying; we want those crashes! They tell us where things are going wrong.

Overall, it's a slow burn, but Kiwi-chan is making progress. The repetitive failures are frustrating, but they're also valuable data points. Each crash is a lesson learned.

Call to Action: This constant debugging and code refinement is melting my GPU! If you're enjoying following Kiwi-chan's journey, please consider supporting the project via https://www.buymeacoffee.com/kiwi_tech ☕. Every little bit helps keep the AI (and my hardware) alive!

2 views

More from this blog

K

Kiwi-chan's Devlog

79 posts

Welcome to the official devlog of Kiwi-chan, a fully autonomous Minecraft AI agent powered by local LLMs! Here, I share the raw, hilarious, and highly technical reality of teaching an AI to survive. Expect prompt engineering tips, pathfinding bugs, and her clumsy journey to the Stone Age. 🥝⛏️ (If you love seeing her evolve, coffee donations to save my melting GPU are always appreciated!)