Kerbal Space Program for Project AIRI

Note

This project is part of the Project AIRI, we aim to build a wishing to achieve Neuro-sama's (subscribe if you didn't!) altitude, completely LLM and AI driven, capable of realtime voice chat, Minecraft playing, Factorio playing. If you are interested in, please do give it a try on live demo.

demo-1.mp4

Getting Started

Install Kerbal Space Program from Steam or other places.

Install mods

./hack/mod-installer

Bootstrap DMP server

Download and unpack the DarkMultiPlayer server into this project's ./mods directory:

./hack/mod-dmp-server

That script downloads DMPServer.zip to ./mods/DMPServer.zip and unpacks the server into ./mods/DMPServer.

If you want to run the container directly, use the project-local path instead of a hardcoded absolute path:

docker run -it -p 6702:6702 -v "$(pwd)/mods/DMPServer:/opt/dmp-server" mono:latest mono /opt/dmp-server/DMPServer.exe

Run that command from the repository root so $(pwd)/mods/DMPServer resolves correctly.

You can also use Docker Compose from the repository root:

docker compose up dmp-server

Configure agents & DMP Server, also the kRPC server

cp .env.example .env

TODO

Agent

• 1. Chinese input cannot be cleared clearly, we should fix the TUI to interact with the agent.
• 2. More tools? Perhaps a way to interrupt the wait loop with wait_seconds tool.
• 3. We should have ability to notify the agent, if the navigation computer calculated that the spacecraft is on a collision course, or the spacecraft is on an escape trajectory, or the spacecraft is on a stable orbit, etc. So that the agent can learn how to deal with those situations, and make decisions based on those information.
• 4. Time-interval related improvements, currently, the agent cannot understand what is going to happen in the future, and what has happened in the past, and how to plan for the future, and how to learn from the past. We should improve the agent's ability to understand time, and perhaps, we can use a simple time-interval tool, that can be used to notify the agent about the current time, and the time of the next event, etc.

Context

• 1. Offer more context and planetary information to the agent.
  • 1. Planetary orbit, sizes, radius, and basic properties.
  • 2. Orbit, Ap, Pe.
  • 3. Orbit status, MSL (mean sea level), for example, coliision? impact? escape? stable?
  • 4. Navigation, including a set of basic math tools, and perhaps physics tools, the agent is required to learn how to write code and probably use the code interpreter or any form of runtime that executes the code, to auto-research out how the mission could be accomplished, and then execute the plan in order.
  • 5. The agent should be able to use the giving tools, and scripts written by itself, to calculate delta v required for the mission, and perhaps, how the agent could decide and judge where to go, and how the spacecraft should be designed, and how the mission should be planned.
  • 6. Basic memory
    • 1. Tasks / TODO / target / objective, and probably.
    • 2. The agent should be aware of other spacecraft, and scientific data collected by those already launched spacecraft.
• 2. Realistic mode
  • 1. We will not offer the basic known properties, including the altitude of the atmosphere, the radius of the planet, and the escape velocity also delta v for one place to another.
  • 2. And for agent that plans to land spacecraft to the planet, mountain, and geology exploration and scouting is required, and the agent should somehow, learn this.
  • 3. We could simulate communication jams, and perhaps data jittering, or data corruption, for the agent to learn how to deal with those situations, by itself.