AI Based Decision Making System for Satellites.

Autonomous decision making system for satellites real-time decision making in orbit so missions run safer, longer, and with far fewer ground interventions to prevent from catastrophic failures.

Comment

Inspiration

I have personally worked at various private space and govt agencies and while working their i realized there was this gap between the current technology and what should have existed. My real life working experience inspired would rather say motivated to work on something like this.

What it does

It receives all the telemetry data and

  1. Prevent from any failure to happen while monitoring all the sensors, gauges, path, altitudes.
  2. In case a failure happens it uses AI to take decision to prevent from any catastrophic failures and get accomplish the mission by keeping ultimate mission success at its goal.

How we built it

Dataset : We have create our own data set keeping different scenarios and conditions in place. For example 1 dataset is of only sensors available on ISS and another one is more of a generic dataset of all the possible errors that it could

Frontend HTML5/CSS3/JavaScript - Core web technologies Three.js (v0.157.0) - 3D graphics rendering for ISS visualization ES modules with importmap for modern JavaScript GLB loader for 3D models Interactive 3D ISS model with clickable components Chart.js - Data visualization and telemetry charts

Backend Technologies Python Flask - Web framework for REST API Flask-CORS - Cross-origin resource sharing Groq API - AI/LLM integration using openai/gpt-oss-20b model RAG (Retrieval Augmented Generation) - Context-aware AI responses ChromaDB - Vector database for embeddings (infrastructure present) JSON-based Knowledge Base - ISS components and fault databases

Challenges we ran into

  • Compute : We actually tried to train the model but didnt had enough compute. Our reuquest for H100 was rejected by GCP. So we were not able to train our model so then went for rag based application.
  • Dataset: Quality dataset was not available so we created our own dataset which took our 50% of time.

Accomplishments that we're proud of

We are actually proud of how it turned out. You can play with the simulation and inject errors and it rectifies it. Also if control through the control panel and any abnormality is detected it blocks off human control and take control of it.

What we learned

I have actually learned a lot about building with rag.

What's next for AI Based Decision Making System for Satellites.

We were able to sucesffully run it on rasberry pi. Now i want to add gyroscopes and thermal sensors and motors. So for example when it feels roll it uses motors to adjust the thrust on its own. Means all the data should be real time.

Built With

Share this project:

Updates