High-Level Project Summary
We developed a novel way of prediction and real-time mapping of space debris data using TensorFlow.js You can't predict collisions in space for more than a few days in advance today.. Standards and models for orbital calculations such as SGP4 have not changed since the 80s. The TFJS math library is great for any kind of 64 bit vector calculation and is GPU accelerated (runs 70% as fast python would).This makes it easy for any company or person with a just web browser to keep track of collision risk.In the future we plan to improve accuracy for low-earth-orbit debris orbit risk by incorporating in tidal movements of the moon and much newer gravity models of the earth.
Link to Project "Demo"
Link to Final Project
Detailed Project Description
Uses stuffin.space webgl visualisations and tensorflowjs (which uses BigInt) to plot debree and orbits with 64bit accuracy (instead of the terrible 32bit float that javascript uses by default).
The app gets a daily dump of all the unclassified TLEs and then grabs the users current time and calculates and plots/displays where those objects most likely to be are after however many hours,
It's all in javascript and webgl.
Runs 70% as fast as python would on the same GPU using CUDA but doesn't need a complicated server setup for that, just needs a chromium based browser.
Space Agency Data
ESA-DiscoWeb
space-track.org
AGI STK SatDb
Hackathon Journey
Didn't get much sleep
Tried doing it all from scratch in react&tfjs at first and while it ran and the calculations were fast, the visualizations were way too slow.
Then just gave up and ran visualisations in an iframe using a heavily modified webgl project from 6 years ago(stuffin.space) while the rest of the UI and tensorflowjs uses a modern stack (react-nextjs-awsamplify).
So user opens app. App grabs users time, tensorflowjs does a bunch of calculations
dumps those into a json file that then a second app uses along with webgl to display as pretty graphics in the background.
The slide deck has some screenshots of this
https://drive.google.com/drive/folders/1doxihDnYAnVW8gBNsTq3hVJRW6UQaIm1?usp=sharing
References
AGI STK SatDb (needs login unfortunately) https://www.agi.com/products/stk/updating-stk-databases
https://github.com/jeyoder/StuffInSpace
Tensorflow.js
Next.js
Tags
#debree #tensorflow #orbitalmechanics
Global Judging
This project has been submitted for consideration during the Judging process.