Inspiration
We wanted to build palantir for space
What it does
Each US satellite has its own agent which decides whether it should take an action (manoeuvre away, decoy, destroy etc.) based on Bayesian inference predictions on how threatening a satellite from China or the USSR appeared due to their actions (moving close to a US satellite, becoming geostationary above the US, etc).
How we built it
Use Bayes’ Theorem to calculate the probability that something malicious was happening given the observations from live satellite data. Calculated each of these probabilities for a number of Chinese and Russian satellites to figure out their threat probability on some select US satellites. Implemented a number of observations and aggregated risk scores to create a risk model. Flagged the satellites most likely to cause harm to a US satellite agent allowing it to make a decision autonomously.
Challenges we ran into
Spent ~4-5 hours validating if any of this was a real problem. Contacted the CEO of Loadstar (space tech defence company) to validate the idea (he was a fan).
There was such a high volume of data that needed to be processed and rendered which caused it to be laggy. We spent a long time playing around with various parameters on the rendering until it worked.
We didn’t have all the data on what a malicious attack was. So we had to build a likelihood ratio and take some realistic assumptions which could be changed via the dashboard to allow the model to work.
Accomplishments that we're proud of
- Live tracking and threat modelling working using various factors
- Deep research agent that highlighted live relevant geopolitical events that may contribute to events
- comparison of historical manoeuvres as a way to create a probability distribution of a live manoeuvre being hostile
- AI agents deciding what response to take based on the threat level and context we provided
What we learned
- How Bayesian Inference works
- The maths behind comparing the similarity of trajectories in spherical polar coordinates and similarity of orbits
- How to maintain and handle a live 3D simulation with lots of data
Running the demo instructions
Click the camera icon in the bottom left
To the right of the camera icon click the demos & select malicious manoeuvre
To the right of the demo button click auto
You'll see two satellites now, the one in front is the potentially hostile one
Increase the time in the top right then you'll see the threat % rise on our satellite.
On the left you'll see five options click the bottom one to view the agent being triggered and its eventual decision output.
Built With
- nextjs
- python
- typescript
Log in or sign up for Devpost to join the conversation.