Working on Spacecraft perception and autonomy | PhD MIT AeroAstro | MIFP Fellow 18 | @NSFGRFP | 🇵🇷🏳️🌈 | 📡 KM4BTG
- Tweets5.9K
- Following881
- Followers1.2K
- Likes39.5K
ALT Graph Neural Network Architecture. Each satellite within the environment is represented as a node, and communications between satellites are represented as edges. Each node represents a particular spacecraft, which means that that specific node can encode specific qualities including size of the spacecraft, its capabilities, and its precise location. Each edge represents communication and sensing abilities, so they can encode qualities related to the relative information between two satellites including their relative distances, the areas they have both surveyed or other characteristics associated with path planning. Our graph formulation is dynamic, meaning that the over the course of the simulation, the satellites can adaptively connect and disconnect with one another depending on the satellites ability to sense those in their proximity. We rely on a unified message passing model (UniMP), a variant of a graph transformer, to allow the agents to selectively prioritize messages coming
ALT “The worst thing I've ever experienced in my life was walking among the corpses to find my daughter. I kept repeating in my mind that she was wearing a white wool sweater so I could recognize her among the hundreds of martyrs and bury her next to her mother and brother—without being distracted by the horrific scenes around me. But the massacre left no trace of white; everything was stained with blood and the colors of death and fire. I didn't find many intact bodies. Most of the martyrs' remains were just body parts. I searched endlessly but couldn't find my daughter. However, I did recognize my sister's body—my daughter had been with her. Imagine having to help bury the martyrs without knowing which of the scattered remains belonged to my little girl.”
