Vyuha AI - The Autonomous Orbital Overseer [Demo by Roshan Sharma] скачать в хорошем качестве

Vyuha AI - The Autonomous Orbital Overseer [Demo by Roshan Sharma]

📋 The Problem Low Earth Orbit is degrading into a congested junkyard. With satellite counts projected to jump from 15,000 to 100,000 by 2030 and over one million pieces of lethal debris in orbit, the volume of tracking data has vastly exceeded human analytical capacity. Traditional manual oversight is too slow to prevent catastrophic collisions in this high-velocity environment. ⚡ The Solution Vyuha is a Space Domain Awareness agent that decouples satellite operations from terrestrial control. By ingesting real-time open telemetry (CelesTrak): Predicts — autonomously identifies collision risks (conjunctions) in real time. Acts — calculates and executes avoidance maneuvers without human-in-the-loop latency. Protects — White Circle security gateway validates every orbital command, preventing indirect prompt injection from hijacking thrusters or causing intentional de-orbit. Addresses US strategic priority · Blaxel serverless · White Circle solves the autonomous insider threat. 🎯 Capabilities Live orbital ingestion — CelesTrak TLE Hybrid demo — real telemetry + forced critical scenarios Cyberattack demo — simulate indirect prompt injection → White Circle blocks → resilience Commander (Blaxel) → FIRE_THRUSTERS / HOLD_POSITION State & memory — /state, /restore, /history Learning — /insights -------------------------------------------------------- Excited to share that my project, Vyuha AI, was selected as a Top 8 Finalist at the Iterate x Columbia AI Club Hackathon! 🏆 By 2030, Low Earth Orbit (LEO) will be crowded with over 100,000 satellites and millions of debris pieces. The risk? A catastrophic chain reaction of collisions where human analysis simply cannot keep up with the velocity of threats. To solve this, I built Vyuha AI (Sanskrit for "strategic battle formation"). It is an autonomous orbital overseer designed to decouple satellite operations from terrestrial control. 🛑 The Problem: High-velocity space environments require sub-second decisions. ✅ The Solution: An agent that follows a strictly autonomous loop: Predict. Act. Protect. 🛠️ Under the Hood: How I Built It This wasn't just a concept; it was a complex orchestration of multi-modal AI agents: The Vision (Overshoot AI): I integrated Overshoot AI to power the satellite's "eyes," enabling real-time visual detection of debris objects (simulated at 0.5km proximity) to trigger collision avoidance protocols. The Infrastructure (Blaxel): The core agent lives in a sandbox environment powered by Blaxel. I used Blaxel to handle state management, recording position history to allow for trajectory restoration and maintaining the agent's "liveness" without latency bottlenecks. The Guardrails (White Circle AI): Autonomous agents need boundaries. I implemented White Circle AI to act as the "Protection" layer against cyber threats. Real-world Scenario: During the demo, I attempted an indirect prompt injection to force the satellite to "fire thrusters towards the debris." White Circle successfully validated the command, recognized the policy violation (self-harm/collision), and rejected the prompt instantly. 🔭 The Future of Space Ops We are entering an era where satellites must be intelligent enough to protect themselves. I see massive potential for Vyuha AI to integrate with the next generation of constellations from NASA, ISRO, SpaceX, and Blue Origin. Building this was an intense challenge in balancing agentic autonomy with strict safety protocols. A huge thank you to the organizers and judges for recognizing the potential of Vyuha. On to the next frontier! 🌌 #AI #SpaceTech #Hackathon #AutonomousAgents #VyuhaAI #Blaxel #WhiteCircleAI #OvershootAI #SpaceX #ISRO #MachineLearning #ColumbiaUniversity