TechnoMania 2.0: 24 Hr Hackathon

TechnoMania 2.0 brings you a cutting-edge 24-hour offline coding marathon that pushes the boundaries of standard hackathons. Unlike traditional open-ended challenges, this edition focuses on "Integrated Autonomy & Deep Learning." Participants will move beyond simple computer vision models to build a complete Specialized Autonomous Pipeline. Leveraging Duality AI’s Falcon Digital Twin Platform, teams will utilize high-fidelity synthetic data to solve complex problems in off-road environments. The Theme: Offroad Semantic Scene Segmentation & Autonomy The core objective is to develop robust AI models capable of navigating unstructured, difficult-to-access environments (like deserts). Teams will train their models using synthetic data generated by Falcon Cloud and test them against novel environment shifts. The Workflow: Every team begins with a mandatory Foundation Phase and then splits into a specialized Elective Track to finalize their prototype. The Challenge Tracks & Problem Statements: Phase 1: The Foundation (Mandatory for All) Problem Statement: Offroad Semantic Scene Segmentation All teams must first build a robust semantic segmentation model capable of identifying navigable terrain versus obstacles in a desert environment. Goal: Achieve high accuracy in segmenting terrain types using Duality’s annotated synthetic datasets. Outcome: A functional vision model that serves as the "eyes" of your autonomous agent. Phase 2: The Elective Tracks (Choose One) Once the foundation is set, teams must innovate by integrating one of the following advanced capabilities: Track 1: Natural Language Processing (NLP) Focus: Scene Reasoning & Explainability Problem Statement: Bridge the gap between vision and language. Your system shouldn't just "see" pixels; it must understand them. Task: Develop a module that generates human-readable text summaries of the scene (e.g., "Obstacle detected 5 meters ahead," "Terrain shifting to soft sand"). Track 2: Advanced Computer Vision Focus: Path Planning & Navigation Problem Statement: Vision is useless without action. The agent needs to know where to go. Task: Implement algorithms for autonomous path planning and dynamic obstacle avoidance based on the segmentation output. The agent must calculate the safest route through the terrain. Track 3: Generative AI Focus: Domain Robustness & Edge Cases Problem Statement: Real-world data is messy. Models often fail in "edge cases" (rare scenarios) that are hard to capture. Task: Use GenAI techniques to synthesize novel "edge case" scenarios (e.g., sandstorms, rare rock formations) to retrain and stress-test the model for superior robustness. Track 4: Speech-to-Text / Voice UI Focus: Human-Machine Interaction Problem Statement: In field operations, operators need hands-free control over autonomous systems. Task: Build a real-time Voice User Interface (VUI) that allows operators to issue verbal commands (e.g., "Stop," "Return to base," "Scan sector") which the system interprets and executes. Detailed Guidelines & Resources: For a complete breakdown of the dataset structure, evaluation metrics, and technical resources for the Falcon Platform, please refer to the official problem statement document.