Kovarnis / SurMap

Problem: Disasters come suddenly and it can be quite challenging to the responders and rescuers. SurMap is a suite of 3 major parts: 1. Autonomous surveyor drone enabler. Go to the places of interest and sample the area, project the data to a real-time map. We take stereo imagery and transform it to data that can be analysed in real time. 2. Simulations engine. It's hard to test physical world and it was a challenge to us. So we've built one too. 3. An agentic harness that can create simulations and can serve as a central piece of the disaster response. Single solution for training and the actual matter. Architecture: 1) Onboard: stereo camera, Raspberry CM5, Google CoralAI TFU. RTAB-map for navigation and depth of view. (~17 frames/s) 2) Station: local GPU or Cloud provider to run SAM 3.1 for extracting 3D objects. (~15 frames/s on RTX 5060 Ti) A significant amount of Code was written with Muse Code. In the collaboration with Cloud Code, Cursor, Warp and Copilot. Major 3rd party components that we've used: 1) RTAB-map for SLAM navigation and detection of depth. 2) Bevy game engine for simulation engine. 3) KIITI dataset bank (licenced for academic use)