Week #1 #
Project description #
Project name: Autonomous Semantic 3D Reconstruction and High-Fidelity Mapping #
Code repository: https://github.com/Mousatat/Autonomous-Drone-Semantic-3D-Reconstruction-and-High-Fidelity-Mapping
An autonomous UAV-based system designed for semantic 3D spatial reconstruction and high-fidelity mapping. The project integrates real-time semantic segmentation via SpatialLM, autonomous UAV navigation using agentic AI, and precise exploration strategies to efficiently generate detailed digital twins.
Problem Statement #
Existing autonomous 3D mapping solutions face limitations in:
Real-time semantic understanding of environments.
Autonomous optimal exploration decision-making.
Efficient and effective data processing for high-quality spatial reconstructions.
Our system addresses these challenges through advanced semantic models, autonomous UAV navigation powered by agentic AI, and two-stage reconstruction techniques for superior spatial accuracy and usability.
Team Members #
| Team Member | Telegram Alias | Email Address | Track | Responsibilities |
|---|---|---|---|---|
| Mahmoud Mousatat | @Mousatat | m.mousatat@innopolis.university | Computer Vision | Semantic understanding from visual data |
| Nikita Sergeev | @nary_2 | n.sergeev@innopolis.university | AI & Backend | Developing agentic AI logic |
| Alexander Rozanov | @alrozanov | al.rozanov@innopolis.university | DevOps | Integrating AI solutions with cloud environments |
| Ilvina Akhmetzianova | @IviUnicorn | i.akhmetzianova@innopolis.university | Simulation & Frontend | Building simulation environments and interfaces |
Brainstorming #
Ideas during brainstorming #
- Autonomous Semantic 3D Mapping (Chosen idea) — UAV system using semantic segmentation, real-time navigation, and agentic AI for 3D mapping and digital twin creation.
Basic requirements #
Target users and their primary needs #
Construction and Architecture Firms: Require accurate digital twins for BIM and project monitoring.
Disaster Response Teams: Need rapid and accurate maps for damage assessment and planning.
Urban Planning and Infrastructure Management: Require updated spatial data for urban management.
Initial scope (MVP) #
Included:
Real-time semantic spatial reconstruction using SLAM and SpatialLM.
Autonomous navigation planning and exploration via agentic AI.
Simple game-server communication setup for simulation integration.
Excluded (future iterations):
Detailed photogrammetric reconstruction using traditional methods.
Neural rendering for interactive visualizations.
Fully scaled cloud solutions and multi-UAV coordination.
Tech-stack #
Frontend / Visualization #
- PlayCanvas: Chosen as an initial game engine for simulation and frontend interactions.
Backend / AI Models #
SpatialLM: Semantic segmentation.
Visual-Inertial SLAM: SLAM3R or similar solutions.
Claude 4 or Gemini 2.5 Pro: Agentic AI path planning and decision-making.
Cloud Infrastructure #
We are utilizing a Virtual Dedicated Server (VDS) to host our cloud infrastructure, specifically deploying an Nginx server to facilitate efficient and secure communication between LLM and agentic AI.
Weekly commitments #
Individual contribution of each participant: #
Mahmoud Mousatat:
- Investigated suitable vision models for semantic understanding.
Nikita Sergeev:
- Explored agentic frameworks for semantic vision integration and spatial data processing.
Alexander Rozanov:
Prepared initial weekly report.
Explored integration solutions for agentic AI and cloud infrastructure.
Ilvina Akhmetzianova:
- Evaluated suitable game engines, initially selecting PlayCanvas for simulation.
Confirmation of the code’s operability #
We confirm that the code in the main branch:
- In working condition.