3D Gaussian Splatting Multi-agent simultaneous localization and mapping

This project focuses on three tasks: 1. Develop a compact descriptor for 3D Gaussian splats to enable robust registration between partial sub-maps. Inspired by prior learned descriptors, the method will encode local Gaussian parameters for fast matching under noise and partial overlap. 2. Extend existing multi-agent Gaussian SLAM system into a decentralized framework where each robot shares and merges local maps without a server. Consensus-based optimization will allow scalable mapping under communication limits. 3. Deploy the system on real multi-robot platforms with edge devices. Evaluate mapping accuracy and communication overhead in real environments.

Faculty Supervisor:

Giovanni Beltrame;Pierre-Yves Lajoie

Student:

Partner:

University of Oxford

Discipline:

Computer science

Sector:

Technology

University:

Polytechnique Montréal

Program: