Collaborative LiDAR SLAM

Autonomous robot teams operating in large-scale environments require the ability to collaboratively estimate their locations while constructing a consistent representation of the surrounding environment. This project developed a Random Finite Set (RFS) based Collaborative LiDAR SLAM framework that enables multiple autonomous ground robots to cooperatively perform simultaneous localisation and mapping.

The framework extends probabilistic SLAM from a single robot to a collaborative multi-robot setting by allowing robots to contribute observations towards a common environmental map while maintaining consistent estimates of their individual trajectories. The work demonstrates how Bayesian estimation techniques can be used to scale localisation and mapping beyond individual robotic platforms.

Research Motivation

Single-robot SLAM algorithms are limited by sensing range, computational resources, and the amount of environment that can be explored within a given time. Collaborative SLAM enables multiple autonomous robots to simultaneously explore an environment, share observations, and jointly construct a consistent global map.

This project investigated probabilistic methods for collaborative mapping using Random Finite Set theory, providing a principled Bayesian framework for handling uncertainty in both robot localisation and environmental mapping.

Research Contributions

• Developed a collaborative LiDAR SLAM framework based on Random Finite Set theory.
• Extended probabilistic SLAM from single-robot to multi-robot operation.
• Integrated particle-filter-based robot localisation with GM-PHD map estimation.
• Demonstrated collaborative localisation and mapping using multiple autonomous ground robots.
• Validated the framework using real robotic platforms operating in outdoor environments.

Experimental Platforms

The proposed framework was demonstrated using multiple Clearpath Husky unmanned ground vehicles equipped with:

• Velodyne VLP-16 LiDAR
• Wheel encoders
• Gyroscope
• Onboard computer
• ROS software framework

LiDAR point clouds were pre-processed to remove the ground plane before extracting stable environmental features for probabilistic mapping.

Technologies

• Random Finite Set Theory
• Collaborative SLAM
• LiDAR-based Mapping
• Bayesian Estimation
• Particle Filtering
• GM-PHD Filter
• ROS
• Velodyne VLP-16

Demonstration Video

Related Research Pillars

• Robot Perception & Localisation
• Probabilistic State Estimation
• Autonomous Robotic Systems

Publications

Add related journal and conference publications.

Future Directions

This research established the foundations for collaborative probabilistic perception using multiple autonomous robots. Current research extends these concepts towards decentralised perception, distributed information fusion, and collaborative autonomous decision-making under intermittent communication.