Collaborative SLAM using a swarm intelligence-inspired exploration method

Abstract

Efficient exploration in multi-robot SLAM is a challenging task. This thesis describes the design of algorithms that would enable Loomo robots to collaboratively explore an unknown environment. A pose graph-based SLAM algorithm using the on-board sensors of the Loomo was developed from scratch. A YOLOv3-tiny neural network has been trained to recognize other Loomos, and an exploration simulation has been developed to test exploration methods. The bots in the simulation are controlled using swarm intelligence inspired rules. The system is not finished, and further workis needed to combine the work done in the thesis into a collaborative SLAM system that runs on the Loomo robots.