Dissertation.pdf

Code

diss-video.mp4

In this project, I:

1. Design and implement a novel distributed monocular visual SLAM system, capable of localization, relative pose estimation, and collaborative mapping, all while being tolerant to degraded network conditions and not reliant on any single leader agent (section 3.2).
2. Evaluate the performance of my system on standard datasets, demonstrating its superior performance over comparable state-of-the-art systems (section 4.3).
3. Create a simulation environment for testing and evaluating my system locally (section 3.6).
4. Develop a custom collision avoidance framework (section 3.3) and deploy it alongside my SLAM system on physical robots, demonstrating the practical use cases of my system and benchmarking real-world performance (section 4.4).
5. Contribute as a co-author to the paper The Cambridge RoboMaster: An Agile Multi-Robot Research Platform. My distributed SLAM system is included in the paper and used to evaluate the robotics platform (section 3.7.1).
6. Develop Multi-Agent EVO – the first open-source evaluation library for multi-agent SLAM systems (section 3.5).
7. Develop the Raspberry Pi Video Publisher – a performant platform for SLAM data collection and augmented reality visualizations – and set up a continuous integration and deployment pipeline to automatically deploy the latest builds to the devices (section 3.7.3).