Collaborative Localisation and Planning for Autonomous Vehicles Using Distributed Optimisation

Abstract

This thesis aims to contribute to the autonomous driving domain in two main areas: collaborative planning algorithms for safe navigation and distributed localisation techniques for accurate position estimation using set theory. For the localisation problem, the use of both Zonotopic and Interval Sets was explored to collaboratively localise a fleet of robots within a common frame. Specifically, set intersection was used to fuse data from multiple onboard sensors. These methodologies were compared in a simulated environment to evaluate their effectiveness. In the context of distributed planning, the thesis proposes several optimisation-based approaches adapted from single-agent scenarios to networks of vehicles, with a focus on collision avoidance and physical constraints of the environment. The foundation of these algorithms is the Model Predictive Control (MPC) controller which, due to its duality with planning problems, was modified to generate realistic trajectories. This led to the development of a novel non-linear planner and its pseudo-linear Linear Parameter Varying counterpart for lane-free environments. The next step involved integrating lanes into the optimisation problem. This posed challenges due to the introduction of integer variables, leading to computationally intensive problems in both local optimisations and their decomposition. Two algorithms were proposed to address this: a decentralised integer planner combining both optimisation problems, and a simpler approximation to the lane selection problem assuming relaxed vehicle dynamics and collision checks. The former was unfeasible due to scalability issues, while the latter showed promising results when combined with a lane-free spatial planner. Finally, lane-free planning algorithms were expanded by embedding system uncertainties in the optimisation problem using set theory. In particular, both system noise and unmodeled dynamics were propagated through the optimisation horizon using Reachability and Zonotopic Sets, ultimately guaranteeing safe trajectories under bounded uncertainties.

Date of Award	19 Dec 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Prasad Potluri (Co Supervisor) & Ognjen Marjanovic (Main Supervisor)