Mr Alastair Finlinson
Academic and research departments
Nature Inspired Computing and Engineering Research Group, Surrey Centre for Cyber Security, Department of Computer Science.About
My research project
Vision-based positioning and navigation via sensor fusionIn conjunction with Saab Group, to address the navigation problem, the research will support Saab and explore computer vision to navigate an environment aboard an autonomous surface vessel. The vessel is equipped with an array of sensors that provide contextual information, such as from a camera and inertial measurement unit (IMU). Though vision as a localisation method has been used in more constrained scenarios where the environment is fully mapped and relatively small, such as a city block, our target environment is an archipelago (an extensive collection of islands).
My work also explores the control aspect of the surface vessel, with work considering the use case for single or multiple surface vessels operating as a fleet to carry out autonomous tasks while negotiating the environment. The vessels' cooperation and communication efficacy is key to completing group tasks, such as patrol missions or supply delivery. The work will focus on reinforcement learning agents communicating effectively to achieve optimal behaviours when negotiating the archipelago. In the single-agent scenario, the agent can operate with no regard for any other vessel and execute the task as required. In the multi-agent scenario, the agents must account for other vessels, agents or not, to achieve the objective and complete the task. The cooperation may be as simple as not crashing into one another or changing course to avoid an obstacle collectively. This allows agents to act flexibly with the larger goal in mind. The task definition of navigation and control is broad, but we focus on how and when the agents communicate and what they communicate.
Supervisors
In conjunction with Saab Group, to address the navigation problem, the research will support Saab and explore computer vision to navigate an environment aboard an autonomous surface vessel. The vessel is equipped with an array of sensors that provide contextual information, such as from a camera and inertial measurement unit (IMU). Though vision as a localisation method has been used in more constrained scenarios where the environment is fully mapped and relatively small, such as a city block, our target environment is an archipelago (an extensive collection of islands). My work also explores the control aspect of the surface vessel, with work considering the use case for single or multiple surface vessels operating as a fleet to carry out autonomous tasks while negotiating the environment. The vessels' cooperation and communication efficacy is key to completing group tasks, such as patrol missions or supply delivery. The work will focus on reinforcement learning agents communicating effectively to achieve optimal behaviours when negotiating the archipelago. In the single-agent scenario, the agent can operate with no regard for any other vessel and execute the task as required. In the multi-agent scenario, the agents must account for other vessels, agents or not, to achieve the objective and complete the task. The cooperation may be as simple as not crashing into one another or changing course to avoid an obstacle collectively. This allows agents to act flexibly with the larger goal in mind. The task definition of navigation and control is broad, but we focus on how and when the agents communicate and what they communicate.