An Infrastructure for Robotic Applications as Cloud Computing Services

Robotic applications are becoming ubiquitous. They are widely used in several areas (e.g., healthcare, disaster management, and manufacturing). However, their provisioning still faces several challenges such as cost efficiency. Cloud computing is an emerging paradigm that may aid in tackling these challenges. It has three main facets: Infrastructure as a Service (IaaS), Platform as a Service (PaaS) and Software as a Service (SaaS). Virtualization is a technique that allows the abstraction of actual physical computing resources into logical units; it enables efficient usage of resources by multiple users. Its role is a key to resource efficiency. Virtualization can be performed at both node and network level. This thesis focuses on the IaaS aspects of robotic applications as cloud computing services. It starts by defining a set of requirements on the infrastructure for cost efficient robotic applications provisioning. It then reviews the state of the art. After pinpointing the shortcoming of the state of the art, it proposes an architecture that enables cost efficiency through virtualization and dynamic task delegation to robots, including robots that might belong to other clouds. Overlays and RESTful Web services are used as cornerstones. The virtualization in the IaaS is achieved by providing a coalition formation algorithm, which is the cooperation between several robots to perform a task that either cannot be solved individually or can be solved more efficiently as a group. Forming the effective coalitions is another big challenge. We adapted heuristic-based Multi Objective- Particle Swarm Optimization (MO-PSO) algorithm to solve this specific problem. As a proof of concept, a prototype is built using LEGO Mindstorms NXT as the robotic platform, and JXTA as the overlay middleware and the prototype architecture is presented along with the implemented scenario (i.e., wildfire suppression). Performance measurements have also been made to evaluate viability. To evaluate the effectiveness of our algorithm, WEBOTS simulation software is used