Real time, dynamic cloud offloading for self-driving vehicles with secure and reliable automatic switching between local and edge computing

Autonomous robots are quickly transitioning from isolated to connected systems, offloading an increasing number of operations to third-party systems. Several historical factors contributed to the fact that robots have had the necessary intelligence built into them. By using the massive amount of data that is generated by the on-board sensors, they locate objects in real time and properly steer clear of them. Today, these robots are quickly transitioning from isolated to connected systems, offloading an increasing number of operations to third-party systems. Offloading computational tasks can actually have a significant positive impact on computational capacity, allowing to handle larger workloads and process them more efficiently. However, it's important to note that offloading typically involves some trade-offs, and one of the common trade-offs is increased latency. The extent of the latency trade-off depends on various factors, including the speed and reliability of the network connection, the distance between the local system and the offloading resource, and the complexity of the computations being offloaded. It's important to carefully consider the requirements and constraints of your specific use case and computational capacity, especially when latency and energy consumption minimization are required such as in collaborative robotics, where people and robots interact in dynamic environments. In the scenario considered in this work, a mobile ROS2-based robot that can sense, compute, and communicate wirelessly, moves from a starting position to a target point in an operating environment, avoiding dynamic obstacles. For this robotic application, it was suggested a mechanism for offloading computation, designing and putting into place a local and remote switching system to accomplish this. Multi-access edge computing (MEC) was the preferred choice for this use case, being a distributed computing architecture that brings computational capabilities closer to the network edge, specifically to the base stations or access points. This proximity reduces the distance between the connected agents (devices) and the data processing server, resulting in lower latencies. In the use case analyzed in this thesis, which involves, among others, the detection of lidar objects, low latency is crucial. The Edge Computing service delivery model enables the robotic agents to delegate their computationally demanding tasks to powerful computing infrastructure nearby. The decision to offload is based on the resources that are available at the network's edge. The aim is to use the resources, both on the local and edge sides, to meet the service level goals, which may include but are not limited to latency, bandwidth, reliability, and privacy. In contrast to on-board computers, edge platforms offer strong computation capacity, real-time data transmission, and the ability to handle enormous amounts of data at very high processing speeds. Additionally, because the distance between the edge-connected robots and the edge datacenters is much shorter, it will offer lower latencies. New architectural and technological developments have been examined, enabling seamless operation of containerized robotic applications at the edge or in the cloud. It is also provided an overview of the systems that enable secure and dependable high-bandwidth low-latency connectivity between automated vehicles and remote servers, ranging from extensions to the ROS 2 tooling to the integration of Kubernetes and ROS.