Leveraging Edge Computing Resources on Computing-Intensive Robotic Tasks

In the previous years the industry has pushed for infrastructure management on the cloud, this meaning centralized computational resources which are then leased on demand. Within this paradigm there’s another option which has become a trend: Edge computing. In edge computing the “cloud” resources are carried geographically closer to the end user, which presents the same benefits as cloud computing with the addition of lower latency. Robotics applications have historically handled the computational burden locally (onboard computers) which comes with its own weight and power restrictions. Edge computing comes forward as a tool to be leveraged given its ability to perform graphically and memory dependent tasks within the critical parameter of latency allowing almost real-time results. This work demonstrates a practical implementation of adaptive edge computing in robotics by combining ROS 2 lifecycle management, efficient computer vision for stop sign detection, peer-to-peer Kubernetes networking via Liqo, and WebSocket-based offloading. The combined design allows the robot to dynamically switch between local and remote (edge computing) processing based on real-time network and peer availability, conserving local compute when possible and maintaining robust perception when operating in degraded network conditions. The integration of Lifecycled Ros Nodes for dynamic resource management and Liqo for peer discovery and networking are central contributions, showing how modern containerization and networking technologies can be applied to create flexible and resource-efficient robotic systems.