Resource Utilization Prediction Model for SLAM Offload to Edge

In recent years, the use of autonomous mobile robots has increased. SLAM (Simultaneous Localization and Mapping) technology is where an autonomous mobile robot estimates its own position and simultaneously maps the surrounding environment. It is widely used in robots that require autonomous movement. However, it is heavily burdened by computational tasks related to various types of image processing and calculations of physical environment recognition, which makes the system slow down due to limited hardware resources such as embedded systems. Various offload methods have been proposed, however they have not been applied to Robot Operating System (ROS) + SLAM applications and there are few examples of robot verification. To avoid the heavy load on the devices, we propose a method that takes the heavily load to the edge by using the prediction model based on the system resource information for a certain period related to SLAM processing. Based on preliminary experiences, we firstly develop a prediction model using system resource consumption data obtained from an actual SLAM execution and evaluate the prediction accuracy. The result shows that the accuracy varies as memory consumption increases, however the accuracy was high and useful in determining offload timing.