Recent trends in agriculture show decreasing availability of agricultural workers, coupled with reduction in productivity and yield due to climate change. Autonomous agricultural robots are a suitable technology to counter these issues, but in order to gain widespread use they must be able to effectively localise themselves in complex rural environments, whilst being able to run for long periods of time. This paper aims to aid in the selection of Simultaneous Localisation And Mapping (SLAM) variants and parameters when used in the agricultural field in order to maximise range/operational time with minimal sacrifice of effectiveness as evaluated by Percentage Tracked Path (PTP) and Average Trajectory Error (ATE). We investigate some of the most promising applications of ORB-SLAM algorithms, running on a low-resource system, providing insights into the energy costs of increased real-time performance in terms of Frames Per Second (FPS) achieved through varying SLAM parameters. Our contributions are: a full Energy Per Frame (EPF) analysis over all ORB-SLAM parameters; SLAM analysis of a benchmark LFSD dataset, entirely performed on a common low-resource robotics computer, Jetson Nano; evaluation of the determined parameters to achieve highest PTP at lowest EPF; and live trials on the Turtlebot robotic platform using the determined parameters. This work demonstrates that the most power efficient and accurate performance of the variants tested can be attained using ORB-SLAM3 with the number of features reduced to 600. This improved on default parameter operation by 20% on EPF and 8.21% on PTP. This outcome was corroborated in live testing achieving a similar 14.4% EPF drop to 321.83±46.14mJ. Therefore, utilising our power performance evaluation, power can be optimised for real-time usage based on the parameters selection, enabling a more accurate and power-efficient SLAM for longer operation time in the field.