Evaluating the Power Efficiency of Visual SLAM on Embedded GPU Systems

摘要

SLAM (Simultaneous Localization and Mapping) on mobile robot requires high performance but low power consumption because it is constrained to the battery. This paper testes the performance and evaluates the power efficiency of enabling GPU acceleration for ORB-SLAM2 on three embedded GPU systems: Jetson Xavier, Jetson TX2 and Jetson Nano.