4DoF Rat-SLAM with Memristive Spiking Neural Networks for UAVs Navigation System

ABSTRACT: Unmanned Aerial Vehicles (UAVs) are versatile platforms with potential applications in precision agriculture, disaster management, and more. A core need across these applications is a navigation system that accurately estimates location based on environmental perception. Commercial UAVs use multiple onboard sensors whose fused data improves localization accuracy. The bioinspired Rat-Simultaneous Localization and Mapping (Rat-SLAM) system, is a promising alternative to be explored to tackle the localization and mapping problem of UAVs. Its cognitive capabilities, semi-metric map construction, and loop closure make it attractive for localization in complex environments. This work presents an improved Rat-SLAM algorithm for UAVs, focusing on three innovations. First, Spiking Neural Networks (SNNs) are incorporated into Rat-SLAM’s core modules to emulate biological processing with greater efficiency. Second, Neuromorphic Computing models the neurons of the SNNs, assessing the feasibility of implementing SNNs on specialized hardware to reduce software processing, a key advantage for UAVs with limited onboard resources. Third, SNNs are developed based on the Memristive Leaky Integrate-and-Fire model, integrating memristors into artificial neurons to leverage their low power and memory properties. Our approach was evaluated through trajectory simulations using the Hector Quadrotor UAV in the Gazebo environment within the Robot Operating System, yielding valuable insights and guiding future research directions.