Neuromorphic LIDAR-based Robot Obstacle Avoidance using Spiking Neural Networks

Neuromorphic computing is rapidly expanding in the field of robotics due to the remarkable efficiency of Spiking Neural Networks (SNNs) which offers high precision yet low memory and compute complexity inference when implemented in neuromorphic hardware. This study investigates the use of SNNs for autonomous robot navigation and obstacle avoidance based on LIDAR data and kinematic state data collected from a custombuilt platform. A key contribution of this paper is the analysis of the neuron membrane leakage constant in Leaky Integrate-andFire (LIF) neurons model which shows that proper tuning significantly improves model performance. With an optimal leakage constant, the SNN achieves statistically comparable performance to a CNN as well as being less computationally expensive.