A Novel Approach for Autonomous Mobile Robot Learning and Control Using a Customized Spiking Neural Network

Abstract

Abstract The application of reinforcement learning in autonomous mobile robots is a challenging task and paid more attention to in previous years. Traditional reinforcement learning (TRL) methods have several limitations, such as extensive trial-and-error searches, complex control frameworks, slow convergence, and prolonged computation time. This article proposes a novel approach for autonomous mobile robot learning and control in unknown environments using a customized Spiking Neural Network (SNN). The proposed model combines spike-timing-dependent plasticity (STDP) with dopamine modulation as a learning algorithm. This study uses the efficient and biologically plausible Izhikevich neuron model, which can lead to the development of more biologically-inspired and computationally efficient control systems that can adapt to changing environments in unknown environments. Accordingly, this paper aims to develop an algorithm for target tracking amidst obstacles. We conducted extensive simulation experiments to evaluate the proposal in the Webots robotic environment simulator. The findings demonstrate that our proposal achieved a remarkable 100% success rate in reaching the target for the SNN trained with one obstacle without any collisions during the 972 simulated seconds. However, the SNN trained with three obstacles achieved a slightly lower success rate of 96%, with collisions occurring approximately 4% of the time during the 214 simulated seconds. These findings suggest that training the SNN with a single obstacle is more effective than training with three obstacles.