Memristive Neural Network Based Reinforcement Learning with Reward Shaping for Path Finding

Robot is becoming an important role in people's daily life and work. Especially, in some extreme circumstances, robots can be used for searching and rescuing to improve the rescue efficiency and decrease additional casualties. Reinforcement learning, especially q-learning, is always employed in robot path finding for the unknown environment. However, the basic reinforcement learning always faces the problem of inefficiency and blindness. In this paper, a novel path finding method called MRNS q-learning is proposed. By leveraging the advantages of the reward shaping q-learning and memristive RBF neural network, this method may provide higher processing efficiency and convergence speed. In addition, a hardware architecture for the MRNS q-learning is also given. Finally, simulation results demonstrate the effectiveness and ascendancy of the proposed scheme.