A Design of Reinforcement Learning Accelerator Based on Deep Q-learning Network

Abstract

Reinforcement Learning (RL) is widely adopted in robotic controlling and decision making due to its ability of adapting to the changing. However, the Deep Q-learning Network (DQN) RL systems heavily rely on CPU and GPU to accelerate the computation, which cannot meet the requirement. In this paper, we proposed a design of RL accelerator based on DQN in CPU-FPGA architecture. In this paper, we propose a design of RL accelerator based on DQN in CPU-FPGA architecture. Compared with other CPU architecture, the accelerator we proposed achieves 1.84x computing speed improvement and reduces hardware resources significantly.