Binarized P-Network: Deep Reinforcement Learning of Robot Control from Raw Images on FPGA

This letter explores a deep reinforcement learning (DRL) approach for designing image-based control edge robots to be implemented on Field Programmable Gate Arrays (FPGAs). Although FPGAs are more power-efficient than CPUs and GPUs, typical DRL method cannot applied since they composed of many Logic Blocks (LBs) high-speed logical operations but low-speed real-number operations. To cope with this problem, we propose novel algorithm called Binarized P-Network (BPN), which learns image-input policies using Convolutional Neural Networks (BCNNs). alleviate the instability caused by BCNN low function approximation accuracy, our BPN adopts robust value update scheme Conservative Value Iteration, is tolerant errors. We confirmed BPN's effectiveness through applications visual tracking task in simulation real-robot experiments FPGA.