Hardware Implementation of On-Chip Learning Neural Network on FPGA

Animals achieve precise locomotion through learning and practice, and many scholars are investigating how animals acquire new skills. Deep Reinforcement Learning (DRL) is often employed in skill learning and locomotion in robotics. However, the relationship between the learning process and neural networks or DRL remains unclear. In this paper, we propose a framework of backpropagation (BP) network based on matrix multiplication, which is implemented and visualized on a Field Programmable Gate Array (FPGA). This allows us to track changes in each neuron by analyzing nonlinear signals. By considering each register in the FPGA as a neuron, we can plot the weights of joints, which provides insights into the learning rules.