Algorithm and RTL Architecture Design of Q-Learning Algorithm for Path Planning

In this paper, we propose two simulations designed for the implementation of Q-learning on path planning. The first simulation of a modeling design using MatLab to find the best route by agent and calculate the updated reward from the environment. The second simulation of a modeling design RTL using Xilinx Vivado and written in Verilog. Q-learning is an off-policy reinforcement learning technique and is used for decisive action by value in path planning’s navigation. The scheme of the environment is a grid world map with a volume-sized 5x5 number state position and has four possible actions. The result illustrates the shortest step route decided by an agent with convergence accumulated reward from the q-table and the goal position achievement after the learning process. Implementation design architecture RTL is implemented to prove the architecture can be applied. From both results, the final condition to achieve the goal’s position by the agent in the model simulation was 64