Abstract—A surveillance drone supervises designated area or sites against dangerous situation. In recent years, it is required to perform autonomous flight to achieve the supervision with path optimization based on minimization of the time lag. In this paper, we propose the reinforcement learning algorithm to optimize path for autonomous flight of surveillance drones. We present a simulation result of a single surveillance drone, which has reinforcement learning algorithm in an unknown grid area. A single surveillance drone finds the optimized path autonomously with minimization of the time lag. This paper provides the following two main contributions for autonomous flight of the surveillance drone. First, the surveillance drone finds the optimized path autonomously using proposed the reinforcement learning algorithm. Second, the traditional reinforcement learning was improved with parameter optimization including learning rate coefficient, convergence criteria, and adaptive error convergence detection for ε-greedy policy process.