This paper explores the application of biomimetic design to autonomous unmanned aerial vehicles (UAV) navigation, leveraging the characteristics of biological systems capable of flight and obstacle avoidance despite limited sensory and computational capabilities. Specifically, focusing on the observation that honeybees primarily utilize optic flow for navigation in complex environments, we trained a reinforcement learning agent to navigate a tunnel containing obstacles using only optic flow as a sensory input. We analyzed the trained agent's attention patterns to determine which areas of optic flow are primarily responsible for making movement decisions. We found that the trained agent focused most attention on discontinuous areas of optic flow and areas with large optic flow magnitudes. The trained agent navigated complex tunnels by maintaining its position in the center of the environment while avoiding obstacles that generated large optic flow, a behavior similar to insect flight behavior. This pattern persisted in independently trained agents, suggesting that it could be a useful strategy for developing simple explicit control laws for real-world UAVs.
