This paper presents a navigation algorithm with an integrated obstacle avoidance module for insect-computer hybrid robots using a monocular camera. The algorithm leverages a deep learning-based monocular depth estimation algorithm to generate depth maps, guiding the robot away from obstacles based on distance distribution. A new dataset, the SmallRobot Dataset, was collected from a small robot's perspective for model training. The proposed method significantly improved the navigation success rate from 6.7% to 73.3%.

Rui Li, Qifeng Lin, Phuoc Thanh Tran-Ngoc, Duc Long Le, Hirotaka Sato