2022 Volume 84 Issue 4 Pages 256-264
Autonomous navigation by vision sensors was realized in a vineyard. We used a neural network to detect vanishing points in images from two vision sensors on the front and back of an electric vehicle robot. The vehicle was steered based on the heading and lateral errors calculated geometrically. Turning was performed by detecting markers on vine hedges with a vision sensor and an IMU. When the vehicle was driven manually to evaluate the accuracy of estimation, the heading RMSE was 1.3°, and the lateral RMSE was 0.12 m. When the vehicle traveled autonomously between vine rows, the lateral RMSE was 0.06 m. When the vehicle completed turning autonomously, the heading error was 1.8° and the lateral error was 0.04 m.
