Image-Based Particle Filtering For Robot Navigation In A Maize Field

Autonomous navigation of a robot in an agricultural field is a challenge as the robot is in an environment with many sources of noise. This includes noise due to uneven terrain, varying shapes, sizes and colors of the plants, imprecise sensor measurements and effects due to wheel-slippage. The drawback of current navigation systems in use in agriculture is the lack of robustness against such noise. In this study we present a robust vision-based navigation method based on probabilistic methods. The focus is on navigation through a corn field. Here the robot has to navigate along the rows of the crops, detect the end of the rows, navigate in the headland and return in another row. A Particle Filter based navigation method is used based on a novel measurement model. This model results in an image from the particle state vector that allows the user to compare the observed image with the actual field conditions. In this way the noise is incorporated into the posterior distribution of the particle filter. The study shows that the new method accurately estimates the robot-environment state by means of a field experiment in which the robot navigates through the field using the particle filter.