Mobile Robot Control for Agriculture Using Autoencoder Sensor Fusion

This work proposes a mobile robot control applied to a seeding task using inertial sensors. The position estimation through these sensors is solved using an autoencoder neural network trained in simulation. The control strategy is tested in the Webots simulation environment. The control strategy uses two PID controllers, one for position and other for direction. Theses controllers are optimized during trajectory using the bat algorithm to update its parameters aiming limited energy cost. The optimization objective involves position and energy cost estimated using the data fusion autoencoder as well as speed imposed to the wheels. The results present a trajectory tracking for a straight path in a rectangular environment.