Predicting Energy Consumption of Ground Robots On Uneven Terrains

Optimizing energy consumption for robot navigation in fields requires energy-cost maps. However, obtaining such a map is still challenging, especially large, uneven terrains. Physics-law-based models work uniform, flat surfaces but do not generalize well to these Furthermore, slopes make the at every location directional and add complexity of data collection prediction. In this paper, we address challenges data-driven manner. We consider function which takes terrain geometry motion direction as input outputs expected consumption. The represented ResNet-based neural network whose parameters are learned from field-collected data. prediction accuracy our method within 12% ground truth test environments that unseen during training. compare baseline literature: using basic physics-based model. demonstrate significantly outperforms it by more than 10% measured error. More importantly, generalizes better when applied new with various slope angles directions.