A Reduced-Order-Model-Based Motion Selection Strategy in a Leg-Wheel Transformable Robot

This article reports on a model-based methodology for a leg-wheel transformable robot to autonomously determine the use of the wheeled or legged mode based on environmental RGB-Depth information. A concentric and reduced-order dual-leg-wheel model was developed to explore the dynamic interactions between the leg-wheel and terrain. The simulation results revealed that terrain height variation acts as the key factor in determining the model's terrain negotiability; therefore, it was utilized as the decision index for leg-wheel motion selection. To facilitate real-time motion selection on the empirical robot, a two-step selection strategy utilizing RGB-Depth information was proposed. The first step utilized RGB images to classify seven types of terrain. If the classified terrain was potentially rough, the second step utilized depth images to compute the terrain height variation, which was then compared to the index derived by the dual-leg-wheel model to select the final operation mode. The strategy was implemented on a leg-wheel transformable robot and experimentally validated on various types of terrain. The results confirm that the robot's behavior closely resembled the model's prediction, and using the proposed strategy to select leg-wheel motion in the robot yielded the most energy efficient locomotion.