Mode Prediction and Adaptation for a Six-Wheeled Mobile Robot Capable of Curb-Crossing in Urban Environments

Abstract

Wheeled mobile robots have a significant role in the mobility industry. When outdoor mobile robots perform given tasks in urban environments, they often face the challenge of climbing and descending curbs. Many commercial mobile robots rely on ramps to navigate curbs. However, for outdoor mobile robots to achieve human-like freedom of movement in urban spaces, they must be capable of climbing and descending curbs. A six-wheeled mobile robot can be a solution to this problem since it can effortlessly climb or descend curbs. When driving on flat terrain, however, driving with six wheels is quite inefficient, wasting energy, and driving with four wheels would be better than using six wheels. Therefore, when operating a six-wheeled mobile robot, we need to determine the appropriate mode to use, specifically the six-wheel mode or the four-wheel mode, depending on the type of the road. In this paper, we present a method that determines the appropriate mode using the video captured from the frontal camera. We employ 3D CNN and the Bayesian fusion to determine the optimal mode for the six-wheeled robot and optimal timing for mode change. We collect a set of video sequences and test our method on these collected video sequences.

Keywords

• Bayesian fusion
• bogie suspension
• curb
• deep learning
• mobile robot
• navigation