Terrain Classification Based on Wheel-terrain Interaction Measurements using Aside-Wheel Camera

Feng Xue, Longteng Hu, Chen Yao, Jin Wei, Yunzhou Li, Peichen Wang, Zheng Zhu, and Zhenzhong Jia

10.1109/ROBIO58561.2023.10354574

2023-12-04

International Conference on Robotics and Biomimetics

979-8-3503-2571-3

2023 IEEE International Conference on Robotics and Biomimetics (ROBIO)

1-6

2023-12-2--2023-12-9

泰国

The scientific applications over a decade have demonstrated the robust practicality of wheeled rovers, which are capable of undertaking exploration tasks in various unstructured terrains. However, sinkage and slippage are extremely common on sandy terrains. Fortunately, these risks can be mitigated in a timely manner through the perception of terrain types. Thus, terrain classification technology has become a key solution to addressing these issues. In this paper, we propose a new terrain classification pipeline for a rover over sandy terrains. It exploits the aside-wheel sensor suite to measure sinkage, contact angle, slip ratio, and normal force. Due to the strong correlation between contact characteristics and terrain types form our experiments, we can infer terrain types based on wheel-terrain interaction parameters. We train various machine learning models and obtained classification results, which is evaluated using the record dataset form a singlewheel testbed. Through Experimental comparison, most models achieve high accuracy (over 90%), we analyze the reasons behind performance discrepancies among different models and recommend LGBM as the preferred approach. In the end, we transfer the models to use data collected from rover field tests and achieve a higher level of accuracy, which proves the effectiveness of our approach.

Analytical models Force measurement Correlation Robot vision systems Pipelines Force Machine learning

第一 ; 通讯

EI

20240315404375

人工提交

南方科技大学

Feng Xue, Longteng Hu, Chen Yao, Jin Wei, Yunzhou Li, Peichen Wang, Zheng Zhu, and Zhenzhong Jia. Terrain Classification Based on Wheel-terrain Interaction Measurements using Aside-Wheel Camera[C],2023:1-6.