文献类型：期刊文献

英文题名：An autonomous navigation approach for unmanned vehicle in off-road environment with self-supervised traversal cost prediction

作者：Zhou, Bo[1];Yi, Jianjun[1];Zhang, Xinke[1];Wang, LianSheng[1];Zhang, Sizhe[2];Wu, Bin[2]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, 130 Meilong Rd, Shanghai 200237, Peoples R China;[2]Aerosp Syst Engn Shanghai, 1777 Zhongchun Rd, Shanghai 201108, Peoples R China

年份：2023

卷号：53

期号：17

起止页码：20091

外文期刊名：APPLIED INTELLIGENCE

收录：;EI(收录号：20231413840980);WOS:【SCI-EXPANDED(收录号:WOS:000960222400002)】；

基金：AcknowledgementsThis paper was supported by the Major Program of National Natural Science Foundation of China under Grant No. 61690214, Shanghai Science and Technology Action Plan under Grant No.18DZ1204000, 18510745500, 18510750100, 18510730600, Shanghai Aerospace Science and Technology Innovation Fund (SAST) under Grant No. 2019-080, 2019-116 and the Natural Science Fund of China (NSFC) under Grant No.51575186 the National Defense Basic Scientific Research Program of China (Grant No. JCKY2021606B002).

语种：英文

外文关键词：Self-supervised learning; Traversal cost prediction; Motion planning; Off-road environments

摘要：This paper presents a self-supervised learning-based terrain traversal cost prediction method that addresses different orientations and velocities to aid autonomous navigation in off-road environments. First, a cost prediction network is proposed to implement a mapping of the local terrain information around a vehicle to the traversal cost. Second, we propose an automatic data collection and self-labelling algorithm to achieve self-supervised learning for this network. Third, we proposed a map-free navigation strategy aimed at the terrain obstacles. This strategy incorporates the traversal cost prediction into a sampling-based trajectory planner, enabling the consideration of traversal orientation and velocity when estimating the traversal cost. Finally, both the proposed prediction method and the navigation strategy are extensively compared. The results show that our proposed traversability estimation method outperforms existing methods using convolutional neural networks (CNNs). Simultaneously, in both simulation and real-world experiments, our approach exhibits effective and safe autonomous navigation capabilities in off-road environments.