Ying Yu

Ying Yu

Auto-generated profile, based on publicly available information

Ying Yu

Key Laboratory of Ocean Observation-Imaging Testbed of Zhejiang Province, Ocean College, Zhejiang University, Zhoushan, China

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
13 March 2024 in IEEE Transactions on Systems, Man, and Cybernetics: Systems
Verified
Journal Article
Closed access
YYYing Yu
Ying Yu
Huarong ZhengHuarong Zheng
Huarong Zheng
WXWen Xu
Wen Xu
Autonomous underwater vehicles (AUVs) are widely used in sampling on-site the seawater parameters, such as temperature, salinity and biomass for better understanding the ocean. The AUV path needs to be carefully planned in order to maximize the sampled information within the power constraints, which is known as the informative path planning (IPP). The existence of ocean currents further complicates the problem. This article proposes an IPP method for AUVs under the influence of ocean currents via combining the probabilistic roadmap and $Q$ -learning. Specifically, the $Q$ -learning algorithm builds an informative optimal AUV path by traversing a learned and updated $Q$ -table. The $Q$ -value in the table represents the expectation of the obtained reward if taking a certain action moving from one position to another. Considering the characteristics of the IPP task, we design the reward matrix in $Q$ -learning using the prior knowledge on the environment information. A convergent $Q$ -table guarantees that only one complete training and learning is required to generate the path between any two positions. This feature facilitates converting the possible repetitive path plannings into simple search problems, and thus the automatic return is easily realized whenever the AUV residual energy is insufficient. Moreover, to improve the efficiency of the $Q$ -learning algorithm, a probabilistic roadmap with random sampling is generated and combined with the $Q$ -learning. Various simulations and comparisons are carried out. The results demonstrate the effectiveness of the proposed IPP algorithm, showing that the convergence of the path planning can be achieved quickly and successfully. The superiority in terms of efficient return path planning over the traditional path planning method, RRT*, is also demonstrated.