An artificial lateral line (ALL) is a bioinspired flow sensing system of an underwater robot that consists of distributed flow sensors. The ALL has achieved great success in sensing the motion states of bioinspired underwater robots, e.g., robotic fish, that are driven by body undulation and/or tail flapping. However, the ALL has not been systematically tested and studied in the sensing of underwater robots driven by rotating propellers due to the highly dynamic and complex flow field therein. This paper makes a bold hypothesis that the distributed flow measurements sampled from the propeller wake flow, although infeasible to represent the entire flow dynamics, provides sufficient information for estimating the lateral motion states of the leader underwater robot. An experimental testbed is constructed to investigate the feasibility of such a state estimator which comprises a cylindrical ALL sensory system, a rotating leader propeller, and a water tank with a planar sliding guide. Specifically, a hybrid network that consists of a one-dimensional convolution network (1DCNN) and a bidirectional long short-term memory network (BiLSTM) is designed to extract the spatiotemporal features of the time series of distributed pressure measurements. A multi-output deep learning network is adopted to estimate the lateral motion states of the leader propeller. In addition, the state estimator is optimized using the whale optimization algorithm (WOA) considering the comprehensive estimation performance. Extensive experiments are conducted the results of which validate the proposed data-driven algorithm in estimating the motion states of the leader underwater robot by propeller wake sensing.