Wendong Niu

Yaw regulation of biomimetic underwater robots is complicated by flexible body motion, nonlinear hydrodynamics, and coupled actuation. This study examines whether a policy trained in simulation can be deployed on an existing robotic sea lion (RSL) without changing its hardware or low-level controllers. A deep deterministic policy gradient (DDPG) controller was formulated from measurable states and available actuator commands and trained in Webots using a model calibrated from previous tank tests. Four manually selected reward-weight settings and command update rates of 1, 2, 5, and 10 Hz were examined as deployment-oriented sensitivity comparisons, after which a 5 Hz policy was evaluated in six tank trials. Performance was reanalyzed using the circular-angle mean absolute error (MAE) and root mean square error (RMSE). In the two straight-swimming trials, the per-trial MAE was 1.01–1.12°, and the RMSE was 1.10–1.46°. In the four turning trials, evaluated from the first target crossing to the end of each record, the MAE was 1.71–3.63°, the RMSE was 2.10–4.20°, and the maximum overshoot was 2.99–7.70°. Despite the transient differences between simulations and experiments, the controller regulated the robot toward the target headings in all six tank trials. These results demonstrate the successful sim-to-real deployment of reinforcement-learning-based yaw control under low-frequency communication constraints and provide experimental evidence for its application to biomimetic underwater robots.

1. Covering approximately 70% of the Earth’s surface, the ocean represents a vast reservoir of strategic resources. Since the turn of the century, there has been a global surge in interest regardin...

A key advantage of heterogeneous underwater glider formations is their differentiated depth coverage, which allows for highly efficient observation of marine areas with steep depth gradients such as continental slopes and submarine canyons. However, the cooperative control of such formations remains a critical challenge due to heterogeneous dynamics, divergent control strategies, and difficulties in inter-agent coordination. Taking the Petrel series of underwater gliders as the research object, this study proposes a cooperative control scheme for heterogeneous underwater glider formations based on deep reinforcement learning. First, we establish a Markov model for the heterogeneous formation under time-lag communication based on a virtual leader architecture. Then a feature attention mechanism (FAM) is integrated into the twin delayed deep deterministic policy gradient (TD3) algorithm to construct the FAM-TD3 algorithm. Finally, the effectiveness of the proposed cooperative control scheme is validated through a sea trial. This study provides a feasible technical path for the refined observation of heterogeneous underwater glider formations in complex sea areas such as continental slopes.

The airdroppable underwater vehicle is a new type of autonomous underwater vehicle featured by rapid deployment and a relatively loose requirement for the sea conditions, which has become a research hot spot in recent years. However, predicting landing points remains challenging due to various factors. This study establishes a 14-degree-of-freedom (DOF) dynamical model for the vehicle–parachute system of the airdroppable underwater vehicle. The established model is utilized to study the distribution pattern of the landing points under the influence of uncertainty factors, and comparisons were made with a naive zeroth-order guess model and a six-DOF model. In low-altitude deployment, changes in deployment height are revealed to have a weak impact on the landing point distribution, while deployment velocity, wind speed, and wind direction show a significant impact. Compared with the naive zeroth-order guess model and the six-DOF model, the 14-DOF model has an average improvement of 66.00% and 52.52% in the accuracy of the 95% confidence circle radius and an average improvement of 65.12% and 47.96% in the circular error probable. Finally, the effectiveness of the landing point distribution pattern of the 14-DOF model was verified to a certain extent through two airdrop experiments using the “Petrel-A” underwater glider. However, more data will be needed in the future to support the accuracy of the model.