The ability to manipulate fluids has always been one of the focuses of scientific research and engineering application. The rapid development of machine learning technology provides a new perspective and method for active flow control. This review presents recent progress in combining reinforcement learning with high-dimensional, non-linear, and time-delay physical information. Compared with model-based closed-loop control methods, deep reinforcement learning (DRL) avoids modeling the complex flow system and effectively provides an intelligent end-to-end policy exploration paradigm. At the same time, there is no denying that obstacles still exist on the way to practical application. We have listed some challenges and corresponding advanced solutions. This review is expected to offer a deeper insight into the current state of DRL-based active flow control within fluid mechanics and inspires more non-traditional thinking for engineering.

Formation path planning is a significant cornerstone for unmanned aerial vehicle (UAV) swarm intelligence. Previous methods were not suitable for large-scale UAV formation, which suffered from poor formation maintenance and low planning efficiency. To this end, this paper proposes a novel millisecond-level path planning method appropriate for large-scale fixed-wing UAV formation, which consists of two parts. Instead of directly planning paths independently for each UAV in the formation, the proposed method first introduces a formation control strategy. It controls the chaotic UAV swarm to move as a single rigid body, so that only one planning can obtain the feasible path of the entire formation. Then, a computationally lightweight Dubins path generation method with a closed-form expression is employed to plan feasible paths for the formation. During flight, the aforementioned formation control strategy maintains the geometric features of the formation and avoids internal collisions within the formation. Finally, the effectiveness of the proposed framework is exemplified through several simulations. The results show that the proposed method can not only achieve millisecond-level path planning for the entire formation but also excellently maintain formation during the flight. Furthermore, simple formation obstacle avoidance in a special case also highlights the application potential of the proposed method.

As UAV popularity soars, so does the mission planning associated with it. Classical planning approaches suffer from the triple problems of decoupling of task assignment and path planning, poor real-time and scalability, and limited adaptability. Aiming at these challenges, this paper proposes a multi-UAV real-time collaborative mission planning method based on UAV states. First, the employed Dubins path accurately represents the distance between tasks and satisfies curvature constraints without smoothing, thus achieving a coupled solution for task assignment and path planning. Then, a series of acceleration techniques are applied to guarantee the real-time performance of the method, including task clustering to reduce the decision space, allocation strategies with fewer iterations, and efficient distance cost calculation methods. To enhance robustness and adaptability, real-time assignment of new tasks and task reassignment due to the reduction of available UAVs are appropriately handled. Finally, simulations highlight that the proposed method only increases the path length by 9.57% compared to benchmark method, while achieving a 4–5 orders-of-magnitude improvement in planning speed, with a single mission planning of about 0.0003 s. Moreover, it easily scales to large-scale scenarios (0.0029 s, with 1000 UAVs and 25,000 tasks).

Inspired by the albatross, this paper presents the construction of a dynamic soaring simulation system with distributed pressure sensors. The advantage of our system lies in harvesting energy from the wind shear layer and estimating the wind information using a pressure-based sensor system. Specifically, the dynamic soaring simulation system contains an offline training stage and an online estimation and control stage. In the offline training stage, computational fluid dynamics simulations are conducted and used as the data source. A surrogate model is established to correlate the local flow conditions and the surface pressure at optimal sensor positions. In the online estimation and control stage, through sensing the pressure information, the real-time wind velocity and wind gradient are estimated by the surrogate model trained in the offline stage. Moreover, wind information is adopted in the simulation of dynamic soaring control. In this study, the simulation system was applied to linear and circular path-following tasks. It was found that the dynamic soaring simulation system with distributed pressure sensors provides an acceptable estimation of wind velocity and wind gradient with a certain time delay caused by numerical differentiation.