Unmanned Aerial Vehicles (UAVs) are expected to fully exploit their maneuvering capabilities to conduct aggressive maneuvers to complete tasks such as racing and aerobatics even in harsh environments. However, most of related works often focus on slow movements only suitable for regular tasks in well-structured environments. This motivates us to propose a highly robust control scheme designed specifically for UAV aggressive maneuvers, encoded by rapid and large changes in either position or attitude. The core of our approach is the data informed incremental dynamics and the quaternion facilitated control scheme. The former facilitates the model-free control that eliminates the requirement to model the complex dynamics at high velocities and accelerations accurately; While the latter avoids potential singularities during large angular changes. In particular, we first utilize one-step-backward data to construct incremental dynamics, a model-free representation of the UAV dynamics. Then, the constructed incremental dynamics serves as the basis for the development of both position-priority and attitude-priority control schemes, wherein the novel quaternion based aggressive maneuver tracking controllers are designed with complete theoretical analysis. The aggressive maneuvers such as roulette, barrel roll, multiple-flip and cobra maneuvers are chosen for tri/quad/hexa/octocopter platforms to evaluate the performance of our proposed position-priority and attitude-priority control schemes, during which the linear velocity up to 20 m/s, the angular changes up to 360 ◦.

In recent years, safe reinforcement learning (RL) with the actor-critic structure has gained significant interest for continuous control tasks. However, achieving near-optimal control policies with safety and convergence guarantees remains challenging. Moreover, few works have focused on designing RL algorithms that handle time-varying safety constraints. This paper proposes a safe RL algorithm for optimal control of nonlinear systems with time-varying state and control constraints. The algorithm’s novelty lies in two key aspects. Firstly, the approach introduces a unique barrier force-based control policy structure to ensure control safety during learning. Secondly, a multi-step policy evaluation mechanism is employed, enabling the prediction of policy safety risks under time-varying constraints and guiding safe updates. Theoretical results on learning convergence, stability, and robustness are proven. The proposed algorithm outperforms several state-of-the-art RL algorithms in the simulated Safety Gym environment. It is also applied to the real-world problem of integrated path following and collision avoidance for two intelligent vehicles – a differential-drive vehicle and an Ackermann-drive one. The experimental results demonstrate the impressive  sim-to-real transfer capability of our approach, while showcasing satisfactory online control performance.