ABSTRACT Land–air hybrid locomotion robots are popular because of their exceptional cross‐domain operational capabilities, although control challenges during air‐to‐ground transitions remain a key research focus. Flying quadrupedal robots exhibit excellent terrain adaptability, rapid mobility, and remarkable static load performance during flight, making them highly valuable for research and development. To increase adaptability to complex terrains, we model the robot's legs as a mass‐spring‐damper system and propose a compliant adaptation strategy based on plantar pressure information for irregular surfaces. Landing is divided into five distinct stages, enabling a smooth transition from flight to ground locomotion. To increase system robustness during landing, we introduce a reinforcement learning‐based self‐tuning generalized PID control framework. In this framework, the critic neural network (NN) approximates the cost function, and the actor NN dynamically adjusts the generalized PID gains via feedback from the critic. The optimized control inputs drive the system towards optimal performance. To address actuator input saturation, we reformulate the problem with a smooth hyperbolic tangent function, transforming it into a variable‐gain control problem with a known control direction. The Nussbaum function mitigates the adverse effects of input saturation. This approach systematically adjusts PID gains for flying quadrupedal robots, yielding a simple yet computationally efficient framework that avoids extensive parameter tuning. Finally, the stability of the closed‐loop system is proven via Lyapunov theory. The numerical simulations conducted on a flying quadrupedal robot prototype—integrating a one‐dimensional vectoring thrust VTOL platform with a Mini Cheetah quadruped—further validate the effectiveness of the proposed strategy.
Liu et al. (Fri,) studied this question.