Bernoulli Effect and Deep Learning Based Ground Height Sensing for Micro Quadrotors

Micro Quadrotors are easily affected by aerodynamic disturbances when flying in narrow, near-ground environments. Currently widely used height sensors, due to insufficient accuracy, large dead zones, and poor environmental adaptability are difficult to meet the demands of low-altitude autonomous flight. To address this issue, this paper proposes a near-ground height sensing method for Quadrotors based on the Bernoulli effect and deep learning. This method embeds multiple pressure sensors in a ring at the bottom of a duct to capture in real time the pressure distribution features between the ducted Quadrotors’ bottom and the ground caused by the Bernoulli effect. Based on momentum principles and Bernoulli’s equation, a pressure-height model is established and corrected through computational fluid dynamics (CFD) simulations to resolve problems of airflow leakage and backflow. To handle errors induced by attitude tilt and turbulence disturbances, this paper constructs a lightweight feedforward neural network (FNN) model to fuse the mechanistic model predictions from multiple pressure sensors, learning the nonlinear mapping relationship between these predictions and the Quadrotors’ actual height, thus achieving comprehensive compensation for system errors. Experimental results indicate that, in the same smooth ground environment, the root mean square error (RMSE) of height predictions with this method improves by 75.6% compared to the accuracy of Time-of-Flight (ToF) sensors. Additionally, in scenarios where ToF sensors fail, such as on transparent surfaces, this method still maintains good data usability (RMSE = 0.159mm). This study provides a new solution for height sensing in micro Quadrotors operating in complex low-altitude environments, particularly suitable for autonomous flight control in height-constrained spaces.