Bidirectional thrust control for quadrotor safety

Abstract

Quadrotors performing aerial tasks are vulnerable to sudden external disturbances, which may lead to instability, control loss, or even structural damage such as broken arms or frame failure. These threats are particularly critical during flight, where recovery opportunities are limited. To address this, we propose a bidirectional thrust control framework that improves mid-air impact resilience. A lightweight recurrent neural network (RNN)-attention module detects and evaluates external forces in real time. When the disturbance is mild, a model predictive control (MPC) + active disturbance rejection control (ADRC) controller ensures stability; when it nears a critical level, the system switches to a flipping recovery policy that exploits bidirectional thrust to regain balance. Experiments validate robustness and safety under sudden external disturbances.