Safety is critical for autonomous drones. However, today's commercial quadrotor drones are still not resilient to severe motor failures. Especially in GPS-denied environments, a quadrotor with loss of one motor will fast yaw spin, causing significant position drift, which will crash the drone and may harm people on the ground. For this reason, we propose a vision-based control algorithm to save the quadrotor after the complete loss of a single motor. This algorithm uses an onboard camera to provide pose and velocity estimates, and a fault-tolerant flight controller to generate control commands for the remaining motors. Despite the inevitable fast spinning motion after motor failure, the algorithm can control the quadrotor to follow waypoints and conduct safe landing without the aid of external localization systems such as the GPS.
We demonstrate the effectiveness of the proposed fail-safe control algorithm in real-flights by autonomously following a designated path after the in-flight failure of one motor. In addition, an event-based camera is also tested as the visual input, showing greater reliability in low-light conditions.