Data underlying the publication: All Eyes, no IMU: Learning Flight Attitude from Vision Alone

Datasets used for training, validating and testing of the networks deployed on the drone for the publication: All Eyes, no IMU: Learning Flight Attitude from Vision Alone as published in NPJ Robotics (link will be provided when it is available). In this work, the authors have demonstrated for the first time that a quadrotor can be controlled solely based on vision inputs. A recurrent network was trained to estimate both attitude and angular rate, which was used in the control loop of a flying drone. The output of the event based camera, together with sensor data from the flight controller and the control commands is given here as rosbags, ulg files from the PX4 autopilot, .csvs from the motion capture system and some videos showing the performance as .MOV and .mp4. <br>

本数据集用于训练、验证与测试发表于NPJ Robotics期刊的论文《All Eyes, no IMU: Learning Flight Attitude from Vision Alone》中部署于无人机的神经网络，该论文的公开链接待上线后另行提供。本研究首次证实，四旋翼无人机可仅依靠视觉输入实现自主控制。研究人员训练了一款循环神经网络（Recurrent Network）以同时估计无人机姿态与角速度，并将该网络应用于飞行无人机的控制回路中。本数据集提供以下类型的数据：事件相机（Event-based Camera）的输出数据、飞行控制器的传感器数据与控制指令，具体格式包括rosbags数据包、PX4自动驾驶仪生成的.ulg文件、动作捕捉系统导出的.csv格式文件，以及展示实验运行性能的.MOV与.mp4格式视频文件。