The developed system uses the Generative Adversarial Network (GAN) and Denoising Autoencoder (DAE) to filter the undesired UAV auto-vibration signal. GAN and DAE are effective to filter the influence of noise and improve recognition performance. However, past research has focused on image or speech signal processing. We measure the ground-truth signal and the real-world noisy signal from a fixed receiver and receiver on the UAV to learn the model. By doing so, we can effectively achieve noise canceling even in an unknown environment. In practice, we consider a device-free human sensing scenario. The UAV is used to analyze the radio map of the environment by scanning the indoor received signal strength (RSS). We can recognize the signal characteristics generated by human activities from the radio map, achieving high-accuracy human detection.

In our research, we extend the existing 2D through-wall wireless sensing to 3D through-wall aerial sensing. Compared to the existing wireless sensing with a fixed receiver, UAV cannot be accurately positioned in a specific location in the air. Due to the vibration characteristics of UAVs, it is challenging to apply the wireless sensing technology in UAV systems.

In the long-term goals of the project, we hope to design efficient and reliable flying robots for mission-critical applications. Through the human-robot collaboration, we can achieve the vision of smart disaster relief and smart security. In addition, the project will develop the UAV communication network testbed for performance evaluation and demonstration. We will design a fully programmable UAV system based on open architecture and open source. The developed system will combine wireless sensing, flight control, and wireless communication technology. It is expected to become the foundation of Taiwan’s autonomous UAV research and achieve the important milestone of AI democratization in the age of drone.