Indoor swarm exploration with Pocket Drones

Pocket drones, weighing less than 50 grams, are small, agile and inherently safe. This makes them suitable for several surveillance tasks such as search and rescue, green-house monitoring and pipe-line inspection. In order for a more efficient search, a swarm of pocket drones would be ideal to explore these types of areas faster. Current methods and navigation techniques will not be suitable due to their extensive requirements for the platform's computational capabilities and memory storage. This dissertation will therefore focus on designing a new strategy for a swarm of pocket drones for both low-level and high-level navigation in an indoor environment. The first part of this dissertation focuses on the low-level-navigation capabilities of the swarm of pocket drones, by first looking at the individual. We developod Edge-FS (Flow \& Stereo), which enabled the stereo-camera to detect both obstacles and the drone's velocity at the same time. A further necessity for swarm operations is for multiple pocket drones to avoid each other. An on-board relative localization scheme based on the Received Signal Strength Intensity (RSSI) of the inter-drone communication was developed to make this possible. Two pocket drones with a forward-facing stereo-camera were communicating with each other by means of Bluetooth and by fusing the RSSI with their velocity (estimated by Edge-FS). With this, two pocket drones were able to fly together in a room while avoiding the walls and each other. The second part of this dissertation focuses on high-level-navigation. Since conventional navigation strategies cannot fit on-board the pocket drones, we investigated an alternative method: bug algorithms. We present a literature survey and comparison of the existing techniques and evaluation on their suitability for deployment for real-world scenarios. We found that with increasing sensor errors and estimation drift, all existing bug algorithms’ performances decreased. This provided us valuable insights for the design of a novel bug algorithm for high-level-navigation. Finally, we developed and demonstrated a bug-algorithm-based navigation strategy for multiple pocket drones for indoor exploration and homing. We named this technique the swarm gradient bug algorithm (SGBA) and it enabled the pocket drones to explore a floor of an inside building and return to its original position by the RSSI-gradient of a radio beacon. Once two pocket drones come into each other's proximity, one will avoid the other and coordinate its own preferred search direction based on the information it has received (from the other).