We are developing a network of autonomous drones which adapt flight patterns in real-time according to sensor measurements and mission objectives. Over the past year, we designed and implemented the first version of our network of drones that: (i) coordinate in autonomous flight via software-defined radios, (ii) fly off-grid without requiring a ground control station or air-to-ground network, and (iii) embed on-board machine learning missions based on measured data which is shared among drones.
Our design focus is on drone networks that are autonomous, perform sensing missions, and are tetherless. These three features have not previously been realized in a single design. Indeed, in contrast to single-drone solutions, we want our networked drones to communicate and coordinate among themselves. They should form a dynamic mesh and employ software defined radios (SDRs) to enable programmability and advanced communication and networking features. This helps drones adapt their carrier frequency in order to realize longer range as needed to maintain connectivity with other drones, at the potential cost of less bandwidth being available at lower frequencies. In addition, unlike existing systems, our drones should operate without the necessary need for human or machine control from the ground. This enables flight in areas not served by Wi-Fi or cellular networks. Moreover, the drone’s decisions and flight paths should be adapted in real-time according to the measured sensor data.
Over the past year of test flights, we learned several lessons regarding the design of robust drone systems. One of the first challenges was the choice of the platform. On one side, commercial platforms such as Parrot, DJI, etc. offer stable solutions, but are closed and single-purpose, leaving no space for extensions. Therefore, we built a custom system relying on open-source software and open hardware. In this process, it has been crucial that the selected components are reliable, and at the same time, meet the weight and power budget requirements. Indeed, lower weight improves takeoff/climb and landing performance during a mission. Given a maximum payload of 1.5 kg and 500 grams of battery, it is crucial to carefully budget the payload. Regarding the power budget, a good rule of thumb when selecting the battery is to use 1,000 mAH (milliamp hours) per motor. Another fundamental task regards maintenance: drones are indeed different from classical tech gadgets that at most require battery checks. Before every flight, it is essential to check the integrity of the drone frame, the battery conditions, and calibration of the Flight Control sensors. We discovered that most crashes that we had during our tests were due to bad wiring to the GPS and bad calibration of the Flight Controller sensors. For example, if the barometer of the Flight Controller yields incorrect altitude, a safe landing is unlikely.
Publications
- Petrolo, Y. Lin, and E. Knightly, “ASTRO: Autonomous, Sensing, and Tetherless netwoRked drOnes,” in Proceedings of ACM DroNet 2018, Munich, Germany, June 2018.