Revolutionizing Drone Safety and Performance
The aim of this research project is to fundamentally improve the safety of drones by developing a new safe envelope predictive (SENCE) flight control system. The goal of the SENCE controller will be to prevent loss-of-control after failures, and allow the drone to conduct a controlled landing, or complete a critical mission such as search and rescue. SENCE will complement existing autopilot or artificial intelligence systems, and will add a new layer of safety to any system it is applied to.
SENCE is a fault tolerant control system that can stabilize the drone after hardware failures. In addition, SENCE will also be capable of making predictions of the remaining maneuvering envelope and acting intelligently on this information. The fault tolerant control aspects of the SENCE controller were recently tested in the TU-Delft Open Jet wind tunnel Facility (OJF). In these tests one of the rotors was suddenly stopped to simulate a motor failure. After this, the fault tolerant control based on the adaptive incremental nonlinear dynamic inversion (A-INDI) controller quickly recovers the drone, allowing it to perform a safe landing. It should be noted that a similar motor failure in a standard quadrotor drone would lead to a fatal crash. This aspect of this research project was recently highlighted on IEEE spectrum.
Fault tolerant control is an important aspect of SENCE, but without online envelope prediction and protection, a damaged drone is still susceptible to loss-of-control. We have been able to demonstrate loss-of-control with a damaged drone in experiments. In this case, one rotor of the drone was removed to simulate a complete rotor loss, after which flight tests were conducted in the OJF. The Adaptive INDI fault tolerant controller is able to maintain position of the drone up to 9.5m/s forward velocity. As the wind speed is further increased to 10m/s the drone suddenly loses control and crashes. The maximum velocity of the nominal drone is 16m/s. The crucial point here is that there currently is no way of predicting that the loss-of-control event would take place between 9.5m/s and 10.0m/s. The complete SENCE system will be able to predict this reduction of performance and prevent the drone from entering into a situation from which it cannot recover.
A new insight has been that the standard flight envelope should be replaced by a probabilistic interpretation of the maneuvering envelope resulting in a ‘landscape’ of safety levels. This in turn will allow for a much more precise balancing of safety and performance to meet specific mission needs.Various new methods for obtaining the probabilistic envelope are currently being explored, including stochastic PDE theory, Monte Carlo methods, and local envelope sensing. To obtain data for our algorithms, flight test experiments with damaged drones are conducted in the CyberZoo motion capturing facility operated by the Delft Robotics Institute and in the TU-Delft Open Jet wind tunnel Facility (OJF). This project will be conducted in collaboration with the TU-Delft MAVlab which is a world leader in the field of small autonomous drones.