Autonomy and UAVs

We are working on the design of more efficient, that is on-board computable motion planning algorithms with constraints, error detection and navigation. These form the basic functionalities for autonomous flying, as is necessary, for example, for future exploration missions in space travel.

Unmanned vehicles and autonomous flight require dedicated functions beyond conventional automatic flight in the fields of navigation, motion planning, controls and fault management, and environment mapping. Our research focuses on the first three topics for small UAVs with limited computational power and sensor information. It covers both, aerial (primarily) and space vehicles.

Flying without a human pilot is enabled by algorithms for perception of the environment and one's own state of motion. In our activities for the autonomous operation of unmanned aerial vehicles, questions of state estimation therefore play an important role. This includes the theoretical examination of various estimation tasks as well as the development and testing of novel algorithms to solve them.

Particle Filters
A particle filter is a nonlinear probabilistic filter that estimates the probability density of one or more random variables. While the widely used Kalman filter and its various variants depend on the fact that the estimation task is (approximately) based on normal distributions, particle filters are theoretically capable of representing any distributions. The density is not described as a continuous function, but in the form of weighted state hypotheses, the so-called particles. However, the required number of these particles is typically high, which in turn is associated with high computational effort.

A special formulation of the estimation task developed at the Institute of Flight Mechanics and Control allows the efficient implementation on an FPGA. For example, the real-time function of a navigation algorithm based on a particle filter was demonstrated inflight, tackling various constraints typical for small unmanned aerial vehicles regarding size, weight and energy requirements of computing hardware.

Integrated motion planning and state estimation
In order to achieve their full potential, future unmanned aerial vehicles must provide much greater autonomy than today. Navigation and motion planning are crucial functions and therefore the subject of intensive scientific studies. In research and development, both areas are still viewed in isolation from each other, although it has been proven that this separation can severely compromise the performance of the overall system. In fact, significant increases in performance are to be expected if motion planning can work systematically towards improving the quality of navigation. The prerequisite for this, however, is a theoretically sound understanding of the underlying interactions. On this basis, the Institute of Flight Mechanics and Control develops novel integrated algorithms for planning and estimation that will enable future systems to work more precisely, reliably and cost-effectively.

As the name suggests, the iFR develops and designs flight controllers for a variety of different vehicles. This includes controllers for different fixed-wing airplanes, multirotors, helicopters and other more unconventional vehicles.

A problem specific to vehicles using more than four engines is distributing the required torques between all the actuators. This so-called allocation problem has been known for some time and there exist several solutions. There are however new problems in conjunction with multirotors. For example, a manned electric VTOL usually has a very low thrust to weight ratio, which causes actuators to easily reach their saturations. To solve this problem, a special allocation algorithm was developed at iFR, which can exploit the full physically possible control space while not changing the direction of the desired torques.

Tackling certain tasks with a swarm of UAVs offers big potentials. For example a swarm of small sized drones could search a relatively big area in a short period of time. Drone swarms, which are exceedingly extending the capability of UAVs, are currently a hot topic for researchers worldwide.

Algorithms enabling a swarm of fixed-wing UAVs to fly in a specific formation are the subject of current activities at the IFR. The task can be divided in two main parts:

  • Gathering of the members of the swarm and forming a starting formation (Flocking)
  • Sticking to the formation while flying over a specific path (Formation Flight)

The challenge of using fixed-wing UAVs is their prerequisite of a certain minimum speed or curve radius. Contrary to a copter they cannot stop in midair.

To validate the formation flight algorithms a simulation environment for drone swarms was developed at the IFR. Additionally the formation flight of two fixed-wing UAVs could successfully be demonstrated in 2019. The necessary communications have also been developed in-house.

To increase the autonomy of small UAVs, suitable sensors are required to detect obstacles during flight. On the other hand these obstacles need to be dealt with in the controller. Since they often only appear during flight, the planned path has to be changed. The iFR uses different algorithms for this, one of which is nonlinear model-predictive control (NMPC), which optimizes the planned path in every time step. It also allows integrating obstacles and other constraints like for example a maximum velocity.

Such an NMPC has been implemented for fixed-wing aircraft as well as Multicopters and both have been demonstrated in real flight tests. In the shown scenario the Multicopter was tasked with reaching a waypoint in the challenging environment of the Grand Canyon without colliding with any terrain or breaching any constraints. The controller was implemented in a way that allows execution on a small embedded computer in real time.

To the top of the page