Alexander Hellemans of IEEE Spectrum recently reported that a number of drone designers are currently exploring a technique known as optical flow, which is based on the speed of objects passing a subject’s field of vision.
“[We] use optical flow to give us a sense of how fast we’re going when we’re driving,” Hellemans explained. “But unlike humans in cars, drones have a third dimension to worry about. They also have to keep track of their height in order to land successfully.”
Stereovision, says Hellemans, allows drones to estimate distances. However, the resulting measurements are imprecise – if the baseline between sensors is too small. This is precisely why Guido de Croon, a researcher at the Delft University of Technology in the Netherlands, is analyzing optical flow data captured by a single camera.
“When [the drone] gets close to the ground, at a certain point, the control system will become unstable; it starts to oscillate,” de Croon wrote in the most recent issue of Bioinspiration & Biomimetics. “It is not that the ground causes an aerodynamical effect; the drone itself is inducing these oscillations.”
[youtube https://www.youtube.com/watch?v=Fm7SMJp8EA4]
As de Croon observes, the oscillation occurs at a specific distance from the ground.
“We could see this as a problem, of course, but actually, it is also an opportunity for the robots,” he added. “If it can detect this oscillation, then it can actually know its height.”
Dr. Patrick Gill, Rambus Principal Research Scientist, told us that the above-mentioned research demonstrates a new method of determining how to land a drone or maintaining a pre-assigned distance above the ground.
“Optic flow is the apparent motion of objects caused by your own motion; as you descend to the ground a downward-pointing camera will see objects on the ground looming larger as it approaches them,” he explained. “In principle, maintaining a constant degree of optic flow as you land will give you a smooth landing; too much flow means you’re falling too fast and should increase thrust. However, this simple feedback loop breaks down when you get too close to the ground – if there is any delay in optic flow sensing or your drone’s reaction.”
The fundamental trick illustrated by de Croon’s research, says Gill, is that one can build a simple landing control system based on optic flow using two control loops.
“The first loop adjusts the upward thrust to try to keep the optic flow constant,” he noted. “However, delay in optic flow sensing or actuation in this first loop will cause instability, leading to oscillations, at a specific height. The specific height oscillations start depends both on the delay in the first loop and also on how aggressive the descent is, or more technically, the gain of the first loop.”
According to Gill, the advance published in the current work by Guido de Croon is that a second control loop can detect instability in the first loop.
“Any fixed flight system with a given delay and gain will become unstable at a certain altitude. The second loop detects this instability and uses it as an indicator that a certain height has been reached,” he said. “You can then ask the second loop to maintain an edge-of-stability operation (in which case your drone will fly at a fixed altitude determined by the first loop’s gain) or you can turn down the gain in the first loop smoothly, staying at the edge of stability, until the drone is known to be low enough that it can land by cutting its thrust.”
As Gill notes, certain sensors Rambus scientists are developing may be appropriate for low-weight optic flow detectors.
“For example, lensless smart sensors (LSS) can replace focusing lenses with tiny diffractive elements and perform a variety of tasks, such as optic flow measurement, using a minimum of optics,” he confirmed. “In the case of a drone landing system, you probably want a lightweight sensor that is able to use information from a pretty wide field of view, to increase the chance there’s some part of the ground with usable features that’s in view.”
Combining lightweight optics with elegant control loops like Guido, Gill concludes, could make for tiny, robust flyers that reliably land with grace.
Leave a Reply