Bio-inspired control and navigation - 10 Flashcards
How do animals and robots differ in terms of sensing and control?
Insects have many sensors with very little computation and rapid processing. Large UAVs/Aircrafts have fewer sensors with a large amount of computing, and MAVs have few sensors, small payloa and limited computing. Animals have remarkable sensing and control capabilities and can achieve very complex tasks.
What type of sensory input do flying animals particularly rely on?
Vision, having thousands of lens per eye, highly developed skills such as long range vision, night vision and wide field of vision.
Why is vision-based navigation and control attractive for MAVs?
Nature provides a baseline of the power of vision based navigation and control. Allows avoiding reliance on external inputs such as GPS, vision sensors are compact, light and inexpensive (goood for MAVs), do not emi radiation/signals that can be detected and can be used for object identification which allows for fully autonomous capabilities.
What is optic flow? Explain the basic concept.
Optic flow consists on the apparent motion of objects seen by an observer (eye or camera) due to its relative movement through the environment. Its a velocity field in an image which transforms one image to the next in a sequance. Allows for the estimation of the motion of objects within an image sequence.
What does the optic flow equation state?
It states that the image gradients of a given pixel brightness (under constant brightness assumption) in respect to x and y times their respective flow velocities u and v are equal to the symmetric of the temopral gradient or I_xu + I_yv + I_t = 0. Our goal is to find the optic flow velocities u and v so that we can infer the motiong between two frames.
Image gradients symbolize directional changes in intensity
Temporal gradient its the derivative of intensity
What is the difference between sparse and dense optic flow?
In spars flow (Lucas-Kanade algorithm) we assume constant flow, meaning neighbouring pixels have the same flow (focus on tracking features) whereas the dense optic flow (Horn-Schunk algorithm) we assume a smooth flow, where the flow is assumed as smooth but can vary, focusing on small motion.
Sparse flow has lower accuracy but is faster and less computationally demanding, whereas the dense flow is higly accurate but very demanding computationally.
What are some of the limitations of “traditional” UAV guidance and control approaches?
control and sensing - Limited power, limited computational capability, limited payload/sensing
guidance and control - limited levels of autonomy, limited robustness to external disturbances, limtied interaction with the nvironment, limited edurance over-reliance on GPS/off-board components.
GPS - Limited by satellite signal coverage, Unreliable in cluttered environments, e.g. forests, cities, etc.
How can optic flow patterns be used to determine egomotion?
Since optic flow is the image motion that a vision system experiences during motion we can infer the current motion depending on the visual field motion, such as identify roll, pitch, translation and disatance to objects (by the magnitude of the optic flow vectors).
How can optic flow be exploited for landing, in hover-capable MAVs?
Inspired by bees, by keeping the optic flow generated by the ground constant, it ensures the flight spped is reduced as the ground is approached reaching zero at touchdown for horizontal surfaces. For inclined surfaces, the rate of expansion of the image must be constant so that the speed is reduced gradually in the approach.
How do gannets exploit optic flow for diving into water?
They use the optimal paramter tau which is the time to contact under constant closing velocity when a body is moving torwards a surface (parameter of the optic flow field). This allows to control wing floding based on this parameter to avoid injury. The greater the risk of injury the larger the margin allowed for error.
