Event-driven visual sensors have attracted interest from a number of different research communities. They provide visual information in quite a different way from conventional video systems consisting of sequences of still images rendered at a given "frame rate." Event-driven vision sensors take inspiration from biology. Each pixel sends out an event (spike) when it senses something meaningful is happening, without any notion of a frame. A special type of event-driven sensor is the so-called dynamic vision sensor (DVS) where each pixel computes relative changes of light or "temporal contrast." The sensor output consists of a continuous flow of pixel events that represent the moving objects in the scene. Pixel events become available with microsecond delays with respect to "reality." These events can be processed "as they flow" by a cascade of event (convolution) processors. As a result, input and output event flows are practically coincident in time, and objects can be recognized as soon as the sensor provides enough meaningful events. In this paper, we present a methodology for mapping from a properly trained neural network in a conventional frame-driven representation to an event-driven representation. The method is illustrated by studying event-driven convolutional neural networks (ConvNet) trained to recognize rotating human silhouettes or high speed poker card symbols. The event-driven ConvNet is fed with recordings obtained from a real DVS camera. The event-driven ConvNet is simulated with a dedicated event-driven simulator and consists of a number of event-driven processing modules, the characteristics of which are obtained from individually manufactured hardware modules.

/pdf/mapping-from-frame-driven-to-frame-free-event-driven-vision-2j69byxuem.pdf

Mapping from Frame-Driven to Frame-Free Event-Driven Vision Systems by Low-Rate Rate Coding and Coincidence Processing--Application to Feedforward ConvNets

Optic flow regulation: the key to aircraft automatic guidance

Neuromorphic engineering [Mead, 1989, Mead, 1990, Douglas et al., 1995] applies the computational principles used by biological nervous systems to those tasks that biological systems perform easily, but which have proved difficult to do using traditional engineering techniques. These problems include visual and auditory perceptive processing, navigation, and locomotion. Typically, current neuromorphic systems are hybrid analog-digital electronic systems fabricated using CMOS VLSI technology [Mead, 1989, Douglas and Mahowald, 1995]. Research has focused on the sub-threshold analog operation of these circuits, because in this regime it is possible to construct compact analog circuits that compute various biologically relevant operations such as logarithms, exponents and hyperbolic tangents.

A pulse-coded communications infrastructure for neuromorphic systems

Flying insects are intelligent micromachines capable of exquisite maneuvers in unpredictable environments. Understanding these systems advances our knowledge of flight control, sensor suites, and unsteady aerodynamics, which is of crucial interest to engineers developing intelligent flying robots or micro air vehicles (MAVs). The insights we gain when synthesizing bioinspired systems can in turn benefit the fields of neurophysiology, ethology and zoology by providing real-life tests of the proposed models. This book was written by biologists and engineers leading the research in this crossdisciplinary field. It examines all aspects of the mechanics, technology and intelligence of insects and insectoids. After introductory-level overviews of flight control in insects, dedicated chapters focus on the development of autonomous flying systems using biological principles to sense their surroundings and autonomously navigate. A significant part of the book is dedicated to the mechanics and control of flapping wings both in insects and artificial systems. Finally hybrid locomotion, energy harvesting and manufacturing of small flying robots are covered. A particular feature of the book is the depth on realization topics such as control engineering, electronics, mechanics, optics, robotics and manufacturing. This book will be of interest to academic and industrial researchers engaged with theory and engineering in the domains of aerial robotics, artificial intelligence, and entomology.

Flying insects and robots

We have developed a visually based autopilot which is able to make a micro air vehicle (MAV) automatically take off, cruise and land, while reacting adequately to wind disturbances. We built a proof-of-concept, tethered rotorcraft that can travel indoors over an environment composed of contrasting features randomly arranged on the floor. Here we show the feasibility of a visuomotor control loop that acts upon the thrust so as to maintain the optic flow (OF) estimated in the downward direction to a reference value. The sensor involved in this OF regulator is an elementary motion detector (EMD). The functional structure of the EMD was inspired by that of the housefly, which was previously investigated at our laboratory by performing electrophysiological recordings while applying optical microstimuli to single photoreceptor cells of the compound eye. The vision based autopilot, which we have called OCTAVE (optic flow control system for aerospace vehicles) solves complex problems such as terrain following, controls risky maneuvers such as take off and landing and responds appropriately to wind disturbances. All these reputedly demanding tasks are performed with one and the same visuomotor control loop. The non-emissive sensor and simple processing system are particularly suitable for use with MAV, since the tolerated avionic payload of these micro-aircraft is only a few grams. OCTAVE autopilot could also contribute to relieve a remote operator from the lowly and difficult task of continuously piloting and guiding an UAV. It could also provide guiding assistance to pilots of manned aircraft.

/pdf/visually-guided-micro-aerial-vehicle-automatic-take-off-4nlhv2as8c.pdf

Visually guided micro-aerial vehicle: automatic take off, terrain following, landing and wind reaction

We present two algorithms for estimating the velocity of a visual stimulus and their implementations with analog circuits using CMOS VLSI technology. Both are instances of so-called token methods, where velocity is computed by identifying particular features in the image at different locations; in our algorithms, these features are abrupt temporal changes in image irradiance. Our circuits integrate photoreceptors and associated electronics for computing motion onto a single chip and unambiguously extract bidirectional velocity for stimuli of high and intermediate contrasts over considerable irradiance and velocity ranges. At low contrasts, the output signal for a given velocity tends to decrease gracefully with contrast, while direction-selectivity is maintained. The individual motion-sensing cells are compact and highly suitable for use in dense 1-D or 2-D imaging arrays.

J. Krammer

Papers

Pulse-based analog VLSI velocity sensors