Bio-Inspired Small Field Perception for Navigation and Localization of MAV's in Cluttered Environments

Loading...
Thumbnail Image

Publication or External Link

Date

2015

Citation

Abstract

Insects are capable of agile pursuit of small targets while flying in complex cluttered environments. Additionally, insects are able to discern a moving background from smaller targets by combining their lightweight and fast vision system with efficient algorithms occurring in their neurons. On the other hand, engineering systems lack such capabilities since they either require large sensors, complex computations, or both. Bio-inspired small-field perception mechanisms have the potential to enhance the navigation of small unmanned aircraft systems in cluttered unknown environments. In this dissertation, we propose and investigate three methods to extract information about small-field objects from optic flow. The first method, \textit{flow of flow}, is analogous to processes taking place at the medulla level of the fruit-fly visuomotor system. The two other methods proposed are engineering approaches analogous to the figure-detection sensitive neurons at the lobula. All three methods employed demonstrated effective small-field information extraction from optic flow.

The methods extract relative distance and azimuth location to the obstacles from an optic flow model. This optic flow model is based on parameterization of an environment containing small and wide-field obstacles. The three methodologies extract the high spatial frequency content of the optic flow by means of an elementary motion detector, Fourier series, and wavelet transforms, respectively. This extracted signal will contain the information about the small-field obstacles.

The three methods were implemented on-board both a ground vehicle and an aerial vehicle to demonstrate and validate obstacle avoidance navigation in cluttered environments.

Lastly, a localization framework based on wide field integration of nearness information (inverse of depth) is used for estimating vehicle navigation states in an unknown environment. Simulation of the localization framework demonstrates the ability to navigate to a target position using only nearness information.

Notes

Rights