Thumbnail Image


Publication or External Link







Sound source localization and tracking using auditory systems has been widely investigated for robotics applications due to their inherent advantages over other systems, such as vision based systems. Most existing robotic sound localization and tracking systems utilize conventional microphone arrays with different arrangements, which are inherently limited by a size constraint and are thus difficult to implement on miniature robots. To overcome the size constraint, sensors that mimic the mechanically coupled ear of fly Ormia have been previously developed. However, there has not been any attempt to study robotic sound source localization and tracking with these sensors.

 In this dissertation, robotic sound source localization and tracking using the miniature fly-ear-inspired sensors are studied for the first time.  First, through investigation into the Cramer Rao lower bound (CRLB) and variance of the sound incident angle estimation, an enhanced understanding of the influence of the mechanical coupling on the performance of the fly-ear inspired sensor for sound localization is achieved. It is found that due to the mechanical coupling between the  

membranes, at its working frequency, the fly-ear inspired sensor can achieve an estimation of incident angle that is 100 time better than that of the conventional microphone pair with same signal-to-noise ratio in detection of the membrane deflection. Second, development of sound localization algorithms that can be used for robotic sound source localization and tracking using the fly-ear inspired sensors is carried out. Two methods are developed to estimate the sound incident angle based on the sensor output. One is based on model-free gradient descent method and the other is based on fuzzy logic. In the first approach, different localization schemes and different objective functions are investigated through numerical simulations, in which two-dimensional sound source localization is achieved without ambiguity. To address the slow convergence due to the iterative nature of the first approach, a novel fuzzy logic model of the fly-ear sensor is developed in the second approach for sound incident angle estimation. This model is studied in both simulations and experiments for localization of a stationary source and tracking a moving source in one dimension with a good performance. Third, nonlinear and quadratic-linear controllers are developed for control of the kinematics of a robot for sound source localization and tracking, which is implemented later in a mobile platform equipped with a microphone pair. Both homing onto a stationary source and tracking of a moving source with pre-defined paths are successfully demonstrated.

 Through this dissertation work, new knowledge on robotic sound source localization and tracking using fly-ear inspired sensors is created, which can serve as a basis for future study of sound source localization and tracking with miniature robots.