Robotic Sensing: Adaptive Robotic Control for Improved Acoustic Source Localization in 2D

Raphael Schwartz, Zachary Knudsen, Phani Chavali, Patricio S. La Rosa, Ed Richter, and Arye

Nehorai

Department of Electrical and Systems Engineering

Robotic Platform

Abstract

In this project we expand our previous work entitled "Design of a Robotic

Platform and Algorithms for Adaptive Control of Sensing Parameters". We

have shown that the performance of our algorithm for acoustic source

location in 2D can be improved by adaptively controlling the microphone

array geometry. To this end, we built a robotic microphone array with

capability of autonomous control of array geometry constrained to

movement in 1D. In this project we increase the degrees of freedom of our

robotic platform and design a new controlling algorithm in order to

improve even further the performance. In particular, our robots move in 2D

and the pair of microphones can also rotate independently of the robot

orientation. A heuristic approach for the control of robot locations is

presented and validated with real experiments. Labview and Matlab are

used for the implementation of the system.

Motivation

It can be shown that given a particular array geometry and sampling frequency, there are a

finite number of possible locations which can be estimated using two pairs of

microphones. This set of possible points is not uniformly distributed, as is shown in

Figure 2. Further the resolution (defined as the number of points around the actual source

location) depends on the orientation of the microphone pairs. In this work, we intend

mount the microphone pairs on a robot and adaptively move them such that they have a

good resolution around the source.

Mounting the microphone array on robots allows us to change the physical

parameters of the system in real time.

Altering the distance between microphone pairs affects the spatial distribution of the

possible estimation points.

Shifting the array brings the source in-line with the array center.

These movements increase the resolution near the source and improve the estimation.

Higher Resolution

Lower Resolution

Overview

Goal:

Linear Array Configuration

Design a system capable of acquiring measurements to estimate the

acoustic source position in real time and adaptively move the microphone

pairs in 2-D to improve localization resolution.

Approach:

Dataflow programming techniques were used to implement signal

processing architectures. A heuristic algorithm was used to control

movement based on our experimental observations.

Background :

y

v (TimeDelay)

cos

d

1 s

Source

P

[tan( 2 ) tan(1 )]

*

x 2

tan( 2 ) tan(1 )

y*

y*

P tan( 2 ) tan( 1 )

tan( 2 ) tan(1 )

R

Real Time Architecture and Controller Algorithm

Adaptive Control Algorithm for Robot movement

A simulation based heuristic algorithm is used to

determine potential resolution improvement of moving

the robots in each of the systems four degrees of

freedom. For each degree of freedom a number of

simulations are performed to model a variety of different

robotic movements. The robotic movements which

produced a significant simulated resolution improvement

are then used as the commands for the next movement.

This process is performed until the microphones have

been moved to a configuration which best optimizes

localization resolution, given the physical constraints of

movement. A threshold is defined to characterize the

optimal resolution.

The four degrees of freedom which the algorithm seeks

to optimize are the:

2

d

x*

P

L

1

x

d

Sensor

Variables

L is the length of the linear array

is the wavelength of the acoustic waveform

R is the radial distance from the source of the array

d if the distance between sensors pairs

Rotated Array Configuration

Figure 2

Distance between the robots

Orientation of the robots with respect to the source

Movement in the horizontal axis with respect to the

source (shifting movement)

Movement in the vertical axis with respect to the

source (approaching/retreating movement)

The adaptive algorithm also utilizes the robots rotational

capability to point each of the microphone pairs to best

face the source, further improving estimation

Statistical Signal Processing &

Position Control Al gorithms

PC

Sound

Output &

Waveform

Generation

Data

Acquisition

&

Microphone

New array

Configuration

Measurements

y

x

Resolution

Figure 4

Error

Compute

Source

Estimation

Controller &

Robots

Calculate

Resolution

Simulation

Processing &

Plotting

Move the

Robot to new

position

Robot

Movement

Command

Resolution

Improvement

?

Figure 5

Figure 3

References

Joshua York, Acoustic Source Location Using Cross-correlation Algorithms, Fall 2008, http://ese.wustl.edu/~nehorai/josh/students.cec.wustl.edu/_jly1/

Chase LaFont, Robotic Microphone Sensing: Design of A Robotic Platform and Algorithms for Adaptive Control of Sensing Parameters, Fall 2009

Converge & Diverge

Sideways

Shift

Rotate

Approach & Retreat

2X