Robotics & Sound Localization II

On this page you can see the first full-scale trials of the sound localization software on the mobile robot.

The following video shows the very first test a colleage and I did with sound source localization and laser-based object detection.

Video: first test of sound source localizer on mobile robot

With its 360° laserscanner, the robot can detect objects which are not part of its environment. These "dynamic" objects can be boxes, chairs, tables ... or people.

The idea was that the robot should only turn towards sound sources which could be associated with a laser-based dynamic object. The assumption being that these sound-producing objects could be humans talking and would therefore be potentially interesting for the robot.

After this first, promising trial, we decided to make a more elaborate experiment in controlled conditions. There were six sound sources (loudspeakers), three of which were put on cardboard boxes, so they could be classified as dynamic objects by the laser scanner. Noise was played randomly through one of the speakers and the robot had to turn to that source if it was associated with a dynamic object (if the speaker was on top of a cardboard box). This can be seen in the video below.

NOTE: The video is quite long (10'15") and gets boring pretty fast ... also note that only five loudspeakers are visible in the video. The sixth is to the right of the foremost source (labeled "6"), outside of the picture frame.

Video: experiment with sound source localization on mobile robot

To summarize, the robot turned in about 60% of the cases to the right source. That might seem like not much, but it could be shown that that was mainly caused by the room acoustics.The room was much larger than any other room in which the sound source localization system had been tested before and had a lot of reverberation. This confused the sound localizer. But still this experiment can be seen as a success, as it showed that simply adding a sensor modality (laser-based object recognition) helped to correct the poor sound source localization performance. The complete details and results of the experiment can be found in my dissertation.

The robot we used was initially designed for robotic soccer in the RoboCup Middle Size League. But it - along with its teammates - was retired somewhere in 2006, because they were too slow compared to the competition. Fortunately, in 2006, RoboCup@Home was created, for which the platform seemed perfectly adequate. "Caesar" went on to become RoboCup@Home world champion in 2006 and 2007, vice world champion in 2008 and winner of the RoboCup@Home German Open in 2007 and 2008!

Below you will find some technical information about the robotic base (as it was used in the experiment above) along with a picture of "Caesar" (as of 2009). Along with the superstructure, it also got an additional computer, based on a more modern dual-core processor. The picture and technical specifications were taken from the AllemaniACs website.

Robot "Caesar", technical details
AllemaniACs RoboCup@Home Robot
Platform size
40 cm x 40 cm x 60 cm
approx. 80 kg
Motor power
2.5 kW
Top speed
12 km/h
Onboard computers
2 Pentium III 933 MHz
Distance sensor
360° laserscanner
CCD on pan-tilt unit
IEEE 802.11a/b/g WLAN