ODAS Multichannel audio-interface for our PR2 robot
Bachelor- or Master-Thesis at group TAMS
Motivation
While audition traditionally has played only a secondary role for robots,
it is increasingly seen as an important input channel
that provides additional information not readily obtainable from
vision or motion sensors.
In particular, advanced signal processing algorithms and deep neural networks
are proposed to extract and localize sound events and sources
from multi-channel multi-mircophone data.
Also, real-time speaker-independent speech recognition is within reach
of deep neural network approaches by now.
Potential application scenarios are plentiful, including
- 2D and 3D spatial localization of (static) sound sources
- 2D and 3D spatial tracking of (moving) sound sources
- robot ego-noise suppression
- speaker tracking and identification
- robot environment detection and localization from reverbation modeling
- etc.
ODAS - Open embedded audition system
We have bought the 16-channel open-source audio processing board
developed as part of the ODASA project,
https://github.com/introlab/odas,
together with a set of microphones.
However, the hardware has never been installed on any of our robots,
and the basic setup and test of the board and accompanying software
will be the first part of the thesis.
Given the current Corona-restrictions, this part of the thesis
could be done at home.
In the second step, we will plan and decide on the actual
hardware installation on our robot(s),
selecting suitable locations for the microphones on our
PR2 two-arm mobile service robot.
At this stage, we could also consider additional boards for installation
on our other robots or in our (fixed) lab setups.
In the third step, to be discussed when starting the thesis,
we will select one or two application scenarios that form the
main part of the thesis, with the research topic selected
from among the items listed above.
Thesis Goals:
- getting advanced embedded system (hardware+software) to work
- working audio input on a mobile service robot
- implementing a robot audition demonstration on a robot
- collect ideas and experiment data leading to a publication (MSc-thesis only)
Requirements
- as always, interest in the topic area
- interest in digital audio and signal processing (e.g. SP lecture)
- basic knowledge about electronics and hardware (e.g. Embedded Systems course)
- interest in advanced robotics (e.g. Introduction to Robotics lecture)
Contact
- Norman Hendrich, Raum F-314, Tel.: 42883-2399
- Andreas Mäder, Raum F-317, Tel.: 42883-2502
- this thesis could be done in a cooperation with the SP signal processing group
(Prof. Timo Gerkmann)
