ASST - 1 Using audio to direct visual attention in a humanoid robot
CSE Project - Dr. Waleed Kadous (waleed@cse.unsw.edu.au) and Professor Claude Sammut (claude@cse.unsw.edu.au)
The Centre of Autonomous Systems group is building a humanoid robot to explore human-robot interaction. Previous research conducted has built a static 4 microphone audition system that allows it to guess where the location and type of the sound (speech vs non-speech).
The scholar will integrate a stereo pan-tilt camera into the process, allowing the robot to identify faces and track different speakers around the room and trying to appropriately track a conversation between two people.
The scholar will be given freedom to explore approaches to the problem under the guidance and assistance of the supervisors and research support staff. New techniques for handling confusing information from different sensors, and a form of interaction that lays the ground work for future research. An expected outcome of the research is real-time software that allows the robot to use both vision and audition to track people in a room.
For further information, please contact the supervisor or consider the following:
Centre for Autonomous Systems http://www.cas.edu.au
ASST - 2 Opening doors: A step to robots at home
CSE Project - Dr. Waleed Kadous (waleed@cse.unsw.edu.au) and Professor Claude Sammut (claude@cse.unsw.edu.au)
Robocup @ Home is a competition to explore approaches to make it easier for robots to be used in the home. One task that must be accomplished in home use is to reliably open doors, a deceptively difficult task.
The scholar will build a system that uses a Pioneer (see below) equipped with a camera and a robotic arm to open a specially built scale door, including depressing the handle, pulling the door open (whether it swings inwards or outwards) and driving through it.
The scholar will be given freedom to explore approaches to the problem under the guidance and assistance of the supervisors and research support staff.
Accomplishing this task reliably with a robot would be a significant step forward in robotic hand-eye coordination, as well as practically important. A potential outcome of the research is control software that would allow a robot to go through a specially built door that could be generalised to a full size robot and door.
For further information, please contact the supervisor or consider the following:
http://www.robocupathome.org/
http://www.activrobots.com/ROBOTS/p2dx.html
ASST - 3 Playing scissors, rock, paper with a humanoid
CSE Project - Dr. Waleed Kadous (waleed@cse.unsw.edu.au) and Professor Claude Sammut (claude@cse.unsw.edu.au)
The aim of this project is to use speech and gestures to program a humanoid robot how to play scissors, rock, paper. While apparently trivial, the task has aspects that touch on many deep and difficult research topics in vision, multi-modal interaction, speech recognition and virtual character scripting.
This robot has been constructed as part of a project in the ARC Centre of Excellence for Autonomous Systems. It includes two arms, two hands, microphones and a stereo camera.
The scholar will work with a small team and will be assisted on the mechanical and low-level aspects of the design. Successful integration of the diverse technologies discussed here to build a working system is an open research topic. Potential outcomes of this research include techniques for integrating diverse capabilities into an effective interaction with a humanoid and software that would control the robot to play scissors rock paper.
For further information, please contact the supervisor or consider the following:
http://www.cse.unsw.edu.au/~claude
http://www.cse.unsw.edu.au/~waleed
http://www.barretttechnology.com/robot/products/hand/handfram.htm
ASST - 4 Automatically identifying disaster victims
CSE Project - Dr. Waleed Kadous (waleed@cse.unsw.edu.au) and Professor Claude Sammut (claude@cse.unsw.edu.au)
RoboCup Rescue is an international competition that requires a team of robots to search through a mock disaster site and automatically identify any victims. The victims are represented as mannequins who emit heat and CO2, make sounds, and wave their arms around.
UNSW has built a robotic software and hardware platform called CASTER that can be used to navigate such arenas and manually detect victims. It is equipped with normal cameras, thermal cameras and a microphone. What would be of great benefit is a system that would automatically detect victims and/or suggest possible victims to the vehicle operator.
The project will require the scholar to develop software to process all these inputs - including image processing and fuse the information to help pick out likely victims.
Building such a system that integrated information reliably from multiple sensors would be a step forward research-wise. In practice, such techniques could also help in technologies that would save people's lives. Potential outcomes of the research include new techniques for combining multiple sensors to detect disaster victims and a software implementation integrated into the existing Robocup rescue infrastructure.
For further information, please contact the supervisor or consider the following:
2005 entry description (2006 entry in preparation):
http://www.araa.asn.au/acra/acra2005/papers/kadous.pdf
Rescue Web site:
http://rescue.web.cse.unsw.edu.au/
ASST - 5 Learning to See - an L2 project
NICTA Project - Dr Bernhard Hengst (Bernhard.hengst@nicta.com.au)
The aim of this research is to program a machine to learn to recognise objects using vision. The research involves experimenting with hierarchical layered learning algorithms allowing a machine to form invariant representation of objects that move in its visual field. The nature of the work is software development - machine learning algorithms that are inspired by neuroscience and able to form object concepts. This project is related to project WASABI, a cognitive robotics project that forms part of the boarder research by NICTA's SMLKA program into the possibility for machines to learn by themselves and to be taught multiple skills. Forming invariant object concepts is an open problem in AI. Any success will make a significant contribution in AI. An expected outcome of the research is that the machine should be able to learn to recognise several different objects from various viewpoints with a high degree of accuracy.
For further information, please contact the supervisor or consider the following:
http://www.cse.unsw.edu.au/db/thesis/topicinfo/BH06.html
whitepaper at http://www.numenta.com/
http://nicta.com.au/director/research/programs/smlka.cfm
[Top of Page]
