Postgraduate Projects

Contents

Preamble

Qualifications

Online Publications

Projects:

Humanoid Projects

Real Time Sonar

Motion Extraction from Video Sequences: Model Selection (Scholarship available)

Robotic odour sensing

Tactile sensing for robots

Mechatronic devices

Robotic mechanisms

Video Plane Robotic Swarms

Autonomous Robot Navigation in Very Rough Terrain

Autonomous Watercraft Navigation

Stereo Vision Navigation of an Underwater Vehicle

Humanoid/Human Cooperative Hand/Eye Coordination

Range and Colour Intensity Data Based Scene Analysis

Preamble

The following list of projects is not complete but does give prospective postgraduate students some concrete directions on what research projects academic staff are offering. Other projects can be negotiated with potential supervisors.

Qualifications

You will need a recognised degree equivalent to a Monash honours degree in Engineering or Computer Science appropriate for the research planned. Please refer to http://www.ecse.monash.edu.au/prosp/postgrad.html for more details on requirements, scholarships and applications.

Online Publications

Some online publications are available for prospective students to review previous work in the IRRC:

Humanoid Robotics Projects

The following academic staff are interested in supervising postgraduate projects in this area: Professor Ray Jarvis, Associate Professor Andy Russell, Associate Professor Lindsay Kleeman and Dr David Suter. The IRRC is developing human-like robots that have similar form to human beings in terms of mechanical and sensor structures. Work on the design of Plasticman and Metalman is well underway. There are many PhD and Research Masters projects available as part of the humanoid ongoing research.

Why Humanoid?

Humanoid robots have a sensor and mechanical structure similar to human beings - stereo vision, stereo ears, two arms etc. The motivation for using a similar design for robotics is many fold:

manmade devices are tailored for human sensors and actuations. A robot designed to operate these can operate similar to a human.

humanoid robots can learn and be taught based on self knowledge that humans possess.

human form is considered the most versatile of all animals and is the most evolved - so evolution has naturally selected this form. A robot can exploit the outputs of evolution by mimicking important aspects of human form.

humans interact naturally and quickly with a human form. We all know how to relate to other humans and therefore this can be exploited in the human machine interface by making the machine humanoid.

Research Aims and Directions

A humanoid robot projects is an excellent vehicle for studying real time machine/environment interaction. This project aims to investigate and implement humanoid robots that can achieve the following goals

accept simple natural commands from a human operator - such as by voice, hand signals or even facial expressions

convey information back via sound, facial expression, bodily motion, eye movement.

actively learn tasks for manipulating the environment, with a particular emphasis on assisting elderly or handicapped individuals. Familiar examples are voice directed feeding, learning manipulation tasks such as fetching and pouring a drink and turning on a tap.

Although these aims appear simple for human beings, they require considerable sensing capability, physical actuation and underlying intelligence. These issues are being investigated by this project.

Sensing Capability

High speed vision and tracking, sound recognition and location, odour/taste sensing and tactile sensing are all considered important for this project. The emphasise is on real time performance, since high speed interaction with the environment is paramount. Distributed and embedded DSP based systems are being investigated. The IRRC has considerable expertise in these areas.

Physical Actuation

Humanoid robots have many degrees of freedom that need actuation. Not only are speed and accuracy important but safety of movement is a key factor. Active compliance techniques are being considered in the implementation in order that the humanoid cannot injure or endanger humans interacting.

Intelligence

In order that the humanoid can perform tasks in different and dynamically changing environments, directed or autonomous learning needs to be performed. Online fast learning schemes are to be investigated to achieve these goals. The assistance of an operator will be encouraged in the learning process since not only will this be readily available in the intended applications, but also great benefit can be obtained since the operator should have insight into the problems faced by a system with similar form.

 

Real Time Sonar Sensing for Simultaneous Map Building and Localisation

Supervisor: Associate Professor Lindsay Kleeman

Air-based sonar sensor has been developed for the purpose of simultaneous map building and localisation (SLAM) in a real time processing context on a mobile robot. The sonar sensing hardware and mobile platform are already available and the project will focus on the algorithms and analysis required to solve the SLAM problem in real time. Other aspects to the project could include software engineering, data association problems, map maintenance and consistency and approximations required for real time operation of estimation algorithms. See Video Clips for work in progress.

 

Robotic odour sensing:

Supervisor: A/Prof R. Andrew Russell. Odour sensing is a relatively unexplored area for robotics. However, there are applications for chemically sensitive robots as, for example, replacements for sniffer dogs. Based on biological models volatile chemicals could also be used to help robots navigate and to organise robot swarms. We are investigating biologically inspired and other techniques for locating the source of chemical plumes, the use of chemical trails as an aid to robot navigation, dispersed chemicals as a means of organising robot swarms, and the design of electronic noses for both mobile and humanoid robots.

 

Tactile sensing for robots:

Supervisor: A/Prof R. Andrew Russell. Touch is a very immediate and unambiguous source of information about our close surroundings. During manipulation operations and in 'difficult' conditions of poor visibility it can provide very useful information for a robot system. We are investigating novel forms of tactile sensing including touch sensory skins and tactile whiskers together with algorithms for analysing and using tactile information.

 

Mechatronic devices:

Supervisor: A/Prof R. Andrew Russell. The essential ingredients of any robotic system are sensors, computation and actuators. Appropriate choices of sensors and actuators can simplify a robotic system or may even be the difference between its success and failure. We are investigating many novel actuators including those based on shape memory alloy, electrorheological fluids, magnetic fluids and the piezoelectic effect as well as a wide range of sensors for measuring quantities of importance for robotic systems.

 

Robotic mechanisms:

Supervisor: A/Prof R. Andrew Russell. All research in robotics should be validated using real robotic systems. Thus, all of the sensors, actuators and algorithms that we develop are tested by incorporating them into a mobile robot platform, humanoid robot or fixed manipulator/ gripper system. We have extensive experience of building legged, wheeled and tracked land vehicles, submersibles and flying robots as well as robotic grippers and complete humanoid robots.

 

Video Plane Robotic Swarms

Supervisor: Professor Ray Jarvis

Being able to control many small robots using the paths traced out on a horizontal video display screen, driven by computer graphics output, permits many swarm control structures to be tested and consequent emergent collective behaviours to be studied. This project concerns the study of robotic swarm behaviour orchestrated by simple control structures both hierarchial and distributed in nature.

 

Autonomous Robot Navigation in Very Rough Terrain

Supervisor: Professor Ray Jarvis

This project concerns the deployment of multiple onboard sensors, predominantly visual, to support the simultaneous mapping and localisation paradigm for autonomous mobile robot navigation using natural landmarks in very rough terrain environments. Appropriate vehicles are available for this research as are suitable very rough terrains for experimental studies.

 

Autonomous Watercraft Navigation

Supervisor: Professor Ray Jarvis

This project is concerned with combining information from GPS systems, maritime radar, active and passive vision sensors and flux-gate compass gear to navigate a

watercraft in a variety of realistic situations in an efficient and collision-free manner. A variety of water vehicles are available for this work and bay water access is

only some twenty minutes away from the campus.

 

Stereo Vision Navigation of an Underwater Vehicle

Supervisor: Professor Ray Jarvis

This project involves developing a stereo vision based navigation system for an underwater vehicle. Both a stereo camera fitted submersible and a 30 foot ( 9.14

metre ) diameter, 7 foot deep ( 2.13 metre ) tank are available for this work. Open waters would be used as the project matures, several watercraft suitable for this

stage being available.

 

Humanoid/Human Cooperative Hand/Eye Coordination

Supervisor: Professor Ray Jarvis

As part of our group's humanoid robotics program this project concerns the development of interactive vision based cooperation between a human and a humanoid

for table top task scenarios. Both human gesture interpretation and human action mimicry through visual processes are central to this study.

 

Range and Colour Intensity Data Based Scene Analysis

Supervisor: Professor Ray Jarvis

This project builds upon previous research and existing experimental apparatus and concerns the processing of range and colour imagery data extracted from scenes

of jumbled objects to determine, where possible, individual object identity, pose and position in a form suitable for robotic manipulation. Both geometric and free

form objects, not all known beforehand, may be involved.

 

Copyright Monash University - All rights reserved Disclaimer.