Research Work done by Albert Diosi

The research I am allowed to talk about comes from these sources:

• My PhD studies at Monash Unversity in Melbourne, Australia.
• 10 moths Post-Doc at the Lagadic Group of IRISA/INRIA Rennes.
• Self sponsored research.

Brief description of the research in reverse chronological order:

Self Sponsored Research

Mappino

Here are photos of the first Mappino prototype (Mappino v0.1) and the first range scan matching results including scans before and after the matching:

The work done on Mappino is not really scientific research as the contributions have more of an engineering and educational character. The questions addressed in the research were:

• How simple can one make scan matching?
• How simple can a robot be while still able to perform localization and mapping?

Mappino is equipped with Sharp IR range sensors and capable of taking 360 degree range scans. Mappino's 8bit micro-controller (programmable with the Arduino IDE) is capable of matching the scans in about 150ms.

The work regarding Mappino has been presented at the 2011 Robotics in Education Conference. More about Mappino including the presented paper can be read here.

Mappino is an open source robot. However for those robotics enthusiasts who would like to have a Mappino, but don't want to build one, the start-up company ABC Robotics, s.r.o. is preparing a version they can buy.

Post Doc in the Lagadic Group of IRISA/INRIA Rennes

In 2006-2007 I have spend about 10 months at the Lagadic Group at IRISA/INRIA Rennes in France as a Post Doc doing research on autonomous car navigation using visual servoing. I was working with Sinisa Segvic, Anthony Remazeilles and Francois Chaumette. As my involvement with the project included getting Cycab running under autonomous control using visual servoing and doing heaps of experiments, I got heaps of practical help from Fabien Spindler who is an excellent research engineer.

Photos of the cyber taxi CyCab driving autonomously under a building and then doing a relaxing drive in the park:-):

A video showing how the vision system on CyCab worked:

A video showing CyCab going at 1.8m/s and parking itself into its own garage (after it was shown what to do):

PhD Research at the Intelligent Robotics Research Centre of Monash University

Thanks to being awarded the MGS and MIPRS scholarships, during the years of 2001-2005 I was doing my PhD studies at the Intelligent Robotics Research Centre (IRRC) of Monash University in Melbourne, Australia. I could not have hoped for better supervision as my supervisor was A/Prof Lindsay Kleeman and my associate supervisor was A/Prof Andy Russell. I submitted my thesis in a bit over 3.5 years which then won the Douglas Lampard Electrical Engineering Research Prize and Medal for 2006 (citation).

During my PhD studies, I have focussed on the following topics:

• Laser gyro calibration and gyro-odometry fusion.
• Modelling random and systematic errors in 2D laser range finder readings. Extracting primitives as lines and right angle corners in polar coordinates and estimating their random and systematic errors.
• Fusing advanced sonar (Lindsay Kleeman's special sonar measuring accurately range and bearing to corners, edges and planes) readings with laser readings and feeding the results into Simultaneous Localization and Mapping (SLAM).
• For fun and as an application of the research, together with Geoffrey Taylor we got the robot SLAMBot to do interactive SLAM where we show the robot around telling it which room is which while the robot builds a map using SLAM. Then the user could send the robot to any of the rooms. Geoffrey Taylor did the voice interfacing to IBM's Via Voice, extraction of legs from scans and segmenting free space into rooms. The concept of Interactive SLAM comes from Patric Jensfelt's thesis, but to our best knowledge our's was the first complete working implementation of the concept. There had been before us an implementation without SLAM - just plain mapping, and without voice recognition.
• During the Interactive SLAM work, a solution to the problem of registering past robot poses with corresponding laser scans to feature based SLAM maps was also developed. This is important for being able to generate accurate occupancy grid maps even if the feature based SLAM map deforms as it improves itself.
• Fast and simple 2D laser scan matching in the laser's own polar coordinate frame called Polar Scan Matching (PSM). PSM was released under GPL and BSD, and thanks to Ivan Dryanovski it even got a wrapper for ROS now.

I was lucky during my PhD, as I for my research I could use my superversor's robot SLAMBot, which worked reliably and had really good sensors. On the following images, you can see SLAMBot (blue boxes are the advanced sonars, yellow sensor is the laser, little black box in front of screen is a fiber optic gyroscope), a features base SLAM map build by SLAMBot while having laser scan overlaid on the map (grid: 10x10m) and the laser calibration tool I have built from salvaged printer parts:-).:

A video showing SLAMBot while performing Interactive SLAM:

To read more, visit my website from the times of my PhD studies at IRRC.

Copyright (c) Albert Diosi, 2011

Last changed: 01/12/2011