Engineering, Technology and Robotics

Project Members

Chief Investigators	Janet Wiles, Gordon Wyeth, Mandyam Srinivasan
Engineer	David Ball
Collaborators	Michael Milford, Tien Luu, Allen Cheung, Peter Stratton, Francois Windels, Chris Nolan
PhD Student	Gavin Taylor
Research Assistants	Gavin Taylor (2009) Daniel Clarke (co supervised) (2009) Jack Valmadre (co supervised) (2009-2010)
Research Experience Students	Scott Heath (2008-2010), Ryan Wong (2008-2009), Chris Lehnert (2008-2009), Gavin Taylor (2008-2009) Nick Calver (2009-2010), Jessica Wrigley (2009-2010) Hilton Bristow (2009-2010), Ezra Zigenbine (2009-2010) Kieran Wynn (2009-2010)
Undergraduate Thesis Students	Scott Heath (2009), Marcel Giermanski (2009) Ryan Wong (2009), Chris Lehnert (2009) Nick Calver (2010), Jessica Wrigley (2010) Hilton Bristow (2010), YeHua Hsu (Mike) (2010) Shao-Kuang Fang (William) (2010), Daniel Clarke (2010)

Engineering, Technology and Robotics

David Ball

(July 2008 – November 2010)

Introduction

My primary role has been to enable new science through collaboration where I provide technology, systems and mechatronics engineering leadership to the Thinking Systems team. In this way I work with the researchers to evaluate and provide appropriate engineering planning and solutions. I also am working on new mechatronics solutions for mobile robots primarily focussed on building a new rat animat robot.

iRat (Intelligent Rat Animat Technology) (with Scott Heath)

The iRat (intelligent Rat animat Technology) is a rat animat robot designed for robotic and neuroscience teams as a tool for studies in navigation, embodied cognition, and neuroscience research (Ball, Heath, Wyeth, and Wiles, 2010, ACRA). The rat animat has capabilities comparable to the popular standard Pioneer DX robots but is an order of magnitude smaller in size and weight. The robot’s volume is approximately 0.08m² with a mass of 0.5kg and has visual, proximity, and odometry sensors, a differential drive, a 1 GHz x86 computer, and LCD navigation pad interface. To facilitate the value of the platform to a broader range of researchers, the robot uses the Player-Stage framework, and C/C++, Python, and MATLAB APIs have been tested in real time. Two studies of neural simulation for robot navigation have confirmed the rat animat’s capabilities.

An industrial design company, Infinity Design, is currently styling the iRat to give a professional look suitable for commercialisation. They are also designing a dock to allow the iRat to recharge autonomously.

 

 

Figure 1. (left) iRat prototype. (right) iRat by Infinity Design

 

In 2010, three undergraduate thesis projects were based around the iRat.

•      Development of a whisker system to sense proximity and texture - Nick Calver.

•      Development of a visual obstacle avoidance system - Daniel Clarke.

•      Initial work on developing legs for a future quadruped version – Jessica Wrigley.

RatSLAM on the iRat (with Scott Heath, Michael Milford)

This study ran RatSLAM on the iRat to map a figure of eight environment which demonstrated the capabilities of the iRat (Ball, Heath, Milford, Wyeth, Wiles, 2010, Artificial Life).

 

 

Figure 2. (left) RatSLAM experience map, (right) ‘non conjuctive grid cells’ cells generated from the pose cells.

Spike Time Robotics (with Peter Stratton, Christopher Nolan)

In this study a spiking network controls the iRat in real time (Wiles, Ball, Heath, Nolan and Stratton, 2010, ICONIP). The study demonstrates how the neural controller directs the rat animat’s movement towards temporal stimuli of the appropriate frequency using an approach based on Braitenberg Vehicles. The circuit responds robustly (after four cycles) when first detecting a light pulsing at 1 Hz, and rapidly (after one-to-three cycles) when primed by recent experiences with the same frequency. This study is the first to demonstrate a biologically-inspired spike-based robot that is both robust and rapid in detecting and responding to temporal dynamics in the environment. It provides the basis for further studies of biologically-inspired spike-based robotics.

 

 

Figure 3 - iRat location and spiking network output while tracking a 1Hz flashing stimulus. (top left) iRat showing two light sensors, their respective resonant circuits and crossed connections to the wheels. (top middle) Tracking camera view. (top right) Tracking data showing three trials, first with the robot directly facing the flashing stimulus, then rotated approximately 45° to the left and right. (bottom) Left and right sensor responses (see text for details).

Blind Bayes (with Allen Cheung, Michael Milford)

Real rodents are able to maintain localisation even when visually deprived. This study is investigating the iRat’s ability to maintain localisation with only sensory information about the walls in close proximity. (Cheung, Ball, Milford, Wyeth and Wiles. In preparation.)

Figure 4. iRat global pose tracked overhead (blue line), integrated wheel odometry (red line) and sensory wall measurements (green crosses). The goal is to maintain accurate global pose by resetting the iRat’s internal map position when colliding with the wall.

RatSLAM – MATLAB version (with Michael Milford, Scott Heath)

There was motivation for a lightweight version of RatSLAM:

•      that could be released publically,

•      that SLAM researchers could use on their own datasets, and

•      a tool to understand how the RatSLAM algorithm works.

MATLAB was chosen for a version that could be publically released due to its inbuilt functions for matrix operations, graphing, and loading video. The result was a lightweight implementation that clearly shows the major functionality of RatSLAM including: View Templates, Pose Cells and Experience Map. The core parts of the RatSLAM have also been written in C and are loaded as a DLL which allows much faster processing. A module is also available to process overhead images to track global pose. The code supports as an input:

•      processing a combination of video and wheel odometry information from files (such a recorded from a robot) , or

•      processing video and performing visual odometry (such as recorded from a moving platform like a car), or

•      real time closed loop control of a robot using Player-Stage.

The code for the offline versions is online at http://ratslam.itee.uq.edu.au/ along with two datasets: a partial St Lucia suburb video and a partial Axon level 5 video. The real time version will be released soon.

Work has begun on integrating a GPU version of RatSLAM into the MATLAB code base. Due to the parallel nature of the GPU this will be much faster.

 

A technical report is in preparation.

 

Figure 5. - St Lucia (left) and Axons level 5 (right) using visual odometry using a MATLB version of RatSLAM.

Omni-directional drive robot platform (with Chris Lehnert)

This project designed a robot research platform with high mobility that can traverse typical office environments and has decent onboard computational resources. (Ball, Lehnert, and Wyeth. 2010. ICRA.) A novel spherical drive mechanism has been designed and tested. The advantage of this spherical drive mechanism is continuous contact with the ground plane to reduce vibration and isotropic rotational characteristics that facilitate improved traversal properties. In 2010 two undergraduate projects worked on turning the drive system into an autonomous robot and adding the ability to autonomously recharge.

 

 

Figure 6. (left) Prototype omnidirectional drive robot.  (right) A CAD drawing of the spherical mechanism.

Telerobot for RatSLAM (with Scott Heath)

A telerobot, is a telepresence device that allows a user to interact with a robot over the internet. There are few active public telerobots on the internet. In general, direct control of the robot is not appropriate because of the communication lag and the ratio of users to robots. The RatSLAM navigation system allows the user to indirectly control the robot by setting navigation goals. A generic streaming and interaction module has been written for the Apache web server which communicates with the robot and an Adobe Flash client in the user’s browser. In the current implementation it allows the user to set navigation goals while the robot’s camera, internal (RatSLAM experience) map, etc is streamed.

 

 

Figure 7. (left) The telerobot architecture showing the connection between the pioneer, the webserver and the user’s client.  (right) The flash client interface showing the robot’s camera view, map, battery charge level, etc. Users can click on the map to add navigation goals (shown as a green dot).

Insect environment replication - Multi-monitor world (with Tien Luu, Allen Cheung, Gavin Taylor)

Neuroscientists investigate honeybee flight behaviour using visual stimulus. (Luu, Cheung, Ball, and Srinivasan. Journal of Experimental Biology. - corrections.) Initial investigations demonstrated that moving an object on an LCD past a Bee resulted in changes to the angle of the Bee’s abdomen. This investigation prompted the development of a setup that would allow a virtual world to be rendered on monitors surrounding the honeybee. My solution was a multithreaded C++ DirectX application that could render a viewport on each monitor. Objects can be rendered relative to the world coordinates or the camera. This system works with up to 6 monitors @ 1920 x 1200 pixels @ 60 Hz. A Python interface was added so that the biologists could adjust major settings such as the number of windows, the textures and planes, the position and velocity of the camera and the platform. A LEGO platform was constructed that can raise and lower on command from the python script.

 

 

Figure 8. (left) Four monitors with moving scene surrounding the Bee tether and LEGO platform. (right) The LEGO platform from another angle.

Rodent Electrophysiology

Digital Wireless Neural Telemetry Phase One (with Ryan Wong, Francois Windels)

Typically, electrophysiologists tether the rodent to their neural recording equipment. A wireless neural recording system would allow for: larger and more complex environments, social interaction studies, and outdoor recordings. Existing wireless rodent recording systems can be classified by the number of channels, recording rate, bit precision, continuous streaming versus spike only, size, weight, battery life, and cost.

A prototype has been developed that allows for continuous digital recording from 16 channels @ 20kHz @ 8 bits, weighs less than 50 grams and lasts over one hour. The prototype has recorded spikes from a freely behaving rodent with comparable results to a tethered system.

 

 

Figure 9. (left) Digital wireless neural recording system architecture. (right) Data recorded using the digital wireless neural recording system from a freely behaving rodent showing a neuron spike.

 

The wireless module is enabling a novel experiment to be setup which is currently in progress. Scott Heath has begun work on a Rat Tracking program suitable for the novel environment.

Digital Wireless Neural Telemetry Phase Two (with Tara Hamilton, Bala Thanigaivelan)

Work has begun on designing the next phase module uses a custom IC to replace the analogue components which make up over 50% of the area and power. The IC is a challenge in itself due to the high gain requirements. The first iteration of the chip has some problems with stability but an updated version will be resubmitted soon.