Centre for Intelligent Systems Research

CISR Research Seminar Series - 2010

  CISR presentation
blue star Professional development
yellow star Keynote lecture
purple star External presentation
red star No presentation

These research seminars were held in the Geelong Haptics Lab (KC1.017)

Date Time Presenter Presentation/topic
Thursday 8th July 3pm Zhenying Guan (Vivienne) 3-D Hydrodynamic Analysis of Biomimetic Robot Fish 


This presentation will discuss a three-dimensional (3D) computational fluid dynamic simulation of a biomimetic robot fish by Fluent 6.3.26. User-defined function (UDF) is used to define the movement of the robot fish and Dynamic Mesh is used to mimic the fish swimming in water. Hydrodynamic analysis has been done in this paper too. The aim of this study is to get comparative data about hydrodynamic properties of those guidelines to improve the design, remote control and flexibility of the underwater robot fish.

Thursday 24th June 3pm Abbas Khosravi Developing Hybrid Modelling and Simulation Methodologies for Complex Systems  


Accurate point prediction is fundamental for the reliable and efficient operational planning. Despite this importance, accurate prediction of system targets is problematic and far remote. The uncertainty prevailing in operation of complex systems significantly degrades prediction performance of models. Besides, there is no indication of the reliability of predicted targets. The objective of this study is to construct prediction intervals for future output/targets of complex system. The delta technique is applied for constructing prediction intervals for outcomes of neural network models. Some statistical measures are developed for quantitative and comprehensive evaluation of prediction intervals. According to these measures, a new cost function is designed for shortening the length of prediction intervals without compromising their coverage probability. Simulated annealing is used for minimization of this cost function and adjustment of neural network parameters. The quality of prediction intervals also closely depends on the neural network structure and its training hyperparameters. A genetic algorithm-based method is developed that automates the neural network model selection and optimal adjustment of the training hyperparameters. The optimization is carried out through minimization of a prediction interval-based cost function, which depends on the length and coverage probability of constructed prediction intervals. Experiments conducted using the real datasets demonstrate the suitability of the proposed methods for improving the quality of prediction intervals in terms of their length and coverage probability.

Thursday 3rd June 3pm Wael Abdelrahman Data-Based Dynamic Haptic Interaction Model With Deformable 3D Objects 


3D deformable objects often exhibit nonlinear deformation behaviour due to the object material, topology, and the external applied force. To generate high fidelity physically based simulation, the finite element method (FEM) was introduced, but latency issues made it unsuitable for nonlinear real-time haptic applications. Data-based techniques were introduced into haptics to handle complex simulation scenarios where parametric (explicit) models cannot be applied. Modelling haptic interaction with empirical data has challenges such as collection methods, large data amount and difficulty of generalizing the collected data to different scenarios. Without planning the data collection, the resulting model will be large and cannot be easily generalized. This paper introduces new techniques of planning the data-based modelling of haptic interaction with deformable models simulation. These techniques enable faster data collection and better utilization of the collected data via discretizing the 3D object and the external force space.

Tuesday 25th May 2pm Prof. Mahmoud Tarokh External presentation  Intelligent Robotic Systems with Application to Robotic Person Following  


In this talk I will give an overview of the research projects being conducted at the Intelligent Machines and Systems (IMS) Lab of the Computer Science Department, San Diego State University. The focus of the research in this lab is intelligent robotic systems with emphasis on mobile robots operating in rough and unstructured environments. The projects fall into five areas of robotic person following, rover navigation, robot path planning, robotic helicopter and robotic security systems. I will briefly review the current IMS lab projects and then give a more detailed description of the robotic person following project. The system, now in its third generation, addresses the problem of tracking and following a person by a robot both indoors and outdoors with significant light variation such as moving from sunny areas into shades, and difficult walking maneuvers such as jumping to a side or making sharp turns. These are achieved through certain image processing technique, and by real-time control of camera parameters affecting the light exposure. Several strategies are formulated and integrated to quickly recognize and segment the person from the environment during difficult walking maneuvers. In addition to the camera exposure control, the camera pan and tilt and the robot speed and steering are controlled using the characteristics of the detected person's image. The system has been implemented on a Segway robotic platform that has been equipped with a camera, an onboard computer, loud speakers, game-pad, etc. I will conclude the talk by showing a video of the robotic person following under a variety of conditions.

Thursday 29th April 3pm Ali Ghanbari Force Pattern Characterization of C. elegans in Motion  


C. elegans is a worm that could be mutated to have different muscle arms, which may generate distinct force patterns when the worm moves. In this talk, an integrated system employing both a novel PDMS device and a visual feedback from the device will be reported. The silicone elastomer-based PDMS device consists of arrays of pillars, which forms open channels for the worm to move in and bend the pillars in contact. Enabled by a single vision sensor (CCD/CMOS camera), the computer vision system is able to transform the forces generated by C. elegans, through detecting the deflection of the pillars with sub-pixel accuracy. The experimental results demonstrate that the current vision-based force sensing system is capable of performing robust force measurements at a full 30 Hz with a 1.52 muN resolution. The framework has the potential to significantly facilitate the study on the relationship between muscle arms and force patterns of C. elegans in motion, and thus gives a better understanding of muscle arms development and modeling.

Thursday 15th April 4pm Dr Reza Hoseinnezhad External presentation  Random Finite Set Approach to Multi-Object Estimation and Tracking in Computer Vision 


In many computer vision applications the number of objects and their states are unknown and stochastically vary with time. This so-called multi-object estimation problem arises in a host of applications areas including aerospace, defense, field robotics, communications, environmental, biomedical research, and are becoming more important with the proliferation of sensing technologies. Multi-object filtering generalises classical paradigms such as Bayesian/Kalman filtering to multi-object systems, and is a challenging problem both in theory and practice. The last decade has witnessed exciting developments with the introduction of random finite set theory to multi-object filtering. The random finite set framework has led to the development of the Probability Hypothesis Density (PHD) filters, which attracted substantial interests from academia and industry alike. The PHD filters have been used by BP in oil pipeline tracking, by NATO in the 'Bold Avenger' defense exercise and Lockheed Martin in the US space fence program. This seminar presents an introduction to multi-object filtering with random finite set and recent advances in filtering from video data.

Tuesday 30th March 3pm Hamid Abdi Strategies for Fault Tolerant Force for Human-Robot Cooperation 


When a robotic manipulator and a human are cooperatively providing a force for a task or a robot is assisting a human for doing a task through cooperation, then the cooperation is more reliable if it is fault tolerant. For specific tasks the fault tolerance promotes into the safety of the whole system. The fault is assumed to occur within the manipulator's joints. To achieve the fault tolerance it is essential to map the effects of the faulty joint/s into the manipulator's healthy joints' torque space and the human force by considering the human force limitation. The objective is to optimally maintain the force within the human robot cooperation. Three human-robot cooperation strategies for fault tolerant force are formulated based on perturbation model and the optimal joint torque reconfigurations for compensating the force jump due to the joint failure/s of the manipulators are introduced. Then the introduced frameworks are validated through different fault scenarios within a joint of a PUMA560 in a human-robot cooperation scenario. It is indicated that the cooperation strategies not only result into a fault tolerant force when it is possible, but also they are consistent to each other as it is expected. .


Deakin University acknowledges the traditional land owners of present campus sites.

27th March 2012