Technology
Robots that are fast and low cost
Research
Overview
Algorithms & Architecture
This program aims to create advanced algorithms and techniques to allow computer vision to be run in real-time on robotic systems deployed in large-scale real-world applications, using distributed sensing and computation resources and to provide efficient and unified software platforms for real-time robot visual SLAM algorithms and techniques development and employment in real-world environment. There are three research projects (AA1, AA2, and AA3) under this program, each addressing a significant aspect of robotic vision research, development, and applications. AA1 (VOS) will provide a common, distributed computational platform that takes advantage of distributed sensing and computational capabilities to solve large complex robotic problems. AA2 (ACRV-SLAM) is focused on the development and integration of robot vision algorithms in robust vision, real-time vision and semantic vision areas, into a single SLAM-centred robot navigation framework. The framework will be demonstrated in real-world robot applications including AUV (autonomous underwater vehicle), UAV (unmanned aerial vehicle, or flying robot), and ground-based autonomous vehicles. AA3 (SIMUL) aims to provide a photorealistic graphics simulation environment to facilitate and accelerate the development of advanced robot vision algorithms and systems.
People
Tom Drummond
Monash University Node Leader, Chief Investigator
Monash University
Peter Corke
Centre Director, Chief Investigator, QUT Node Leader, Project Leader (Manipulation Demonstrator)
QUT
Hongdong Li
Chief Investigator
Australian National University
Richard Hartley
Chief Investigator
Australian National University
Ian Reid
Deputy Director, University of Adelaide Node Leader, Chief Investigator, Project Leader (Scene Understanding)
University of Adelaide
Matthew Dunbabin
Chief Investigator
Queensland University of Technology
Niko Sünderhauf
Chief Investigator, Project Leader (Robotic Vision Evaluation & Benchmarking)
Queensland University of Technology
Viorela Ila
Former Research Fellow
University of Sydney, Australia
Laurent Kneip
Research Affiliate
ShanghaiTech University, China
Trung Pham
Former Research Fellow
Nvidia, United States
Vincent Lui
PhD Graduate, Former Research Fellow, Monash University
Sentient Vision Systems, Australia
Feras Dayoub
Former Research Fellow
Queensland University of Technology, Australia
Yasir Latif
Research Fellow
University of Adelaide
William Chamberlain
Former PhD researcher
Boeing Aerostructures, Australia
Steve Martin
Research Engineer
Queensland University of Technology
Mina Henein
PhD researcher
Australian National University
Andrew Spek
Research Assistant
Monash University
Sean McMahon
PhD Graduate, Queensland University of Technology
Brisbane, Australia
John Skinner
PhD Researcher
Queensland University of Technology
Jun Zhang
PhD researcher
Australian National University
Projects
Robotic Vision Evaluation and Benchmarking
2018 onwards
Niko Sünderhauf,Feras Dayoub,David Hall,John Skinner,Rohan Smith,Ben Talbot
This project will develop new standardised benchmark tasks, evaluation metrics, and a new robotic vision challenge competition that can be one of the legacies of the Centre.With the new tasks, metrics, and benchmarking competitions, we aspire to recreate for robotic vision the positive effects competitions such as ILSVRC or COCO had for the advances of deep learning and computer vision. We will run our new competition annually in conjunction with a major computer vision and robotics conference.The project will combine the variety and complexity of real-world data with the flexibility of synthetic graphics and physics engines. We will put a focus on evaluating algorithms developed in other Centre projects on real robot hardware under realistic conditions, and provide valuable feedback on the robustness of the evaluated algorithms and approaches.The project aims to overcome the current lack of meaningful standardised evaluation protocols and benchmarks which is a significant roadblock for the evolution of robotic vision, and impedes reproducible and comparable research in robotic vision worldwide.
Previous project: AA2: ACRV SLAM Framework
- 2017
Viorela Ila,Richard Hartley,Hongdong Li,Tom Drummond,Ian Reid,Laurent Kneip,Matthew Dunbabin,Yasir Latif,Vincent Lui,Feras Dayoub,Mina Henein,Andrew Spek,Jun Zhang,Sean McMahon
This project will develop novel SLAM algorithms which can perform in challenging environments (large-scale, dynamic, dense, non-rigid). ACRV-SLAM is a common framework that integrates efficient implementations of the proposed algorithms with the goal to facilitate their distribution to the robotics community and the industrial partners, and to produce high quality demonstrators.
Previous Project: AA3: Computer graphics simulation for robotic vision
- 2017
Peter Corke,John Skinner,Steve Martin,Niko Sünderhauf,Trung Pham
The performance of a robotic vision system depends on the initial state of the robot and the world it perceives as well as the lighting conditions, unforseen distractors (transient moving objects) and unrepeatable sensor noise. A consequence is that no robotic vision experiment can ever be repeated and the performance of different algorithms cannot be rigorously and quantitatively compared. For machine learning applications a critical bottleneck is the limited amount of real world image data that can be captured and labeled for both training and testing purposes.This project investigates the potential of photo-realistic graphical simulation based on state-of-the-art game-engine technology to address both these challenges.
Previous project: AA1: VOS-distributed robotic vision in the cloud and embedded mobile platforms
Sept 2016 - 2017
Tom Drummond,Peter Corke,Vincent Lui,William Chamberlain,Steve Martin
The goal of AA1 is to create a Vision Operating System that provides a framework for bringing together multiple sensing and computational resources to solve complex robotic vision problems. This will enable robots to make use of external sensing resources (e.g. CCTV cameras in the environment, or sensors mounted on other robots) as well as computation resources (either attached to those sensors, or provided as a large computing resource within the network). This kind of framework enables novel solutions to complex problems in which the various resources are combined collaboratively to solve complex localisation, navigation, understanding and planning problems.
Australian Centre for Robotic Vision
2 George Street Brisbane, 4001
+61 7 3138 7549