Aerial Robotics & Aerospace Systems
The ACFR conducts some of the world's largest and most innovative R&D in aerial robotics and autonomous aerospace systems.
We tackle a broad range of research and application areas, collectively aimed at increasing the capability of aerial robots and aerospace systems to support national and international government, industry and civilian needs.
Program Leader: Salah Sukkarieh
Research Leads: Mitchell Bryson, Ali Goktogan, Robert Fitch, Nicholas Lawrance, Alistair Reid.
Research Students: Tariq AbuHashim, Calvin Hung, Jason Gan, Prasad Hemakumara, Zhe Xu, Abdallah Kassir, Angela Lui, Jen Jen Chung, Ken Ho, Joe Nguyen, Chanyeol Yoo, Daniel Wilson, Layth Awin.
Previous Researchers and Students: Todd Lupton, Kwang Jin Yang, Andrea Abel, Shaun Brown, Airlie Chapman, Dave Cole, Michael George, Jason Held, Jong-hyuk Kim, Suresh Kumar, Lee Ling Ong, Hai Ngoc Pham, Matthew Ridley, Paul Thompson, Jake Toh, Ben Upcroft.
International Visitors: Dae-Yeon Won (KAIST), Francesco Delle Fave (Southampton), Patrick Kranzinger (Stuttgart), Franzi Ullrich (Switz), Prof Mark Campbell (Cornell), Dr Terresa Vidal (LAAS), Christopher Mei (INRIA), Florian Renk (Stuttgart), Jeurgen Shultz (Stuttgart), Il-Kyun Kim (LAAS), Richard Giroux (EDTS), Pedro Pines (Zaragoza).
Research Engineers: Muhammad Esa Attia, Jeremy Randle.
Collaborators and Partners
Government: Australian Research Council; Australian Space Research Program (DIISR); QLD BioSecurity, Australian Plague Locust Commission; Department of Agriculture, Fisheries, and Forestry; Land and Water Australia.
Industry: QANTAS Airways; BAE Systems; Meat and Livestock Australia; ST Aerospace; US Air Force; US Office of Naval Research, Ministry of Defence UK.
Academia: University of Seville; DLR, Berkley University; Biological Sciences University of Sydney.
Mawson the Planetary Rover
Mawson is a planetary rover that is both a teaching aid for high school students studying science and technology, and a research platform at the University of Sydney.
Teachers - download the latest study guide that lets you use Mawson as part of your teaching.
An operational model of a Mars Exploration Rover provides the opportunity to develop and test new algorithms in autonomous perception, control and machine learning.
The aim of the project is to develop intelligent approaches to search and classification by promoting higher order decision making, and in particular to deal with opportunistic science. The goals of the project include:
- The creation of data fusion algorithms for terrain estimation, environment representation and localisation;
- Developing and demonstrating an integrated guidance and control sensor architecture including; high-order probabilistic decision making linked to the science mission; and
- Developing and implementing middleware architecture for human-robot interaction.
The project is funded by the Australian Space Research Program as the "Pathways to Space" and is in collaboration with the Australian Centre for Astrobiology, Powerhouse Museum, and Cisco, to also include educating students on Space.
Long Endurance Multi-UAV Systems
The ACFR is part of the recently awarded European FP7-PEOPLE-2011-IRSES research exchange program: "Multi-UAV Cooperation for Long Endurance Applications"
The joint exchange program will be with DLR (Germany) and CATEC-USE (Spain) and has the objective of conducting research and development in technologies that will help to create Long Endurance Multi-UAV applications for the future. This will include:
- Research on control algorithms for the extension of the endurance applied to autonomous aerial robots or UAVs using wind energy.
- Research on control and estimation algorithms for all-weather UAV operation that includes estimation and planning techniques to avoid weather hazards and advance control techniques to overcome extreme weather conditions.
- Research on new fully distributed methods for real-time cooperation of entities involving fault adaptive reconfiguration of the trajectories for long endurance applications
Terrestrial Weed Detection, Mapping and Analysis
A number of ongoing projects with Meat and Livestock Australia to look at the impact of autonomous remote sensing from a UAV would have on farming practices.
During 2009 and 2010 flight trials were conducted in the Mitchell Downs grasslands near Julia Creek, West Queensland. During the trials high-resolution airborne imagery was collected from a fixed-wing UAV. The project demonstrated that classified, geo-referenced maps of the area could be built with quick turnaround time and terrestrial weeds could be classified from native vegetation.
Current research focuses on automatic detection, classification and quantification of woody weeds. Recent advances in perception and machine learning will be applied to improve discrimination between weed species.
Aquatic Weed Detection and Eradication
For the 2007/08 Defeating the Weed Menace program (DWM), we proposed to build and test a prototype robotic aircraft and surveillance system to detect aquatic weeds in inaccessible habitats.
Defeating the Weed Menace - Cost-Effective Surveillance of Emergent Aquatic Weeds Using Robotic Aircraft
Currently we are investigating the use of fixed-wing and rotary-wing UAVs pairs for large scale mapping and precision classification of woody weed infestations in a project with Meat and Livestock Australia.
We also have a project with the Australian Weeds Research Council to see how this technology can be applied to detecting cacti in the Flinders Ranges.
The focus of all this research is to develop an intelligent surveillance system which uses various forms of machine learning algorithms for the detection and classification of vegetation, as well as information driven control actions for improving classification and mapping.
We currently have an ARC Linkage Grant in collaboration with colleagues at the Biological Sciences in the University of Sydney and the Australian Plague Locust Commission.
The objective is to develop an aerial surveillance system to track individuals within the locust swarm and deliver data to the biologists to improve swarm modelling.
Initially the focus of the RUAV platform is to simply fly over the swarm and collect imagery data. Specific locusts are tagged as to make identification easier for the tracking algorithms. Eventually the goal of the of the project is to develop an intelligent surveillance system which will choose which part of the swarm to observe in order to improve data collection completion as well as accuracy.
Two types of surveillance systems will be developed: imagery based, and radar based.
Temporal/Spatial Correlations for Terrain Reconstruction
The aim of this research is to develop algorithms for real-time smoothing and mapping using vision and inertial sensors over rich, dense, large-scale and unstructured environments, and to develop intelligent control policies which aim at maximizing accuracy and consistency of the reconstructed terrain information.
The research will also focus on overlaying this reconstruction with track information of moving features so that potential temporal correlations between moving and stationary features can be identified.
The aim of this research to develop energy based control laws for a fixed-wing UAV so that it can actively capture energy from the atmosphere to prolong endurance.
The research work is based on understanding the mechanics of energy capture with a flying platform. That is, how an aircraft gains and loses energy during flight and what type of conditions are required to capture energy from the atmosphere. This also includes understanding of natural energy sources such as thermals and wind shear.
The second part focuses on designing planners and controllers to capture energy autonomously. This includes characterising the energy available and generating paths that take advantage of energy sources. The ultimate goal is to develop a controller to manage platform energy whilst considering an external mission goal (such as a simple point-to-point flight with insufficient starting energy).
Human-UAV Knowledge Sharing
The aim of this research was to determine the appropriate architecture which allowed for information sharing between human ground operators and UAV systems, as well as how tasking can be achieved in a multi-objective optimisation scenario.
The research focussed on determining the value of information and how this can be used as the underlying metric to determine whether certain actions or human knowledge would be informative and whether such information should be incorporated into the mission.
Furthermore, in any complex scenario there are multiple objectives that need to be weighed up against one another and techniques in multi-criterion decision making were employed.
Multi-UAV Active SLAM
This project extended on the work in Decentralised Data Fusion and Control to deal with the problem of mapping terrain in a GPS-less environment.
The focus of the research was on Active Simultaneous Localisation and Mapping for a single UAV plaform and how this could be extended to the multi-platform scenario.
The difficulty lies in how multiple platforms can negotiate over their action space which has to take into consideration map quality, localisation quality as well as the ability for a platform to help improve the localisation accuracy of another platform.
Cooperative UAV Systems
In 2007 we concluded a world first demonstration of 2 UAVs cooperating in real time in flight with the objective of maximising information of ground based features.
The project drew on over 10 years of research activity in the area of Decentralised Data Fusion and Control.
Each UAV was equipped with visual sensors and had the sole objective of mapping ground based features using bearing only tracking. The guidance and control module had the task of determining the appropriate sensor-feature assignment along with the trajectory to the feature, as well as negotiating with the other UAV.
Decentralised Air/Ground Surveillance System
The aim of this project was to extend upon the previous work of Air-based Decentralised Data Fusion network to also include ground platforms, and hence dealing with the difficult problem of fusing information coming from various visual aspects.
Furthermore, we moved away from the simple point feature scenario to also include extended or non-point features.
There were three research thrusts: using machine learning techniques to classify natural features and develop a probabilistic inference machine; develop appropriate data fusion algorithms for the fusion of non-point feature based observations and models; and extend this development to provide for decentralised sensing and data fusion capabilities.
Decentralised Air Surveillance System
This project marked our first venture into aerial robotics and UAV systems. The objective was simple - build and demonstrate a decentralised air surveillance system - however the task of doing so was very complex.
The platforms (Brumbies) were modified and built in-house at the ACFR. They were fitted out with various imagery sensors, along with a complex suite of processing capability to deal with image acquisition, detection, and data fusion. A complex communication network was developed to allow the various sensors to communicate with one another within the platform and between one another.
The platforms would fly along fixed flight paths and acquire imagery of the terrain below. This information was shared between the sensors on the platforms and a map of features was constructed in real time, to form a decentralised air sensor network. We flew single, dual and triple UAV configurations and compared the performance to a centralised sensor network system.