Autonomous Ground Vehicles – the next wave of commercial field robotic systems will be in agriculture and we have a number of projects with agriculture government institutes and industry collaborators. Students would be interested in autonomous UGV and UAV applications for conducting farm wide mapping, sampling and harvesting. Students would have an interest in working with the robots in field trials on farms in NSW and VIC. Areas of research include perception system for fruit classification, vegetation mapping, and animal tracking; control for robot motion, harvesting and spraying; and multi-robot systems for farm wide applications. The projects call for students in mechatronics and aeronautical engineering. Contact: Salah Sukkarieh

Autonomous Perception - With recent advances in robotic engineering and the science of precision agriculture, we are at the start of a revolution in agricultural technology. Automated data gathering and picture compilation systems provide significant efficiency gains to large and complex processes that are common in many industry sectors. Gains are achieved by increasing the accuracy and timeliness of information. The defence sector first saw the benefit of “Tactical Picture Compilation” many years ago, and the mining sector is currently adopting “Mine Picture Compilation” systems. Australia is now perfectly positioned for the development of automated Agricultural Picture Compilation. You will conduct research on the flow of information throughout entire operations such as orchards or farms, working on new algorithms for autonomous robotic sensing and information fusion. Spanning precision agriculture and robotics, you will establish new methods to acquire and fuse in-ground data (e.g. natural gamma and conductivity for soil health) with above ground data (e.g. hyperspectral imagery for tree and crop health), and monitoring produce (e.g. counting fruit or tracking livestock with range sensors), to create a real time digital picture of the orchard or farm. This research will help the Australian agricultural sector to establish the farm of the future. Top-up scholarships may be available for exceptional candidates already holding an APA. Contact:

Autonomous Ground Vehicles

Advanced Perception - Localisation and mapping is a key enabling technology for autonomy, and robust solutions lead to new applications and opportunities for the industrial deployment of robotics. Technology such as GPS has demonstrated this, by paving the way for autonomous land, air and sea-surface vehicles in agriculture, mining and defence. Due to the reliance on weak signals from satellites, GPS is not sufficiently robust in many environments (e.g. Urban canyons, deep pit and underground mines, etc), so there is a race to create reliable GPS-less localisation systems to exploit the opportunities they will create. The well-established approach in robotics is called simultaneous localisation and mapping (SLAM), but dependable solutions for very large-scale environments do not yet exist. A key problem is determining whether the robot is exploring new territory or is revisiting a previously mapped region, since it may arrive at an old location by a new and circuitous route. A promising approach using vision sensors is to treat place recognition as an image database retrieval problem. You will research the use of recent advances in image database retrieval, combined with state-of-the-art feature extraction and “visual word” dictionaries to improve the scale, speed and accuracy of vision-based place recognition. Further research will augment this with a graph data-structure to represent spatial connectivity, to generate place recognition hypotheses. Success in this research will advance the agenda for greater industrial deployment of mobile robots. Contact:

Multi-Robot Perception - For large scale operations covering a wide or distributed geographic domain, the quality of the information that is available to decision makers determines the efficiency of the operation or probability of success. Examples include expansive mine sites or farms, in which distributed tasks are carried out by individuals and teams to achieve common medium and long term objectives. On a shorter timescale, the same is true for disaster response, such as bush-fires, freak weather events, search and rescue or urban terrorism. Teams of robots can provide a rich composite source of information by covering larger regions than a single agent can and combining different types of sensor data as appropriate. You will research new algorithms and methods to fuse information from different robotic platforms and sensors to provide a common operating picture over large scale environments, to assist human decision making. A key area of research will be on predicting the future value of information that each team member could provide, to enable system wide autonomy for the team. Solutions will cater for multiple sensor modalities (e.g. thermal, vision, LIDAR, RADAR,environmental) and heterogeneous teams with different capabilities and perspectives (e.g. ground and air vehicles). Contact:

Perception Using Radar - Unmanned ground vehicles (UGVs) that are designed for industrial deployment require perception systems that are robust to adverse weather conditions. RADAR is the most capable sensing technology in harsh environments, with rain, fog, snow, dust and smoke, because at that wavelength, the signal is able to penetrate significantly further than visible light or laser based technology. Due to the relatively low availability of off-the-shelf RADAR sensing technology, more research needs to be done in UGV RADAR perception. You will be working at the forefront of radar algorithm development for UGV perception. The algorithms will focus on matching the physics of how the data is produced by the sensor and environment to support the key autonomous capabilities in agriculture, mining and defence. Contact:

Intelligent Vehicles & Safety Systems

Intelligent Vehicles and Safety Systems Group

The next few years will see a significant transformation of the automotive and transportation industry. The introduction of communication between vehicles and infrastructure and new sophisticated perception technology will enable cooperative operations dramatically changing the way we will design and use transportation systems. The introduction of different level of autonomy in urban roads will require breakthroughs in position, perception and communication.
The first stage will see the introduction of cooperative ITS application with the introduction of Vehicle to Vehicle (V2V) and Vehicle to infrastructure (V2I) technology. This will enable cooperative safety applications demanding positioning with a required accuracy not available in most areas. We will expect vehicles retrofitted with standard communication capabilities and a number of internal and external perception sensors. Although the sensors will be part of a particular vehicle functionality, their information will also be available for other applications.
The second stage will include the deployment of fully autonomous vehicles and the interaction with all existing road users: vehicles, bicycles and pedestrian. Truly intelligent vehicles must be capable of developing a high-level understanding of the traffic scene in order to drive safely and effectively amongst human drivers as well as other intelligent vehicles. Acquiring the necessary information is a challenging task since contextual variables are not directly observable and must be inferred from low-level data.
The projects below present some of the areas that will be addressed as PhD / Masters research projects. Contact

Cooperative Situation Awareness - Situation awareness involves the sensing of the local environment, understanding the situation and predicting the future state. For Intelligent Transportation Systems, situational awareness is essential for detecting unsafe behaviours and for allowing the introduction of autonomous systems into complex traffic scenarios. Cooperative situation awareness involves the sharing of information between local groups of vehicles to improve the understanding of the current scenario. By fusing the information received over communication networks it is possible for vehicles to have a better understanding of the risks, allowing safer operation and reducing accidents. Furthermore, existing fleet of vehicles will be able to share high level perception capabilities provided by smart autonomous vehicles operating in the proximity area. This project will research into multimodal perception and efficient communication of vital information between vehicles. Contact or

Vehicle Localisation and Sensor Fusion - One of the fundamental requirements of Cooperative Intelligent Transport Systems (C-ITS) is navigation ie knowing the position, velocity and attitude of the vehicle at all times with a figure representing the uncertainty. A probabilistic estimate of these properties can be obtained by fusing information from GPS/GLONASS receivers and local reference sensors such as gyroscopes, accelerometers, wheel encoders and other vehicle sensors. C-ITS systems will require different levels of accuracy for different applications. The accuracy required has been categorized according to the following levels:

  • Road level: (5 meters, 1-5 sec). Required to know which road the vehicle is navigating and which other vehicles are in proximity. This enables proximity awareness capabilities in automotive and other industry applications
  • Lane-level: (1.5 meter, 1 sec). Enables applications relating lane level vehicle interactions
  • Where in the lane: (< 1 meter , 0.1 sec). Enables applications for warning of crossing lanes etc.

An additional specification can be stated for high performance C-ITS and autonomous vehicles:

  • lateral / longitudinal road location: ( < 0.1 meters, 0.1 sec ). Enable high speed interaction / autonomous applications.

Current GNSS solutions can only provide information for basic applications such as GPS direction assistance. The position estimate to satisfy other requirements can be improved by incorporating information from a map or environment model. A key area of research in ITS is the fusion of available sensor information to provide a high integrity position and state estimate. An additional important constraint is the use of low cost sensors to increase the uptake of this technology into the general vehicle population. This information is essential in modern vehicle safety and automation applications. Contact or

Collision prediction and avoidance under uncertainty - All positioning systems feature uncertainty in the reported position. Any system for predicting and avoiding collisions between intelligent vehicles must account for this uncertainty. Probabilistic models of vehicle conflicts present a promising avenue for tackling this problem. This work can be further extended to examine trustworthiness of state information shared by other vehicles, and develop a method of obtaining maximum value from the information given its trustworthiness. Trustworthiness may be influenced by sensing uncertainty, sensor failure or deliberate falsehood. Detecting and acting on this will be vital for any future intelligent vehicle fleet. Contact or

Urban Street Mapping - This project addresses the evaluation of global maps to assist vehicle navigation. The main goal is the determination of complete set of feature to achieve high integrity navigation with the level of accuracy for autonomous operations.

Road/Lane Maps: The first stage will look at mapping of all vehicle / cycle / ramps lanes. The project will investigate algorithm to be used with high quality perception sensors. It will further extend these techniques to be used with low cost sensors for rapid deployment in all roads around Australia. It will also research into other type of infrastructure that could be sensed to improve the integrity of localisation. The aim of this infrastructure will be reliable detection by multimodal sensing to improve the integrity under all weather conditions. This will also look at the combination of radar signal processing and beacon design to be able to localise under all weather conditions.

3D salient Features: Road infrastructure will play an important role for localisation, specially in motorways, country roads and open areas. Urban areas will present different challenges since high dense traffic could prevent detection of visual artificial landmarks at road level. Nevertheless, the combination of building and infrastructure usually presents very high density salient features that remain static with time. This project will develop methods to obtain navigation maps that can be downloaded in an efficient manner through V2I infrastructure to be used by map matching algorithm to register current mobile position. Contact or

Data Analytics for Naturalistic Driving - Large scale trials of Intelligent Transportation Systems will generate vast amounts of data that is reflective of real world scenarios rather than artificially constrained by experimental design. Methods will need to be developed to capture, warehouse and analyse this information. Vehicles are equipped with sensors which can directly measure the vehicle state. This data allows algorithms to generate hypotheses about the behaviour of the driver and the vehicle. However, measurements of the vehicle state only form indirect observations of driver behaviour. It is difficult to directly measure human actions without biasing their behaviour. Utilising non-invasive measurement technologies such as estimating head orientation, gaze tracking and driver posture provides a means for collecting direct observations of the driver. This expressive data can be used by learning algorithms to model driver behaviour in natural driving conditions. Contact , or

Driver Intent - Human drivers possess a natural ability to perceive our surroundings from the point of view of other people and reason about their intentions. To drive safely, we must develop a high-level understanding of the situation. The estimation of driver intent is a key research area in the domain of Intelligent Transportation Systems. Mathematical tools are required to take a range of vehicle sensor information and attempt to match the reasoning carried out by a human driver. The resulting estimates will form the basis for collision risk assessment and probabilistic decision-making. Contact or

Transport Safety Evaluation and Monitoring - As Intelligent Transportation Systems are developed and deployed there will be a need for evaluating the impact upon safety that they bring. High quality, high frequency data will be available from a variety of sources (in-vehicle sensors, roadside infrastructure) that have not been analysed before. By developing techniques for analysing this data, transport safety can be evaluated and resources targeted at the highest risk areas. This analysis will also be necessary to justify the investments in ITS as they are deployed, and to monitor their effect and ensure that they are achieving their aim of increasing safety. Contact or

Pedestrian / bicycle detection / intent prediction - Autonomous systems can be designed to operate reliably in highly structured environments. Although road-rules provide structure to traffic situations, roads frequently contain unpredictable conditions and scenarios which can violate assumptions. Pedestrians and cyclists are an important example, particularly in densely populated urban environments. To minimise the risk autonomous systems impose on these vulnerable road users, systems must be designed specifically to consider them. The goal is to identify pedestrians, model their trajectories and predict their intent. To achieve these goals, research in this area will draw from the fields of perception, computer vision, filtering and machine learning. Contact or

Urban Perception - For the first time in history, more than half of the world’s population are living in urban centres, so the development of autonomous robotic technology in urban environments represents a significant opportunity. Solving the existing technical barriers to deployment will enable a suite of new applications in diverse areas such as urban disaster response (police, fire, ambulance), surveillance and security, intelligent transportation and autonomous passenger vehicles. The primary challenges are in robust perception (addressed by this topic) and localisation (addressed by “Robust Multi-Modal Localisation and Mapping). You will conduct research in urban perception, addressing the difficulties of modelling a complex, cluttered three dimensional, dynamic environment. Areas of study include recognising components of the scene at different scales (e.g suburb, carpark, car, wheel), tracking complex sequences of motion and interaction and reasoning about change on different temporal scales. Furthermore, research will be done to understand the relationship between sensor information from different sensors and perspective, to enable active robot control for optimal picture compilation. Contact:

Intelligent Systems

Convex Optimization for Nonlinear System Identification - System Identification is the extraction of a compact mathematical model of a dynamical system from possibly incomplete or noisy records of experimental data. This presents many challenges not found in classical statistics or machine learning, in which the central problem is approximation of a static function. For example, even in the simplest case of linear systems with Gaussian noise, system identification presents a highly nonlinear and nonconvex optimization problem, which are generally impossible to solve in practice. Recently, Dr Manchester and colleagues at MIT have discovered a new convexification of the problem of nonlinear system identification. This means that efficient (polynomial time) algorithms can be developed which are guaranteed to find a good model. However, many difficult theoretical questions remain, and many exciting applications await. Current applications considered by collaborators include high-fidelity modelling of RF circuits and model reduction for computational fluid dynamics. This project would require very strong programming ability and knowledge of (or at least interest in) optimization algorithms, machine learning, or convex analysis. Top-up scholarships may be available for exceptional candidates already holding an APA. Contact:

Design and Control of Dynamic Walking Robots - Dynamic walkers are a new breed of biped robots designed to be efficient and agile, far more so than well known robots such as Asimo or Big Dog. In a dynamic walker, the physical mechanism itself creates most of the motion and control actions are only used for small corrective actions. The extreme nonlinearity and the complex target motions mean that standard linear methods for feedback control and stability analysis cannot be applied directly. This project will involve designing and building a series of modular biped robots, and using them to develop new techniques in advanced nonlinear control. Outside robotics, dynamic walking principles are being applied in biomechanics, active prosthetics, animation, and computer games. Strong mathematical skills and a thorough understanding of dynamics and control are essential for this project. Mechanical/mechatronic design experience is also desirable. Top-up scholarships may be available for exceptional candidates already holding an APA. Contact:

Experiment Design and Information Gathering - For an adaptive autonomous system to act optimally in its environment, it is sometimes necessary for it to perform "probing" actions, which are intended to illicit an informative "response". For example, for the model-predictive control system running a large petrochemical plant, the dynamical model used is critical to performance but often loses accuracy. Can the control system "probe" the various chemical reaction processes in such a way that it can re-estimate parameters without disturbing the production? Another example is optimal sampling of an environmental field (e.g. underwater temperatures, or air pollution) by teams of autonomous robots. How should their motions be planned so it is most informative about the underlying physical process? These are very difficult but highly important problems. This project will focus on developing motion planning and control techniques for these and other "information gathering" problems. The petrochemical plant application involves collaboration with Royal Institute of Technology, Stockholm. Strong programming and mathematical skills are essential. Top-up scholarships may be available for exceptional candidates already holding an APA. Contact:

Nonlinear System Identification in Computational Neuroscience - Reverse engineering the brain is one of the US National Academy of Engineering's "Grand Challenges for Engineering" for the twenty-first century. The brain is an extraordinarily complex interconnected dynamical system. What (static) machine learning and statistics have done for genomics, we aim to do for the brain. This project is based on an international collaboration with Harvard and MIT developing arrays of silicon nanowire sensors that can probe and measure from individual neurons in a rat's brain. Given the ability to experiment with individual neurons, how can one rapidly develop a high-fidelity simulation model that captures its behaviour? This is a system identification task, but the extremely complex dynamics of neurons means it is beyond the currently developed methods, so a new approach is needed. An interest in neuroscience or biology and strong mathematical skills are essential for this project. Top-up scholarships may be available for exceptional candidates already holding an APA. Contact:

Marine Robotics

The ACFR, as operator of a major national Autonomous Underwater Vehicle (AUV) Facility, conducts AUV-based surveys at sites around Australia and overseas in collaboration with institutions including the University of Tasmania, the University of Western Australia, the University of New South Wales, James Cook University, AIMS, CSIRO, the Woods Hole Oceanographic Institution, the University of Rhode Island and the University of Nottingham. These AUV surveys are designed to collect high-resolution stereo imagery and oceanographic data to support studies in the fields of engineering science, ecology, biology, geoscience, archaeology and industrial applications. One of the major challenges with this program is managing, searching through and visualizing the resulting data streams. Our recent research has focused on generating high-fidelity, three-dimensional models of the seafloor; precisely matching survey locations across years to allow scientists to understand variability in these environments; and identifying patterns in the data that facilitate automated classification of the resulting image sets. Providing precise navigation and high-resolution imagery lends itself to novel methods for data discovery and visualization. As a result, we have a strong focus on methods for interacting with and discovering patterns in the data using machine learning techniques. We also have a strong record of engagement with end users in a variety of domains interested in understanding marine environments. We have a number of opportunities for postgraduate students to join our team and make significant research contributions in the area of marine robotics. Prospective students will work closely with research staff in the group to develop an engaging study topic. Recent examples of studies facilitated by our research can be found here. Contact:

Multi-Robot Systems

Coordinated Information Gathering with Decentralised Multi-Robot Systems - We are interested in decentralised information gathering using heterogeneous teams of robots with complementary capabilities. Such capabilities might include sensor payload such lasers and thermal cameras, and mobility capabilities such as ground robots and flying robots. The advantage of heterogeneous teams is that they can perform tasks that are difficult or impossible for single robots working independently. Searching, tracking, object classification, and mapping of geometric or chemical features in cluttered outdoor environments are best addressed collaboratively because heterogeneous teams can provide multiple simultaneous viewpoints to handle occlusions and moving targets, multiple sensing modalities to reduce classification uncertainty, and redundancy to improve system robustness. Research topics generally focus on decentralised planning and control algorithms that consider realistic assumptions about communication as well as various forms of uncertainty. Contact:

Security & Defence

Cross-Spectral Imaging and Classification of Concealed Targets in Cluttered Environments - It is known that by operating over different regions of the EM spectrum we can leverage positive aspects available at each frequency. For example, at low frequencies signals can penetrate solid materials, whilst at high (optical) frequencies we can achieve good imaging resolution. A radar system that could operate like a software defined radio to adapt its waveform generation abilities over a wide frequency band to optimise image quality would be of great utility in many domains such as non-destructive testing, geo-physics, ground-penetrating/wall-penetrating radar. The main research question to be answered in this project would be to determine whether multi-spectral techniques can be used to improve penetrative imaging capability by iteratively resolving layers of obstructing material that become translucent as frequency decreases. This project would suit a graduate with a strong background or interest in physics and mechatronics. Contact: David Johnson

Electromagnetic Simulation of Complex Dynamic Environments to Assist Cognitive Active Perception - An accurate sensor model is required to better understand the interaction of high resolution millimetre-wave radar within complex dynamic environments, particularly in instances where both the radar-platform and target (for example a pedestrian, animal, vehicle or vegetation is in motion). Using the wealth of sensors available to the ACFR and access to High Performance Computing facilities within the University, the intention for this project is to develop a GPU based physical-optics model capable of operation in real-time ‘in the cloud’. Using this model, a Bayesian framework may be constructed to identify potential manoeuvres to seek out areas of increased novelty/information while minimising noise. This project would suit a graduate with a strong background or interest in physics, mechatronics and/or computer-science. Contact: David Johnson

Measures to Increase Trust and Reliability of Robotic Systems in the Presence of Adversity - In the defence and security context it is likely that robotic systems must at some point operate in the presence of a technically adept adversary. The need for stealth and constrained electro-magnetic spectrum will necessitate passive means of localisation and an increased level of autonomy. The potential for operating while damaged will also provide an interesting challenge. This project will investigate novel low-profile sensing techniques and other methods for evading detection. This project is most suited to mechatronics graduates. Contact: David Johnson

Active Perception at the System Level, Including Human-(on-the-loop)-Unmanned Teaming - This project will investigate the use of multi-modal sensor fusion for situational awareness of extended environments. The project will focus on improving the efficiency of air, ground and human assets to solve high-level tasks. This is likely to involve an examination of predictive intent and communication between the system’s own components in addition to that of external entities. This project is likely to suit graduates with a strong interest in mechatronics, computer-science and human-robot-interaction. Contact: David Johnson


Integration and synchronisation of a pair of 77GHz imaging radars to produce bistatic images of targets, with particular emphasis on foreign object detection on runways. Contact