Mechatronic engineering research internships

Supervisors: Professor Stefan Williams, Dr. Lachlan Toohey, Dr. Gideon Billlings, Dr. Jackson Shields

Eligibility: Control systems knowledge (e.g. AMME3500/AMME5520). C++, Matlab, Python recommended.  An interest in marine robotic systems, machine learning or system characterisation are desirable.

Project Description:

The marine robotics group at the Australian Centre for Robotics (ACFR) undertakes fundamental and applied research in a variety of areas related to the development and deployment of marine autonomous systems. We conduct AUV-based surveys at sites around Australia and overseas. 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.

We have up to two positions available through the VRI program.  Specific projects will be designed in collaboration with students to complement their strengths and research interests, but may include one or more of the following:

• This project will focus on characterisation of AUV system dynamics to allow our AUV systems operate close to the seafloor over complex terrain, aiming to hold a consistent altitude of 2 meters. Optimising the AUV’s control system for both maneuverability, power efficiency and obstacle avoidance is necessary for operating in these conditions. 
• Visual scene reconstruction is important for scenarios in which AUVs need to make decisions about survey targets online or when interacting with the environment.  This project will explore the use of a embedded devices to accelerate an image feature encoding network as a backbone for visual methods, such as Simultaneous Localization and Mapping, semantic segmentation, and object detection, that will run onboard underwater vehicles.
• Fast and reliable quantitative estimates of marine environmental health are needed for scientific studies, design and management of protected areas, and regulatory compliance of industrial activity in the ocean. Australia is collecting seafloor images at increasing rates but expert annotations are not keeping up, meaning that typical machine learning approaches struggle. This project will explore self-supervised techniques that use large amounts of unlabelled data to enhance performance.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr. Stewart Worrall, Dr. Julie Stephany Berrio Perez and Dr. Mao Shan

Eligibility: Programming Skills: Basics in Computer Vision and ML

Project Description:

The main objective of this project is to enhance data augmentation for semantic segmentation in images, specifically for autonomous driving applications. The project involves applying cutting-edge machine learning techniques to modify the appearance of locally annotated images into various domains. One of the core tasks is image translation, which entails training a machine learning model to transform an image from one domain to another. For instance, this could involve converting a daytime image to a night-time or rainy condition image.

By performing such data augmentation, the project aims to create a more robust and diverse dataset, which can improve the performance and generalization capabilities of semantic segmentation models used in autonomous driving scenarios. This augmentation process will expose the models to different environmental conditions, preparing them to handle various real-world situations effectively.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr. Stewart Worrall, Dr. Julie Stephany Berrio Perez and Dr. Mao Shan

Eligibility: Programming Skills: Python or C++, basic knowledge of robotics

Project Description:

The project aims to advance digital twins for Autonomous Vehicles (AVs) by harnessing combined capabilities, facilities, and equipment. Students will be able to gather data, process it, and create an accurate model of the environment. Data collection includes various perception sensor data, such as lidar, inertial units, cameras, and GPS localisation.

Experienced researchers specialising in robotics and sensor fusion will mentor the students. Their guidance will assist in developing scalable digital twin models that represent the physical features of the local environment. Applying state-of-the-art techniques will be central to overcoming perception, localisation, and 3D reconstruction challenges.

In addition to the technical aspects, the project will incorporate virtual reality (VR) to effectively communicate the research's potential and showcase the project's results. Ultimately, the models, algorithms, and tools developed during the internship will be publicly available. View current visualisation.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr Donald Dansereau, Jack Naylor

Eligibility: Strong programming and mathematics skills. Basic understanding of Robot Operating System (ROS) is desirable.  Basic understanding of optics, electronics and computer vision techniques is recommended. Basic knowledge of probability and information theory is beneficial, but not required.

Project Description:

Robotic platforms deployed in the wild are subject to conditions that are sometimes unexpected, and may affect their sensor readings - for example, magnetometers operating in industrial environments, autonomous underwater vehicles localising around moving seagrass, drones clipping tree branches during close operation, and multipath interference on GPS sensors in urban environments. Traditionally, these readings are seen as outliers which are excluded from state estimation and downstream processing.

In this project, we will overturn this assumption. By recognising the relationship between sensor readings, their probabilistic and information-based interpretations, and differences between sensor modalities, this project will develop metareasoning-inspired approaches to incorporate readings otherwise discarded during robot operation.

Students will have an opportunity to engage with both theoretical and applied aspects of the work, working from concept to deployment.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr Donald Dansereau, Dr Alexandre Cardaillac

Eligibility: Programming skills. Experience with one or more of imaging, image processing, and/or computer vision would be an asset.

Project Description:

Sensor noise is an important characteristic of the imaging process, and modelling it is a fundamental problem for image processing. It can significantly impact a robot's ability to perceive and interact with its environment. This field is not well studied when it comes to underwater imagery.

This project is an opportunity to dive into a field with a lot of possibilities. By studying how water and its turbidity affect camera noise, we can develop a better understanding of the underwater environment and improve downstream tasks.

In collaboration with researchers at the ACFR, this project will capture images in both above and underwater environments to compare and characterise the sensor noise and the impact of water. There is a potential to work both with conventional and emerging camera technologies including event cameras and light field cameras.

Depending on interest and ability, there is scope to capture images in a water tank, use existing datasets from field trips, and model the underwater sensor noise in modeling software.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisor: Dr Donald Dansereau, James Gray, Jason Lai, Jack Naylor

Eligibility: Programming skills. Experience with one or more of computer vision, robotics,  Blender or CAD would be an asset.

Project Description:

Can we use photos to capture the shape of buildings, cars, or other shiny objects? It’s harder than it sounds—reflections often confuse 3D reconstruction algorithms by violating brightness constancy assumptions.

Currently, there is a significant gap in understanding how reflections affect 3D reconstruction algorithms. Most 3D reconstruction datasets and benchmarks do not isolate the impact of surface reflectivity, leaving us with an incomplete picture.

In collaboration with researchers at the ACFR, this project seeks to understand how different state-of-the-art 3D reconstruction algorithms are impacted by reflectivity. Ultimately we seek to combine insights from across approaches to successfully model even the most challenging objects.

The project includes constructing an experimental setup, either by designing and rendering photorealistic 3D scenes, or fabricating physical scenes and using a high-precision robotic arm.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr Donald Dansereau, Jack Naylor

Eligibility: Strong programming skills. Intermediate understanding of optics, electronics and computer vision techniques is recommended.

Project Description:

Light carries rich environmental cues beyond appearance, including wavelength dependence, polarization, and view-dependence. These signals could help robots across a range of applications by allowing them to estimate material and surface properties.

Computational imaging enables novel sensing modalities that extract otherwise inaccessible information under data, size, and capability constraints. This project focuses on developing polarization cameras with coded apertures to estimate material, stress, specularity, and 3D surface shape.

These sensors can enhance robotic perception for tasks such as inspecting reflective spacecraft, assessing material failure in industry, and mapping complex environments.

Students may contribute to various aspects, including simulating polarized coded-aperture masks, designing and characterizing optical prototypes, or applying the sensor to downstream tasks like 3D reconstruction and material classification.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr Donald G. Dansereau, Chengyang Yan and Bina Rajan

Eligibility: Programming skills. A basic understanding of image formation process. Experience with computer vision applications and robot operation would be an asset.

Project Description:

Motion blur can make it harder for robots to see and understand their surroundings, especially when moving quickly. If the camera captures a blurry image, important details may be lost, making it difficult for the robot to detect objects or navigate safely. By studying how motion blur affects the robot’s vision, we can find ways to reduce its impact, improve camera settings, take control over the blur, and enhance the robot’s ability to see clearly.

In collaboration with researchers at the ACFR and using a recently added nimble unmanned ground vehicle, this project will build a deep understanding of how motion blur manifests and impacts robotic vision. The ultimate goal is to control and harness motion blur in robotic applications like search and rescue and disaster recovery.

Depending on interest and ability there is scope for field deployment with a nimble unmanned ground vehicle and precision lab-based experiments using a robotic arm.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisor: Dr Donald Dansereau

Eligibility: Strong programming skills. Experience with one or more of imaging, image processing, and/or computer vision would be an asset.

Project Description:

What does your robotic vacuum cleaner see, and who else has access to those images? In homes, hospitals, and secure industrial sites, the uptake of autonomous robots is limited by privacy concerns.

Working with researchers at the Australian Centre for Robotics, this project will develop novel sensing technologies to enable robots to visually understand their environments without capturing privacy-revealing images.

Building on existing work in the group, you will advance the design of an opto-mechatronic privacy-preserving robotic vision system.

Depending on interest and ability there is scope to focus on optics, analogue electronics, or algorithms to allow an outdoor ground vehicle to drive autonomously using privacy-preserving vision. There is also scope to quantify privacy by constructing sophisticated attacks that exploit the points of weakness of alternative approaches.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisor: Dr Viorela Ila

Eligibility:

• Strong Python/C++ (C++ preferred) skills
• ROS2 (preferred)
• Astrodynamics simulation frameworks (preferred)

Project Description:

This project utilizes two UR5e robotic arms to simulate the approach of an intervention satellite toward a target satellite in a controlled environment. One arm represents the intervention satellite, executing precise maneuvers using visual and force feedback, while the other mimics the target satellite’s motion and response.

The setup enables testing of guidance, navigation, and control algorithms, docking strategies, and adaptive control techniques. The system integrates real-time perception and predictive modeling to enhance autonomous rendezvous and capture capabilities, supporting research in on-orbit servicing, debris removal, and satellite maintenance applications.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisor: Dr Viorela Ila

Eligibility:

• Strong Python/C++ (C++ preferred) skills
• ROS2 (preferred)

Project Description:

Dynamic environments are at the forefront of robotic autonomy research. To advance robotic systems, simulating these environments is crucial. In this project, student will create a script to generate moving obstacles within a Gazebo simulation, following specific patterns or dynamic behaviours.

Although Gazebo does not provide off-the-shelf dynamic obstacles, it provides the necessary tools to simulate them. Student will enhance the simulation by creating moving obstacles, mimicking real-world challenges.

Through scripting and customization, student will gain hands-on experience in improving the realism of robot testing scenarios, which is crucial for developing and evaluating autonomous systems in dynamic environments.

Requirement to be on campus: Yes *dependent on government’s health advice.

Supervisors: Dr Viorela Ila and Jesse Morris

Eligibility:

• Strong Python/C++ (C++ preferred) skills
• ROS2 (preferred)

Project Description:

Imagine a self-driving car navigating traffic or a delivery drone avoiding pedestrians. To operate safely, robots must estimate the motion and pose of dynamic objects while localizing themselves—this is the challenge of Dynamic SLAM.

Current methods represent moving objects as sparse point clouds, which help estimate position and velocity but lack shape and volume details, making collision avoidance harder. This project aims to replace sparse clouds with lightweight geometric primitives (e.g., rectangles, cylinders, or meshes) to improve decision-making in complex environments.

You will develop techniques to estimate object primitives from sparse point clouds, gaining hands-on experience with SLAM and 3D data processing. Key outcomes include:

• A review of primitive representation in Object/Dynamic SLAM
• Exploration of suitable primitive types
• Implementation of a filtering/smoothing algorithm for object primitives
• (Bonus) Integration into a Dynamic SLAM pipeline for online estimation.

Requirement to be on campus: Yes *dependent on government’s health advice.