Machine Learning-based 3D Object Detection for Navigation in Unstructured Environments

Summary

The Australian Centre for Field Robotics (ACFR) at The University of Sydney, in collaboration with Rio Tinto, a global mining company, established the Rio Tinto Centre for Mine Automation (RTCMA) with an aim to develop and implement the vision of a fully autonomous, remotely operated mine.

Our research programs at RTCMA are dedicated to addressing critical challenges in the mining industry. For more detailed information about our research initiatives, please visit our website at https://www.sydney.edu.au/cma

Supervisor

Dr Andrew Hill.

Research location

Aerospace, Mechanical and Mechatronic Engineering

Synopsis

We are seeking a highly motivated and talented PhD student to join our research team in the field of machine learning for autonomous navigation, with a specific focus on 3D object detection in challenging, unstructured environments. This project aims to investigate the application of deep neural networks to enhance the detection and tracking of objects in specialized domains, such as open-pit mines, where conventional approaches face unique challenges.

The primary objectives of this research project are as follows:

• Data Acquisition - Explore and evaluate two distinct methods for acquiring data suitable for training and evaluating deep neural networks. This includes semi-automated annotation of recorded LIDAR data and synthetic data generation.
• Object Detection - Develop and test various deep neural network architectures for the task of 3D object detection in unstructured environments, focusing on large, dynamic objects such as wheel loaders and excavators.
• Integration with Autonomous Platforms - Propose and implement a seamless integration of a ROS2 detector module into an autonomous driving platform, allowing real-time deployment and evaluation of the trained models in a practical setting.
• 3D Multi-Object Tracking - Extend the object detection work into 3D multi-object tracking to estimate the location, orientation, and scale of objects in the environment over time, considering temporal information to improve robustness against partial or full occlusions.
• Motion Pattern Analysis - Utilize the resulting object trajectories to infer motion patterns and excavation/driving/reversing behaviours, enabling enhanced forecasting and operational training for autonomous decision-making.

This research is expected to make the following contributions to the field of autonomous navigation:

• Novel methods for data acquisition and annotation in specialized domains
• Proposed 3D object detection models for unstructured environments and for objects with dynamic shapes.
• Practical integration of deep learning models into ROS/ROS2 for real-world deployment.
• Advancements in 3D multi-object tracking techniques.
• Insights into motion patterns and behaviour analysis for autonomous decision-making.

At RTCMA, we are committed to advancing science and engineering in the field of mining through our specialised PhD Top-Up Scholarship. These scholarships are designed to provide exceptional graduates with unique opportunities to excel in their doctoral studies at University of Sydney.

Benefits of the RTCMA PhD Top-Up Scholarship.

• Access to our world-class team of scientists, engineers, and mentors in academia and industry.
• The opportunity to elevate your research to new heights using our state-of-the-art facilities.
• First-hand experience in an environment where groundbreaking research is actively contributing to positive changes in the mining industry.

Additional information

Offering:

The RTCMA PhD Top-Up Scholarship is an additional award that complements the support received from a primary scholarship, such as an RTP scholarship.

Successful candidates must:

• Have a Bachelor degree (Honours or Honours Class 1 equivalent) or a Master degree,
• Have a strong background in computer vision, machine learning, or related fields, along with a passion for solving complex problems in autonomous navigation.

How to Apply:

To apply, please email andrew.hill@sydney.edu.au or mehala.balamurali@sydney.edu.au, with the subject line “RTCMA PhD Top-Up Opportunity” and your name. Include the following:

• CV and letters of recommendation.
• Transcripts (can be unofficial) and statement of purpose.

Want to find out more?

Opportunity ID

The opportunity ID for this research opportunity is 3437