Re-localisation in natural environments

Overview

Re-localisation in robotics involves the process of determining a robot's current pose, consisting of its position and orientation. This can either be within a previously mapped and known environment (i.e. prior map) or relative to another robot in a multi-agent setup. Re-localisation is essential for enabling robots to perform tasks such as autonomous monitoring and exploration seamlessly, even when they encounter temporary challenges in precisely tracking their location in GPS-degraded environments. For instance, consider the 'wake-up' problem, where a robot must determine its location after being turned off or losing power.

In natural and unstructured environments, the effectiveness of re-localisation is challenged by the absence of distinctive features, making it difficult to differentiate between locations that appear similar. Additionally, changes over time due to factors like vegetation growth and weather conditions further affect the robustness of re-localisation. Current approaches predominantly rely on a single sensory modality which is susceptible to errors.

Specific areas of interest:

• Leveraging cross-modal perception to enhance re-localisation through multi-modal and multi-view techniques.
• Developing robust methodologies that require minimal human intervention.
• Developing learning algorithms that demand minimal manual annotations or labels, thus increasing their generalizability and adaptability.

Skills and experience

• Strong background in programming (preferably Python).
• Machine learning experience.
• An appreciation of concepts in computer vision and deep learning.
• Understanding of SLAM and re-localisation concepts is preferable.

Scholarships

You may be eligible to apply for a research scholarship.

Explore our research scholarships

Keywords

• deep learning
• machine learning
• relocalisation
• SLAM
• computer vision
• robotics
• re-localisation

Contact

Contact the supervisor for more information.