Robust, rich, and data efficient dynamic object description systems for enhanced robot navigation [SELF-FUNDED STUDENTS ONLY]

Seamless deployment of robots in complex dynamic environments hinges on their capability to estimate the state of dynamic objects such as pedestrians or vehicles. A description of all objects around the robot is a requirement to safely plan and execute their navigation tasks. In general, this description should not only be metric, describing objects poses and velocities, but also semantic, describing their class (e.g. pedestrian, cyclist, car), dynamic behaviour (e.g. rushing, distracted, interacting with others), and crucially predicted motion trajectories. [1] introduced seminal work on simultaneous multi-object tracking and classification from multimodal data (depth and appearance), tackling the problem utilizing a graph formulation and variational inference. However, this work suffers from several of the drawbacks of classical perception systems, including the fact that intermediate tasks are optimised independently from the main task.     

While there has been significant progress in end-to-end mono-sensor methods for multi-object tracking, which focus on improving tracking performance [2,3,4], fewer research works have tackled the problem of simultaneously estimating rich metric and semantic descriptions and predictions of moving objects from multi-modal data, while also approaching the robustness problem formally. By leveraging the state-of-the art in self-supervised learning for multi-object tracking [5] and certifiable geometric perception [6], this project will develop robust and multi-modal self-supervised approaches that simultaneously accomplish:  

● multi-object tracking, 

● motion prediction, and  

● appearance and dynamics-based classification 

● providing performance guarantees like those already achieved for other perception problems like object pose estimation.  

This project aims to introduce globally optimal dynamic object description systems that work well with both depth and appearance sensor data, are robust and can be trained without a large labelling effort while being ready for seamless integration into planning algorithms, such as those currently under development within Cardiff University’s Human-Centred Computing (HCC) research unit. Additionally, the project seeks to promote the presence of robots in diverse human-populated environments, including those found in the hospitality, catering, tourism and service industries. The results will be validated using mobile robots available at Cardiff University’s HCC laboratory. 

Keywords: Robot perception, Simultaneous multi-object tracking and classification, Self-supervised learning 

Contact for more information on the project: Dr Victor A. Romero-Cano: [Email Address Removed]

Academic criteria: A 2:1 Honours undergraduate degree or a master's degree, in computing or a related subject. Applicants with appropriate professional experience are also considered. Degree-level mathematics (or equivalent) is required for research in some project areas.

Applicants for whom English is not their first language must demonstrate proficiency by obtaining an IELTS score of at least 6.5 overall, with a minimum of 6.0 in each skills component.

How to apply:

This project is accepting applications all year round, for self-funded candidates.

Please contact the supervisors of the project prior to submitting your application to discuss and develop an individual research proposal that builds on the information provided in this advert. Once you have developed the proposal with support from the supervisors, please submit your application following the instructions provided below.

Please submit your application via Computer Science and Informatics - Study - Cardiff University

In order to be considered candidates must submit the following information:

·       Supporting statement

·       CV

·       In the ‘Research Proposal’ section of the application enter the name of the project you are applying to and upload your Individual research proposal, as mentioned above in BOLD

·       In the funding field of your application, please provide details of your funding source.

·       Qualification certificates and Transcripts

·       References x 2

·       Proof of English language (if applicable)

Interview - If the application meets the entrance requirements, you will be invited to an interview. 

If you have any additional questions or need more information, please contact: [Email Address Removed]