This project will explore how the behaviour of autonomous vehicles may be informed by “live” satellite observations of earth (i.e. captured within a few seconds): these remote observations can reveal hidden dangers that may not be visible from the vehicle’s point-of-view.
Studentship group name
Autonomous Robotics and Vehicles
Department/School
School of Computer Science and Electrical Engineering
Research group(s)
Centre for Vision, Speech and Signal Processing
Project:
This project will explore how the behaviour of autonomous vehicles may be informed by “live” satellite observations of earth (i.e. captured within a few seconds). These remote observations can reveal hidden dangers that may not be visible from the vehicle’s point-of-view. For example, vehicles or pedestrians emerging from hidden turnings, or cars overtaking in the opposite lane.
The project will explore both the terrestrial and on-orbit aspects of the project. Specifically, it will 1) assess the validity and mechanisms that might enable the broadcast of such satellite imagery in a timely manner (i.e. specialist hardware and compression systems). 2) explore how the data might be effectively used by the AI system of the autonomous vehicle. For example, predicting the upcoming behaviours of other road users, and using this to inform the vehicles behaviour.
Program of research:
The project comprises 4 interlinked research problems. The first runs from months 1-6 developing a solution for road, car and pedestrian detection from satellite data. Existing birds-eye-view semantic segmentation techniques and datasets will be adapted for this purpose. Any modifications necessary to mimic satellite data (such as randomised cloud cover) will be integrated into the training pipeline. We will also explore the feasibility of undertaking this processing step on-orbit versus on the ground.
The second work package within the project will run from months 7-18. Here we will predict the likely future motions of other vehicles and pedestrians in the scene. This will be based on the road and car layouts detected from the satellite data above. The behavioural prediction step will be evaluated using the Waymo prediction dataset.
The third novel contribution of this project will be to develop an autonomous Reinforcement Learning (RL) agent to control the vehicle. This will run from months 19-30 and the agent’s decisions will be informed by the predicted behaviours of the other actors. This will be evaluated using a custom RL environment based on bounding boxes, with the other car’s motion based on the waymo dataset.
During months 30-36 the student will integrate all of these components into an end-to-end system. The RL environment will be expanded (potentially using the CARLA simulator) to produce simulated satellite imagery of the driving scene. The detection framework will be run on this data, which will produce the bounding boxes used by the motion predictions, which will finally inform the agent’s behaviour. After this integrated system is completed, the project will allow the final 6 months (36-42) for completion and thesis writeup.
Candidate Profile
Candidate needs a strong academic background in software from studying Computer Science, Electronic Engineering or a related subject. A first-class undergraduate degree or a Masters in one of these areas is expected (by 1st October 2023).
How to Apply
Open to UK and International students starting in October 2023.
Applications should be submitted via the Vision, Speech and Signal Processing PhD programme page. In place of a research proposal you should upload a document stating the title of the projects (up to 2) that you wish to apply for and the name(s) of the relevant supervisor. You must upload your full CV and any transcripts of previous academic qualifications. You should enter ’Faculty Funded Competition’ under funding type.
Funding
The studentship will provide a stipend at UKRI rates (currently £17,668 for 2022/23) and tuition fees for 3.5 years. An additional bursary of £1700 per annum for the duration of the studentship will be offered to exceptional candidates.
Continue with Facebook
