Position filled - Towards Full Autonomy: Deep Learning Enhanced Scene Understanding for Underwater Robots
One fully-funded, 4-year industry-sponsored studentship is available in the EPSRC Centre for Doctoral Training (CDT) in Robotics and Autonomous Systems in the Edinburgh Centre for Robotics (http://www.edinburgh-robotics.org) starting September 2018. Sponsorship will be provided by SeeByte (http://www.seebyte.com), a global leader in underwater robotics (operating from offices in Edinburgh, Scotland and San Diego, California), creating some of the world's most advanced smart software for underwater vehicles, including solutions for offshore, military, security, subsea and renewable energy sectors.
Underwater robots are now required to conduct complex, autonomous missions in unknown environments, where the robot must rapidly understand their surroundings through the analysis of data collected from their on-board sensors.
The goal of this PhD is to develop novel data analysis approaches to allow automated scene understanding and decision making within the underwater robotics domain.
Analytical scene understanding is required for many emerging applications in autonomous or human-assisted operation such as driving, safety, security and defence. This problem becomes even more challenging in the underwater domain due to limited data availability, severe communication constraints and the noisy nature of many of the sensors.
Deep Learning approaches such as Convolutional Neural Networks (CNNs) typically require large amounts of data to achieve high accuracy and generalize well to previously unseen scenarios. This is challenging in the underwater domain where deploying underwater robotics is expensive and time consuming. The PhD may consider data augmentation, transfer learning and few shot learning approaches to address the data constrained limitations within the underwater domain. Specifically, the PhD will aim to address the dynamic definition of datasets and network layers to expedite training and incremental definition of these CNNs to modularly perform different tasks (e.g. automated object recognition, scene understanding and robot control).
The successful candidate will investigate state-of-the-art data driven machine learning techniques, e.g., Convolutional and Recurrent Neural Networks, as well as Deep Reinforcement Learning techniques, extending these novel approaches to be applicable to underwater robotics domain.
The candidate will be expected to provide innovative methodologies/techniques that are applicable to the challenges and restrictions imposed on robotic solutions within the underwater domain. This project will pursue the following objectives:
1. To propose innovative adaptive deep learning techniques to automatically model systems for specific underwater applications such as object recognition, scene recognition and robotic intervention tasks underwater (using manipulator arms).
2. To design incremental deep learning strategies based on hierarchical definition of application-specific datasets in order to optimize the training process for a variety of tasks (e.g. recognition in a cluttered environment).
3. To apply the methods and strategies defined above to enable a multi-stage modular system to perform modelling at different levels (e.g. scene understanding)
Applicants should have, or expect to obtain a BSc/MSc Electrical Engineering, Computer Science or Mechanical Engineering 1st class Honours / 70% average or above, and should have experience and/or expertise in one or more of the following areas:
- Image processing
- Machine learning
- Computer vision
- Sonar Imaging
- 3D data/perception
- Computer programming (Python, C/C++)
Other engineering, science or mathematical backgrounds studied to a suitably qualified level may also be considered. Strong robotics, computer vision, computer programming and software engineering skills are desirable.
This is a fully funded 4-year UK Robotics and Autonomous Systems Scholarship covering UK fees and stipend (£14,777 for 2018/19). EU students who meet eligibility criteria may also be considered.